fix: stabilize V3.9 and bump package to 3.9.1

Preserve Second Pass audio and scoped Review conditioning, strengthen Take compatibility, and align Review Driving Audio.

Keep v5 history readable, document v6 restart requirements, and describe the fixes at the top of both READMEs. Remove workstation-specific paths and private operational records from public documentation.
This commit is contained in:
ukr8b3g-cmyk
2026-10-04 19:05:33 +09:00
parent b88c5b4d97
commit 190141e03e
45 changed files with 2211 additions and 4267 deletions
+1
View File
@@ -2,4 +2,5 @@ __pycache__/
.pytest_cache/ .pytest_cache/
*.py[cod] *.py[cod]
*.safetensors *.safetensors
!tests/fixtures/v39_sampling_v5/**/*.safetensors
*.tmp *.tmp
+33 -103
View File
@@ -1,112 +1,42 @@
# AGENTS.md # Contributor rules
## Scope ## Scope
MiniMax H3 Continuum for ComfyUI. Current `main` is package 3.9.0: it keeps V3.8X2 and adds a separate V3.9 Sampler and Reference Images V3.9 helper. This source publication is not a new GitHub Release or Registry publication. Treat the accepted V3.7 Production engine and V3.8 GPU Gates as protected execution baselines. Pre-V3.8 public Node IDs and saved workflows remain historical Release/tag contracts, not automatic V3.9 migrations. Experimental controls must not change Production behavior. Still Image Guide remains Experimental / Production HOLD.
Current release-hygiene baseline: O1 and AUDIO-R1 are PASS; the original V3.8 launch had seven IDs, V3.8X2 has ten including deprecated loader compatibility IDs, and V3.9 registers twelve in total. Sampling Contract v5, Run Storage v3, State, Session, Assembly, Driving Audio, and Second Pass/O1 remain protected. V3.9 browser/default-template GPU acceptance and Release / Registry readiness remain HOLD. MiniMax H3 Continuum for ComfyUI. Package 3.9.1 provides the V3.9 Sampler and Reference Images helper, while retaining the separate V3.8X2 workflow. Source publication does not imply a GitHub Release or Registry package publication.
## Non-negotiable rules
- Follow ComfyUI Core behavior for user-facing validation. Continuum-only prompt restrictions, model allowlists, compatibility gates, and policy-based execution stops are prohibited.
- Do not implement a new user-facing hard stop. If execution is structurally impossible or continuing risks corrupting stored data, document the exact broken internal contract and obtain explicit approval before changing the stop.
- Prompt content must never block execution. Empty, malformed, incomplete, or unknown prompt syntax falls back to a Fixed prompt and is passed through as entered. Valid Timeline/List/JSON syntax is still parsed normally.
- Keep warnings diagnostic-only unless execution cannot continue safely. Hard stops are limited to corrupt internal contracts such as invalid latent/layout topology, incompatible Assembly Plan data, broken stored revision schema/hash, unusable required payloads, or decode formats that cannot be assembled.
- Unknown upstream nodes, models, wrappers, or merged models must not be rejected solely because they are unknown.
- Keep legacy `strict_compatibility` inputs loadable for old workflows, but ignore their value. They must never enable blocking behavior.
- Before every change, run `tools\snapshot.ps1` against the authoritative source. Preserve the current V3.8 backend schema and serialized widget order until a separately approved UI/public-surface implementation changes them.
- In public templates, connection examples, screenshots, and documentation, keep ComfyUI Core nodes on their standard display titles. Do not assign custom titles to Core nodes, because users must be able to distinguish Core nodes from Continuum and third-party custom nodes immediately.
- Runtime tests must record the workflow, prompt, media inputs, model/LoRA/sampler/step settings, dimensions, chunk count/duration, elapsed time, and observed result. Inspect embedded workflow metadata when present.
- Do not globally replace `PackedLayout`, `MiniMaxH3.extra_conds`, or other ComfyUI classes.
- Keep SageAttention, Sol-Attn, and Spectrum external; integrate through public ComfyUI wrappers and H3 payload contracts.
- Preserve the identity of `layout.position_ids` when changing MM-RoPE coordinates.
- Never silently resize a continuation State or Session to a different resolution.
- Unknown H3 layout contracts must fail clearly instead of rendering a subtly incorrect join.
- State and Session files remain safetensors + JSON and are written atomically.
- UI nodes call reusable non-UI modules. Legacy nodes and V3.4 must share temporal, state, continuation, and layout logic where their contracts overlap.
- V2 must not decode Video/Audio VAE between H3 chunks.
- Accepted full chunk latents remain on CPU; do not accumulate generated chunks in VRAM.
- Spectrum history must remain run-scoped through its public sampler wrappers; never import Spectrum private runtime functions.
- Do not treat V1-V3.7 public Node IDs as V3.8 compatibility requirements. Removing their V3.8 registration still requires an explicit audited public-surface change; historical implementations may remain internally while V3.8 depends on them. Preserve State/Session data safely unless a separately approved migration replaces it.
## Project files
- `AGENTS.md`: durable rules and prohibitions only.
- `PROJECT_STATE.md`: current paths, version, active direction, and known gaps.
- `WORKLOG.md`: append-only summaries of completed work and evidence.
- Do not duplicate long history across these files. Update `PROJECT_STATE.md` when facts change and append one concise entry to `WORKLOG.md` after completed work.
## Product direction Sampling Contract v6 and Graph Contract v4 apply to new generation. Keep original v5 history readable, without reusing its Takes in v6 generation or rewriting stored manifests. Public node IDs, socket/widget order, official workflows, Run Storage v3 and State/Session formats remain unchanged.
- Product definition: H3 Continuum is a Production Sampler that generates, reviews, partially regenerates, and resumes long-form video without restarting the entire work. ## Product and data contracts
- Keep one discoverable Main surface plus an in-node `Advanced` disclosure. Do not require users to discover a hidden node property before normal Production controls become usable.
- Evaluate new work against three pillars: `Continuation Quality`, `Non-destructive Production`, and `Open Integration`. Prefer external integration or `Deferred` when work does not clearly strengthen one of them.
- `Main / Production` must be simple by default and expose Production controls when expanded. Do not develop a separate Easy product path. Its primary flow is Continuation Quality, Review/Regenerate, Resume/Branch, and Finalize.
- `Advanced` contains Second Pass, Selective/Temporal Refine, external integration, and diagnostics. Experimental or development-only nodes do not become public merely because their code and tests are retained.
- Do not build an all-in-one node pack. Keep external Upscaler, TensorRT, FBC, Sol-Attn, Spectrum, Director, Prompt Planner, Save/Encoder, and Core or external VAE Decode outside Continuum unless an explicit integration contract is approved.
- Do not grow the Main Sampler with unbounded sockets or serialized widgets. Prefer stable latent/assembly/refine/run contracts and external adapters.
- V3.8X2 remains available as a separate workflow and Sampler alongside V3.9 on `main`. V3.7 and earlier remain obtainable from historical GitHub Release/tag rather than being carried as automatic V3.9 Node-ID compatibility.
- A saved workflow that references a Node ID not exported by V3.8 may load as an unknown node. State this explicitly in the V3.8 README, Release Notes, and migration note, and direct users to the corresponding historical Release/tag. Do not restore old registrations solely to hide this intentional support boundary.
- The accepted V3.8 launch surface is seven searchable nodes: Sampler V3.8, Reference Audios, Finalize, Load Image, Load Audio, Load Video, and Second Pass. AUDIO-R1 is the approved one-node modular-input exception; existing six node IDs and schemas remain compatible.
- Public-surface cleanup changes export mappings only. Do not physically delete legacy classes/modules while current V3.8 imports or inherits them, and do not exclude runtime-imported legacy modules from the package archive.
- Main UI disclosure state is presentation-only. It must not enter Sampling or Run Storage identity, alter backend kwargs, or rewrite values of hidden widgets. Connection-specific size controls must be visible when their corresponding input is connected.
- Keep the accepted V3.8 Main facade explicit rather than inference-driven. Its first controls are `Prompt Format`, `Continuity`, `Base Seed`, `Control After Generate`, and `Audio Continuity`, followed by independent `Chunks` and `Seconds per Chunk`, a read-only calculated `Total Length`, an always-visible `Size Source = First Image | Manual`, the applicable Resolution or exact Width/Height controls, Run, Progress, a clear Queue-ready summary, and one Advanced disclosure. Do not show permanent connected/unconnected input badges, infer an Input Mode, derive chunks from a total-duration selector, or automatically change Size Source when an upstream node is bypassed or disconnected. When the canonical backend revision is `review_ready`, replace the setup facade with the amber review status and the three explicit Queue actions; do not infer this state in the frontend.
- Keep Main UI numbers compact: Chunks and pixel dimensions are integers, and Seconds per Chunk shows a decimal only when one is meaningful. Use user-facing `Full Video`, `Review by Chunk`, and `Resume & History` labels while preserving their existing backend values. Review action buttons must remain hidden until the canonical Run Storage revision reports `review_ready`; never infer readiness from frontend widgets alone. Hide empty Render History and keep Advanced settings behind the visible in-node disclosure.
- Every visible V3.8 Main, Advanced, Review, and Render History control, plus every public Sampler input, must have non-empty mouse-over help. Context-sensitive controls must explain the currently selected behavior; `Output Size` must explicitly distinguish First Image for I2VA/FL2VA from Manual Width/Height for T2VA or workflows without a First Image.
- Completed-review UI exception (2026-09-07, user-authorized repair): a canonical backend revision with status `complete` must retain its completion panel, Back to Settings/Return to Review, and non-empty Render History. Show Try this chunk again only when that revision carries a valid review_unit, as already supported by the backend. Hide the no-op Continue/Finish buttons on completion. This extends the earlier review_ready-only presentation rule; never rewrite backend status, infer readiness from widget counts, or invent a retry unit for a full-run completion.
- Keep Core node display names unchanged in every public workflow.
- User-approved distribution exception (2026-09-07): publish the supplied Spectrum graph as `MiniMax_H3_Continuum_V38.json` and the user's matching ZIP, byte-for-byte, including its existing titles, prompts, media names, and saved settings. It is one graph with external Spectrum, rgthree, KJNodes, and ComfyUI-Easy-Use dependencies, not a Core-only template. Document setup differences instead of silently rewriting the workflow. Keep unrelated backup ZIPs out of new commits/packages.
- Hi-Res Fix is not part of the new standard workflow. Preserve its historical implementation until cleanup is approved; use Second Pass as the external latent-processor bridge.
- Keep P1a First Block Cache and P2 boundary-aware Sol-Attn `Deferred`. Keep A5 and A8b `HOLD` under their recorded release contracts. Issue #13 reporter-exact R3 remains the highest Continuation Quality investigation when its required workflow is available.
## H3 completion review handoff - Keep normal generation, Review, retry, continuation and finalization discoverable. Advanced controls must not rewrite hidden values or generation identity.
- Follow ComfyUI Core validation. Do not reject prompts, unknown models, wrappers or merged models solely because they are unknown. Unrecognized prompt syntax falls back to Fixed and passes through as entered.
- Keep legacy strict_compatibility inputs loadable but ignore their values. Warnings remain diagnostic unless an unusable internal payload or corrupted data makes execution impossible.
- Preserve accepted chunks and prior Takes. State and Session remain safetensors plus JSON, written atomically.
- Keep full accepted chunk latents on CPU. Do not accumulate generated chunks in VRAM or decode Video/Audio VAE between chunks.
- Preserve native PackedLayout behavior and position_ids identity. Never silently resize continuation State or Session.
- Keep Reference conditioning, First/Last Image, Driving Audio and generated Audio contracts separate. Second Pass must preserve First Pass audio when the selected path requires passthrough.
- Keep SageAttention, Sol-Attn and Spectrum external and use their public wrappers. Do not replace Core classes globally or import private Spectrum runtime functions.
- UI nodes delegate to reusable modules. Preserve legacy implementation modules while current registered nodes import or inherit them.
- Main UI disclosure is presentation-only. It must not change backend kwargs, Sampling Contract or Run Storage identity.
- Keep standard Core display titles in public examples so users can identify node providers.
- Experimental changes default OFF and must not change Production behavior. Still Image Guide and deferred continuation interventions remain outside accepted Production scope.
- Keep V3.8X2 and V3.9 as distinct workflows. Do not automatically migrate historical node IDs, workflows, Runs or Takes.
- Keep the shipped workflow JSON/ZIP files byte-for-byte unless a workflow change is explicitly requested.
- User decision (2026-09-16): do not send implementation or Gate reports to `H3情報チェック` automatically, ask for handoff approval, or wait for a reply. Send only when the user explicitly requests it again. ## Validation and contribution
## A8b and A5 HOLD release contracts - Preserve unrelated work and take a verified snapshot before changes. The snapshot destination is configurable through tools/snapshot.ps1.
- Reuse the existing pytest configuration, tests and runtime verifier. Distinguish CPU checks, offline registration, live loading, browser acceptance and GPU results.
- Runtime checks record prompts, media, model/LoRA/sampler settings, dimensions, chunk timings and observed results. Check saved workflow metadata when available.
- Verify all hashes declared by MANIFEST.sha256 and REGISTRY_MANIFEST.sha256 after modifying their payloads.
- Current public registration contains twelve nodes. Do not register experimental implementations implicitly.
- Do not claim whole-release acceptance from a targeted repair test. Browser save/reload and the unchanged default workflow remain separate acceptance checks.
- Keep machine-specific paths, credentials, service addresses, process identifiers and private operational records out of public documentation.
- Record public behavior and version history in CHANGELOG.md and WORKLOG.md. PROJECT_STATE.md describes the public source status.
- Commit and publish only when explicitly requested. Release/tag/Registry publication is a separate action.
- 2026-09-05 explicit user exception ("許可、許可。"): Issue13 R3B may run one isolated Video-only weak tapered context-noise A/B experiment before reporter-exact R3. This waives only the reporter-exact/independence prerequisite for that diagnostic. Keep Production/source runtime unchanged, Audio/masks/depth/Seed/SIGMAS/grouping unchanged, default OFF bit-exact, and pass CPU/Shadow/Replay before GPU. Restore clean exported prefix after the diagnostic perturbation. No combined interventions, A5 work, public controls, or Production promotion are authorized. General A8b and Release HOLD remain; GPU-audit sharing is limited to the fixed H3情報チェック task. Run the standard validation entry point from the repository root:
- A8a is the accepted read-only baseline: full pytest `1041`, Manifest `242/242`, every observer record has `execution_applied=false`, and Production/Sampling/Run Storage/State/Session/Assembly parity is preserved. ```powershell
- Keep A8b execution policy on HOLD until one intervention type is specified and approved. Do not change context, transport, mask, latent, or conditioning in combination. Seed, SIGMAS, and physical grouping remain unchanged unless separately approved. powershell -ExecutionPolicy Bypass -File tools/validate.ps1
- Exclude the failed Issue #13 candidates from A8b: effective 13-frame history, 22-frame Video prefix with 75% preservation, Soft Re-anchor, and Production Fixed Prefix/teacher forcing. ```
- Before A8b GPU Sampling, either complete reporter-exact Issue #13 R3 or prove by static and replay evidence that the proposed intervention is independent of the recursive-feedback path. Then pass an A8a Shadow/Replay Gate with no unnecessary action, Terminal misclassification, or Run Storage replay difference.
- Any approved A8b implementation starts Experimental and Default OFF. Existing Production must remain bit-exact, public schema changes require separate approval, and external Sage/Sol/Spectrum wrappers must remain intact.
- Keep A5 Reference Video Influence Schedule/Fade on HOLD until reporter-exact Issue #13 R3 classifies Reference Video as unrelated, an amplification-only factor with a defined safe range, or a necessary/strong cause candidate. Only the first two outcomes may unlock Experimental A5 work; the last keeps A5 on HOLD.
- A5 OFF must remain bit-exact with Production. Initial A5 may change only Reference Video influence; it must not change Video continuation, Audio, Seed, SIGMAS, Terminal Merge, Assembly, or Run Storage. Do not implement A5 and A8b concurrently.
- A5 promotion requires at least `3x5` plus a long-form continuation GPU Gate covering trajectory, identity, saturation, contrast, sharpness, and highlight drift. Short-form evidence alone is insufficient.
## Validation
```bash
powershell -ExecutionPolicy Bypass -File tools\validate.ps1
```
Runtime installer verification must check native PackedLayout behavior and all registered nodes.
GPU checks must cover Sage only, Sage+Sol, Sage+Spectrum, and Sage+Sol+Spectrum with identical prompts and seeds. V2 must also be compared against the equivalent V1 3×5-second graph.
- Once the user has explicitly approved a GPU run, free-VRAM and process-RSS readings are advisory. Do not refuse to start or stop an otherwise healthy run solely because a numeric free-VRAM/RSS threshold was crossed. Continue unless an actual CUDA/PyTorch/Windows OOM or allocation failure, backend crash/health failure, NaN/Inf, or a structural layout/AV/duration integrity failure makes the run unsafe or invalid.
## Reusable local environment
- Source repository: `D:\Codex\_git_push_work\ComfyUI-H3-Continuum`
- Primary tested runtime: `D:\StabilityMatrix\Data\Packages\ComfyUI_W\custom_nodes\ComfyUI-H3-Continuum`
- Secondary runtime: `D:\StabilityMatrix\Data\Packages\ComfyUI_WAN\custom_nodes\ComfyUI-H3-Continuum`
- User-organized V3.8 test videos and related test artifacts are under `D:\output\video\comfy_video\v38`. Reuse that path when inspecting prior V3.8 outputs; do not move, rename, or delete its contents without explicit approval.
- Restart ComfyUI_W with `D:\StabilityMatrix\Data\Packages\ComfyUI_W\venv\Scripts\python.exe`. Do not launch the base executable reported by `Get-Process.Path`; it omits the venv dependencies even though it is the resolved process image.
- When the source adds or changes runtime imports, do not synchronize only the visibly changed files. Install the complete current `REGISTRY_MANIFEST.sha256` set, or first prove a dependency-complete narrow set; then verify registration through the live ComfyUI `object_info` endpoint. A standalone verifier launched from the source checkout is not sufficient evidence for a mixed installed runtime.
- Reuse the repository's existing pytest configuration and helper scripts instead of rediscovering paths on every task.
- Read this file before each task. Record durable project decisions, paths, commands, failure causes, and validation methods here so later work reuses them instead of restarting from zero.
- GitHub authentication for this repository is already configured. Do not request a new login or token unless an actual authentication command fails.
- Commit and push only when the user explicitly requests publication.
## User-approved loader migration (2026-09-21)
- Keep H3EasyLoadImage and Enable Image; delegate inherited mode accessors and leave Core widget collections, drawing and serialization intact.
- New official workflows use Core LoadAudio and Core LoadVideo -> H3ContinuumVideoAdapter. The adapter has no Enable/file/upload UI and reuses the existing frame-rate conversion. Preserve the original Audio/Video IDs as deprecated compatibility nodes; never infer lost filenames or saved OFF states.
- The approved export exception adds H3ContinuumVideoAdapter to the existing nine-node mapping. Do not change Main Sampler, Sampling, external wrappers, Decode Cache or Run Storage semantics for this repair.
## User-approved Timeline Video experiment (2026-09-21)
- This ZIP adds one optional appended Video Reference Mode to V38. Missing mode stays Repeat Reference. Follow Timeline is opt-in and uses the existing physical visible range; do not change grouping, seed, audio or masks.
- Keep current public workflow files unchanged. Separate Experimental Follow/Repeat templates are used until browser and GPU acceptance.
- CPU module/patch-fixture tests do not constitute a full ComfyUI or GPU gate. H3 information-check handoff remains pending.
+11 -2
View File
@@ -1,5 +1,14 @@
# Changelog # Changelog
## 3.9.1 — stability fixes (2026-10-04)
- Preserve audio-only and mixed Second Pass keyframes while spatially adapting only video.
- Use Sampling Contract v6 / Graph Contract v4 for safe Take reuse, including sampler closures and MODEL CFG/wrappers. Keep v5 history readable without adopting its prefixes into v6 generation or rewriting old manifests.
- Slice Review Driving Audio once from physical-group natural time; use Core resampling and document explicit Reference Encode Cache invalidation for direct VAE weight mutation.
- Match V3.9 partial Review conditioning and Reference routes to actual output-group identities before Second Pass. Preserve complete-run behavior, original audio LATENT objects and saved Takes.
- The repaired runtime passed 1696 CPU cases (1 skipped) and three targeted 512×512 Review Second Pass GPU cases. Browser acceptance remains pending. Optional latent-upscale doubled outlines are tracked separately in Issue #27 and are not a required gate for this stabilization.
- Bump package metadata to 3.9.1 while preserving V3.9 public node IDs, widget order and official workflows. Restart ComfyUI and refresh the browser after updating. Keep historical V3.8 releases untouched; this main-source update adds no LoRA Plan and does not create a GitHub Release or Registry package.
## 3.9.0 — V3.9 source on main (2026-09-29) ## 3.9.0 — V3.9 source on main (2026-09-29)
- Publish the separate V3.9 Sampler and nine-slot Reference Images V3.9 helper with fixed `@R1`–`@R9` identities and All / Per chunk assignments. V3.8X2 keeps its existing Sampler and Reference connections; saved Runs/Takes are not automatically migrated. - Publish the separate V3.9 Sampler and nine-slot Reference Images V3.9 helper with fixed `@R1`–`@R9` identities and All / Per chunk assignments. V3.8X2 keeps its existing Sampler and Reference connections; saved Runs/Takes are not automatically migrated.
@@ -11,12 +20,12 @@
- Add a separate nine-slot Reference Images V3.9 helper with fixed `@R1`–`@R9` IDs, in-node All chunks / Per chunk table, and one typed connection to the V3.9 Sampler. Both choices use the existing group-effective routing engine; disconnected or unassigned slots do not renumber later images. - Add a separate nine-slot Reference Images V3.9 helper with fixed `@R1`–`@R9` IDs, in-node All chunks / Per chunk table, and one typed connection to the V3.9 Sampler. Both choices use the existing group-effective routing engine; disconnected or unassigned slots do not renumber later images.
- Supply a separate V3.9 workflow JSON/ZIP and user guide. V3.8 Sampler and V38X2 workflow JSON/ZIP remain unchanged. Older V3.9 prototype graphs require manual rewiring and saved Runs/Takes do not migrate automatically. - Supply a separate V3.9 workflow JSON/ZIP and user guide. V3.8 Sampler and V38X2 workflow JSON/ZIP remain unchanged. Older V3.9 prototype graphs require manual rewiring and saved Runs/Takes do not migrate automatically.
- Local CPU `1530 passed / 1 skipped / 0 failed`, frontend Reference Images and Review Queue contracts, and standalone runtime verification passed. Live-browser and GPU acceptance of this new helper/UI remain pending. No commit, push, Release or Registry publication. - Local CPU `1530 passed / 1 skipped / 0 failed`, frontend Reference Images and Review Queue contracts, and standalone runtime verification passed. Live-browser and GPU acceptance of this new helper/UI remain pending. No commit, push, Release or Registry publication.
- Rebase the separate V3.9 JSON/ZIP on the supplied V38X2+Decode Cache Helper workflow, preserving its appended Repeat Reference video setting, media loader values, Image bypass and Decode Cache Helper. Route Reference Images 1–9 through the new helper; keep all other graph nodes and settings unchanged. The running 8188 backend still exposed the old V3.9 schema until a full backend restart. - Rebase the separate V3.9 JSON/ZIP on the supplied V38X2+Decode Cache Helper workflow, preserving its appended Repeat Reference video setting, media loader values, Image bypass and Decode Cache Helper. Route Reference Images 1–9 through the new helper; keep all other graph nodes and settings unchanged. Restart ComfyUI fully after updating to load the new node schema.
## Prior local prototype — V3.9 Reference Routing RR-R6B (2026-09-25) ## Prior local prototype — V3.9 Reference Routing RR-R6B (2026-09-25)
- Add a separate V3.9 Sampler ID with default All (unchanged V3.8 Reference path) and opt-in Custom R1–R9 per-chunk selectors. Keep the V3.8 schema/widget order and official V3.8X2 workflow files unchanged. - Add a separate V3.9 Sampler ID with default All (unchanged V3.8 Reference path) and opt-in Custom R1–R9 per-chunk selectors. Keep the V3.8 schema/widget order and official V3.8X2 workflow files unchanged.
- Carry the frozen RR-R5 group contract through the private runtime to a read-only, post-Queue Plan Inspector. Pending groups remain configured/unverified; accepted groups display backend-verified generated/reused status, effective Picture mapping, SHA and warnings. - Carry the frozen RR-R5 group contract to a read-only, post-Queue Plan Inspector. Pending groups remain configured/unverified; accepted groups display backend-verified generated/reused status, effective Picture mapping, SHA and warnings.
- Capture matching group-local routed conditioning in same-Queue Refine Context; Second Pass distinguishes verified Reference inheritance from prompt-only fallback. No cross-session reconstruction is claimed. - Capture matching group-local routed conditioning in same-Queue Refine Context; Second Pass distinguishes verified Reference inheritance from prompt-only fallback. No cross-session reconstruction is claimed.
- Local full CPU: `1524 passed / 1 skipped / 0 failed`. RR-R6 GPU/API integration passed 11 Queues, including Review Q1–Q3, Second Pass, and the Sage/Sol/Spectrum wrapper matrix. The subsequent real-browser B0–B5 Review, completion, and save/reload gate passed using a corrected Ref2V 8-step LoRA / eight-step workflow; this is not a six-step rerun. Subjective identity and long-form image quality remain ungraded. - Local full CPU: `1524 passed / 1 skipped / 0 failed`. RR-R6 GPU/API integration passed 11 Queues, including Review Q1–Q3, Second Pass, and the Sage/Sol/Spectrum wrapper matrix. The subsequent real-browser B0–B5 Review, completion, and save/reload gate passed using a corrected Ref2V 8-step LoRA / eight-step workflow; this is not a six-step rerun. Subjective identity and long-form image quality remain ungraded.
- V3.8X2 workflow JSON/ZIP and the V3.8 Sampler remain available. V3.9 `All` uses the V3.8 shared-reference path; changing to V3.9 `Custom` is a manual workflow change and does not automatically migrate an accepted Run or Take. - V3.8X2 workflow JSON/ZIP and the V3.8 Sampler remain available. V3.9 `All` uses the V3.8 shared-reference path; changing to V3.9 `Custom` is a manual workflow change and does not automatically migrate an accepted Run or Take.
+3 -4
View File
@@ -1,14 +1,13 @@
ComfyUI-H3-Continuum 3.9.0 — V3.9 / V3.8X2 インストール ComfyUI-H3-Continuum 3.9.1 — V3.9 / V3.8X2 インストール
V3.9.0はmainの現行ソースです。新しいGitHub Release/Registry公開は別工程で、過去のRelease/tagは残します。V3.8X2 WorkflowをV3.9へ自動変換せず、別々のWorkflowとして保存してください。保存済みRun/Takeも自動移行しません。 3.9.1はV3.9の安定化修正版です。旧v5 Takeと履歴は読めますが、新しいv6生成の前半には再利用できません。新規Full VideoまたはStart again from Chunk 1で開始してください。新しいGitHub Release/Registry公開は別工程で、過去のRelease/tagは残します。V3.8X2 WorkflowをV3.9へ自動変換せず、別々のWorkflowとして保存してください。保存済みRun/Takeも自動移行しません。
1. ComfyUIを終了します。 1. ComfyUIを終了します。
2. 既存の `ComfyUI-H3-Continuum` を更新する場合は、作業中のworkflowとRun Storageを先に退避します。 2. 既存の `ComfyUI-H3-Continuum` を更新する場合は、作業中のworkflowとRun Storageを先に退避します。
3. ZIPを展開し、`ComfyUI-H3-Continuum` フォルダーを `ComfyUI/custom_nodes/` に配置します。 3. ZIPを展開し、`ComfyUI-H3-Continuum` フォルダーを `ComfyUI/custom_nodes/` に配置します。
4. ComfyUIを再起動します。旧UIが残る場合はブラウザをハードリロードします。 4. ComfyUIを再起動します。旧UIが残る場合はブラウザをハードリロードします。
StabilityMatrixの配置例: インストール先は、使用しているComfyUIの `custom_nodes/ComfyUI-H3-Continuum/` です。
`D:\StabilityMatrix\Data\Packages\ComfyUI_W\custom_nodes\ComfyUI-H3-Continuum\`
PowerShellによる自動インストール: PowerShellによる自動インストール:
`install_windows.ps1` `install_windows.ps1`
+41 -31
View File
@@ -2,9 +2,9 @@ dca65fb1393e67f59c6f81f9411e3e745eb8f1097e9a2adf7ffc6f4e3901ca9f __init__.py
54ba2a58388c42063d24335aa2737e86bd4c15fe9632dcc9ffba18c328e04641 .comfyignore 54ba2a58388c42063d24335aa2737e86bd4c15fe9632dcc9ffba18c328e04641 .comfyignore
3d681ded5d4c50cc20c2382e29fdc98d8a48e76d552c41ad5a195d941e66e612 .gitattributes 3d681ded5d4c50cc20c2382e29fdc98d8a48e76d552c41ad5a195d941e66e612 .gitattributes
8296f0afc7be47171e439c02fb2988cfeb0df904acbe3fc4f507f426d709157e .github/workflows/tests.yml 8296f0afc7be47171e439c02fb2988cfeb0df904acbe3fc4f507f426d709157e .github/workflows/tests.yml
0a2f69d4cec89d440f5d5aa1ec68a7c9a741382d313f80dcddf96d13f70b3e68 .gitignore a28c4c592d56c57e0967f0adce5655313b942b8cc04f77766c0db1bec96c13e0 .gitignore
3e7a1ec3380be2a222c8a3e9b8a6bb184d3a760478957f48186438f15235bdba branch_provenance.py 3e7a1ec3380be2a222c8a3e9b8a6bb184d3a760478957f48186438f15235bdba branch_provenance.py
593ec5bafb8e506cc7b326bdc7e3f024a142569f24f8d39c1aaa3b5acfd0beca CHANGELOG.md 16f886cd4b296290880e549a3f4683308246dd1fd93e479c3bbd6ab29307beda CHANGELOG.md
c1ed3a0342730450958fb62fd21854a876991f3a4aefb24e5383b0c73cd2312d compatibility.py c1ed3a0342730450958fb62fd21854a876991f3a4aefb24e5383b0c73cd2312d compatibility.py
4d62508f900af9f93ec320afb900c27cc9a300d1e0c0a94714258498c5ede5fc conditioning.py 4d62508f900af9f93ec320afb900c27cc9a300d1e0c0a94714258498c5ede5fc conditioning.py
294251c40b1be2e296c5e5f627b68091d087bdb39179e3a0c3cebd2ee8a344f5 constants.py 294251c40b1be2e296c5e5f627b68091d087bdb39179e3a0c3cebd2ee8a344f5 constants.py
@@ -75,7 +75,7 @@ ae3338f720fc22c05ab4a661c68f9c5dcb20d751a835f0e60d4738333be48e63 docs/images/v3
b765b37502b012b462606c4d2dcef43f13950e5fd682824fb18cdb19d542c121 docs/images/v39/v39-sampler-node.png b765b37502b012b462606c4d2dcef43f13950e5fd682824fb18cdb19d542c121 docs/images/v39/v39-sampler-node.png
77c225d1e49e06d00b04acf811177c9a4317e792d4bfe92df539de99420c3563 docs/LOADER_PERSISTENCE_REPAIR.md 77c225d1e49e06d00b04acf811177c9a4317e792d4bfe92df539de99420c3563 docs/LOADER_PERSISTENCE_REPAIR.md
b30fe4640359bebfcdeeb94dbeec94cb346c8e3646509e1dcdb87d7021ba04ff docs/N2B_BRANCH_PROVENANCE_CONTRACT_V1.md b30fe4640359bebfcdeeb94dbeec94cb346c8e3646509e1dcdb87d7021ba04ff docs/N2B_BRANCH_PROVENANCE_CONTRACT_V1.md
a884c0bb08fca1f3a2f2b8f0e607f5d2c9ea27feec902fdce720743026fae019 docs/R0_RUNTIME_CHARACTERIZATION.md 994b0b969089247921ca21306496cc1e8e6f729b587a8206b6403e0b7a11734c docs/R0_RUNTIME_CHARACTERIZATION.md
e6b8cc3250776868f473357cd97ce299fcea31ae18b879ec6e6466f47dbf704e docs/R6_NODE_FACADE_DELTA.md e6b8cc3250776868f473357cd97ce299fcea31ae18b879ec6e6466f47dbf704e docs/R6_NODE_FACADE_DELTA.md
91ee4adf828415e84e2f413e3a4e20dc16356f139e8ef1f6756f584cad04f3c8 docs/REVIEW_RESTORATION_REPAIR.md 91ee4adf828415e84e2f413e3a4e20dc16356f139e8ef1f6756f584cad04f3c8 docs/REVIEW_RESTORATION_REPAIR.md
93f07aa727b34802bce3d333426b8ba2af522500882a6d222d00cf577ca82e49 docs/SEAM_TEST_SOURCE_JA.md 93f07aa727b34802bce3d333426b8ba2af522500882a6d222d00cf577ca82e49 docs/SEAM_TEST_SOURCE_JA.md
@@ -89,9 +89,9 @@ ca9f1aecd1032f0f0b237948c7fc1e4e693440dff35f5f17fd5e6d5ca4dab7f9 docs/V38_BEGIN
717577efdec5e61600504b77bc2a099ef022c4fe2333e936c3310c7182d8f2a0 docs/V38_OPEN_INTEGRATION_CONTRACT.md 717577efdec5e61600504b77bc2a099ef022c4fe2333e936c3310c7182d8f2a0 docs/V38_OPEN_INTEGRATION_CONTRACT.md
3496636f645b39336fd95a4f21a562810c56f8e1d92fa6eeb1e4b0786ad11b37 docs/V38_RELEASE_AND_MIGRATION.md 3496636f645b39336fd95a4f21a562810c56f8e1d92fa6eeb1e4b0786ad11b37 docs/V38_RELEASE_AND_MIGRATION.md
4d17d74107e78404b190fec9494b443249e5aff4f58670c50d54dfdfb064334b docs/V39_REFERENCE_INLINE.md 4d17d74107e78404b190fec9494b443249e5aff4f58670c50d54dfdfb064334b docs/V39_REFERENCE_INLINE.md
f0806bb2b999794ddda072e98424dbd1b613df1cd8a096b12d979f124e066d8a docs/VALIDATION_JA.md 1f5d49944c6b1dd0ff1c2edd1fcb54b6c2698f0b8cba85f5f59425c80df18460 docs/VALIDATION_JA.md
3fec541a675c1e4e5e287cc8e498036001f34f7a62460321e16f9ba27b5e4c18 docs/VALIDATION_RESULTS.md 3fec541a675c1e4e5e287cc8e498036001f34f7a62460321e16f9ba27b5e4c18 docs/VALIDATION_RESULTS.md
cedd7b07901d6065079ef37380f0062d535a39c98c8f614e1b26582f549d7183 driving_audio.py a8c56379abe23b98c8ceed6ea5e60a02ecc3c6d9636232d7371615f4eb306e81 driving_audio.py
6771c9905d35666121c4892d2ec26202505f96d244ce814ad9a87a02878952cd examples/H3_CONTINUUM_V1_WIRING.md 6771c9905d35666121c4892d2ec26202505f96d244ce814ad9a87a02878952cd examples/H3_CONTINUUM_V1_WIRING.md
be419d6c773784823af7b58a3423ef2e21d3cc3e23c40e60e274560895ca034b examples/H3_Continuum_V2_README.txt be419d6c773784823af7b58a3423ef2e21d3cc3e23c40e60e274560895ca034b examples/H3_Continuum_V2_README.txt
4b94f779a98bfc555f6999113a694da07746cc3cf400bfc70ebc580592034d52 examples/H3_Continuum_V2_reference.webp 4b94f779a98bfc555f6999113a694da07746cc3cf400bfc70ebc580592034d52 examples/H3_Continuum_V2_reference.webp
@@ -116,38 +116,38 @@ f3d8dd0f0872afdd97fe24e58b5cb74190d03e3583a66a5d013fee41f6894cda examples/workf
65689fd800a60ab6939bea020d630616b03a453cf6d48022cd1361992a765abc examples/workflows/MiniMax_H3_Continuum_V39_Reference_Inline.zip 65689fd800a60ab6939bea020d630616b03a453cf6d48022cd1361992a765abc examples/workflows/MiniMax_H3_Continuum_V39_Reference_Inline.zip
40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/workflows/MiniMax_H3_Continuum_V39.json 40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/workflows/MiniMax_H3_Continuum_V39.json
f69b734b5bf19fb6db78bbe4737db0bef69c2a6cef2710f28767882ef35ec847 examples/workflows/MiniMax_H3_Continuum_V39.zip f69b734b5bf19fb6db78bbe4737db0bef69c2a6cef2710f28767882ef35ec847 examples/workflows/MiniMax_H3_Continuum_V39.zip
7861fac4c8fe82372c5459aa574d35ea003bdb4fd846d567bd891c2e38571b3d graph_contract.py a344da581b3c2d77558a460d2b5256e6c9a63b5008c98b52999ceda2e0b2b767 graph_contract.py
bf0556b076dbaca67c3405a7d86c333ca7fdf3d4381529b0a9edc906519943e9 guide_timeline.py bf0556b076dbaca67c3405a7d86c333ca7fdf3d4381529b0a9edc906519943e9 guide_timeline.py
6e889c42490ccb5f376694b5944c693a84e7febedf38e631d687e96bc1e76376 H3-Continuum-LLM-System-Prompt-v1.zip 6e889c42490ccb5f376694b5944c693a84e7febedf38e631d687e96bc1e76376 H3-Continuum-LLM-System-Prompt-v1.zip
144324850d4ba086938da5f4ed7b02ff0668dec554f092695328240e43fec3ec H3-Continuum-Skill-v1.zip 144324850d4ba086938da5f4ed7b02ff0668dec554f092695328240e43fec3ec H3-Continuum-Skill-v1.zip
1473c384082700f422f2cf9141e0d012d87d2b43ac8a8a433edc7172e6ca7bf6 hardening.py 1473c384082700f422f2cf9141e0d012d87d2b43ac8a8a433edc7172e6ca7bf6 hardening.py
9fca2d9c04be387160dee84e5ee9ce538cb910e79e356b9236e6d2240ea6d4ee INSTALL_JA.txt 199f946243bed42786319517f26e628612877df61ce36d55552485155b5df44b INSTALL_JA.txt
23bddc5a41d2e997f4fbf789c7128e32f4f197f1405b2aef1c76794f57482446 install_windows.bat 23bddc5a41d2e997f4fbf789c7128e32f4f197f1405b2aef1c76794f57482446 install_windows.bat
12bf572a65d57bbc9a72b459f29691c2057768f4fb7a74f9db87f4775e995bda install_windows.ps1 12bf572a65d57bbc9a72b459f29691c2057768f4fb7a74f9db87f4775e995bda install_windows.ps1
139546ec1ff459945aa89b50f74b5ee69f62a0e07bbddfcae06936091a409a83 layout_adapter.py 139546ec1ff459945aa89b50f74b5ee69f62a0e07bbddfcae06936091a409a83 layout_adapter.py
c44abbd789091bbead01250dbd5453e60539d2665c69f162335fdbd2936019c2 LICENSE c44abbd789091bbead01250dbd5453e60539d2665c69f162335fdbd2936019c2 LICENSE
d3e213005670e94dee33a240349a4d245cf9a3b6602099f11ed30cf5a32e2c21 media.py d3e213005670e94dee33a240349a4d245cf9a3b6602099f11ed30cf5a32e2c21 media.py
8ce5a7c255088fb6038925a93b71b87972bf2f02e450af86da5c317e95506c41 metadata.ini ad0835f8fb6d171d8ff8d7295fb48291abc93ef02fd46a2e283265c15b93366a metadata.ini
fd23c7f32810788aaaf1709c4959c40ec72303aecc6f6afa5fa68d824cca97a8 minimax-h3-continuum-dance-director.zip fd23c7f32810788aaaf1709c4959c40ec72303aecc6f6afa5fa68d824cca97a8 minimax-h3-continuum-dance-director.zip
d7e613e2cc24110213b477bdd1d390234a41f2af3b1ed006a809066d71d658eb model_patch.py d7e613e2cc24110213b477bdd1d390234a41f2af3b1ed006a809066d71d658eb model_patch.py
ecb65a5dbf9e1683c5a3cadd82daaea3a90a2b776f6ebabfbe8b07aa93177422 nodes.py ecb65a5dbf9e1683c5a3cadd82daaea3a90a2b776f6ebabfbe8b07aa93177422 nodes.py
99405e4357a143b0f5d0b374c9c0d2089435a7055bea2f3d1216dbbad25e4339 PACKAGE_INFO_JA.txt 99405e4357a143b0f5d0b374c9c0d2089435a7055bea2f3d1216dbbad25e4339 PACKAGE_INFO_JA.txt
ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt
b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py
667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml 1047d05b0aedc405e3a36ee571a8939111d69bb5565e09a3e1caa3c40bf33170 pyproject.toml
94fdc2d545b2d4077a4d70229fd22ba3ccac21d49d1097d2873f76d04e9a3c7a pytest.ini 94fdc2d545b2d4077a4d70229fd22ba3ccac21d49d1097d2873f76d04e9a3c7a pytest.ini
47f5e57dc6c84b0103d89d6549e87ad45ab4c263cadb36f21a952843a65204a7 README_JA.md cea43f63792a28b9d19035b3ee64ff482cf0357aec9edd6fab34eb84956b2973 README_JA.md
b89e0cc5348a8dbab8781a1b57c99c4c42f47598efe218cc087a77bc58419839 README.md 828fc5d51539c3bb5c4bddaa741b25fcda3faf2c1dec89ed545a63247b7b02a1 README.md
44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py 9539f5cb9fb695841073fa58d9246d5a7410bc372f6cd340378185004dfaf978 reference_audio.py
5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py 5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py
caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py
108e455b9df90d677e43381b5443675500e1acce0cee2a5a327bb4b56dd078b4 REGISTRY_MANIFEST.sha256 136a59c87ba789b5ddf5046a27811c4700c77d52f7fb16291d83af774a7dfeec REGISTRY_MANIFEST.sha256
467edb2c188d024c2392aafe7423b3f835e16b10f570da76e2f7dda8a962c1c5 requirements.txt 467edb2c188d024c2392aafe7423b3f835e16b10f570da76e2f7dda8a962c1c5 requirements.txt
53827c94b3f588f35bb8eacb404e1bd5e233d25f4ee0fdd286efc7d8ddb52b6f run_storage.py 1845cbf7ba4d3b35afd087a15f5a4d92dc73d4858e61982f8ccf431a836a22b1 run_storage.py
6056de6c0a5322a8a51ba5a114538dca2ef67788cd1afba10fedd483b95f6afa state_io.py 6056de6c0a5322a8a51ba5a114538dca2ef67788cd1afba10fedd483b95f6afa state_io.py
3a359a0467725cc292ccb4c86cf0c3d30bb6a114d01f5345792b09ea3d45441d state.py 3a359a0467725cc292ccb4c86cf0c3d30bb6a114d01f5345792b09ea3d45441d state.py
43ec4023dc06c941e5d7a04db4b8a89fb58d744ceb986f6eb9bf534d4fdcb87b temporal.py 43ec4023dc06c941e5d7a04db4b8a89fb58d744ceb986f6eb9bf534d4fdcb87b temporal.py
e8badc6a30c3006904944a27e905a02ffd45b3abe63cd1a0e9d97f75a2508ff4 tests/conftest.py 0c10bed921279860b0c9f330a4d4c2ec619f8bdf1bbcaff0a49f81c4dc693063 tests/conftest.py
7f34c43ea74c43cfc4f9453848c69c8ed7546de5b78c7c7d063d643f2c483c36 tests/decode_cache/conftest.py 7f34c43ea74c43cfc4f9453848c69c8ed7546de5b78c7c7d063d643f2c483c36 tests/decode_cache/conftest.py
8d0cced8c945c4333ef78688b51a988d7e1d560e5f9e6cceec8310b17c98ad53 tests/decode_cache/helpers/cache_process_worker.py 8d0cced8c945c4333ef78688b51a988d7e1d560e5f9e6cceec8310b17c98ad53 tests/decode_cache/helpers/cache_process_worker.py
a41a4de12e6fe5690845f9f0795fa82c4907f0afe2551034fa2a4c95617ac39c tests/decode_cache/test_adapter_nodes.py a41a4de12e6fe5690845f9f0795fa82c4907f0afe2551034fa2a4c95617ac39c tests/decode_cache/test_adapter_nodes.py
@@ -158,10 +158,10 @@ db08ce77cc5c5835f2403d4d589394090fa5d25ff3b1840312b2808e1bc203b7 tests/fixtures
606b3985f2b644fac40c84de3e08d1453e87a988acca51ce9595516a7c60a577 tests/frontend_decode_cache_helper.mjs 606b3985f2b644fac40c84de3e08d1453e87a988acca51ce9595516a7c60a577 tests/frontend_decode_cache_helper.mjs
868c613654467c2c739e07de5dd9b4679b36f98cd96d631d91e1d4afe5698cf5 tests/frontend_external_int_size.cjs 868c613654467c2c739e07de5dd9b4679b36f98cd96d631d91e1d4afe5698cf5 tests/frontend_external_int_size.cjs
da45ac4b9422f04a6867ff4cff4b12cf1dff88422522b8010a960b5ecfce046a tests/frontend_reference_images_v39.cjs da45ac4b9422f04a6867ff4cff4b12cf1dff88422522b8010a960b5ecfce046a tests/frontend_reference_images_v39.cjs
19bb25dea15f20fac9874b9d18c9b75156446dda7e6da91e8a9e758282c5f777 tests/frontend_review_queue.cjs 751f38be4351d459145beb3a533530e0c667552fe148de2f732efbb04494b660 tests/frontend_review_queue.cjs
36f74c03a40c4b62b90eede070d976bcd1b7f0ea828983d1e61a6cb6d80a0ea8 tests/helpers/r5_process_crash_worker.py 36f74c03a40c4b62b90eede070d976bcd1b7f0ea828983d1e61a6cb6d80a0ea8 tests/helpers/r5_process_crash_worker.py
706029af8c8d979b2b700f5bc2b96a3064fe7c728f487698d9f079abb02e9409 tests/terminal_merge_diagnostics.py 706029af8c8d979b2b700f5bc2b96a3064fe7c728f487698d9f079abb02e9409 tests/terminal_merge_diagnostics.py
fe23752312d62ddc8c23a3c598ed2f92f2d46e2e4460f06aea91921659092451 tests/test_audit_boundary_repairs.py 02e6286857353b76080b73928c0d1e7ec48b1ed16339231a4a450743f193315b tests/test_audit_boundary_repairs.py
041289556cc621ea1857cbb8931ae41d817f69e084b2e8a7622e184a3197b61b tests/test_compatibility.py 041289556cc621ea1857cbb8931ae41d817f69e084b2e8a7622e184a3197b61b tests/test_compatibility.py
875ac33b5458ec5e8d845850ed495707c394436aaadb668a44f36f6a1b73764d tests/test_conditioning_modes.py 875ac33b5458ec5e8d845850ed495707c394436aaadb668a44f36f6a1b73764d tests/test_conditioning_modes.py
2be669b6d21ec40a8b0e172b8ab431f82681f3c97c7de0935d87ad43b0ae8b4e tests/test_context_diagnostics.py 2be669b6d21ec40a8b0e172b8ab431f82681f3c97c7de0935d87ad43b0ae8b4e tests/test_context_diagnostics.py
@@ -193,7 +193,7 @@ c2560cbafc5962bce5f063e262cf6a173dd8df18f7b58035ec326f616617d61c tests/test_p25
2d16f6de7bcaf1f92718d3719654a746438f6218bba8d25b3632c318a1ab9c98 tests/test_packed_layout_preflight.py 2d16f6de7bcaf1f92718d3719654a746438f6218bba8d25b3632c318a1ab9c98 tests/test_packed_layout_preflight.py
1ae0369943ba2fb13f21ad0cf56ee58621ba09bf20174379595e157a1775c7fd tests/test_public_api.py 1ae0369943ba2fb13f21ad0cf56ee58621ba09bf20174379595e157a1775c7fd tests/test_public_api.py
892d3e3e022a527652dd995508c8bbb650c370caf9ee66fa3eb1f7fd288bba38 tests/test_public_workflow_integrity.py 892d3e3e022a527652dd995508c8bbb650c370caf9ee66fa3eb1f7fd288bba38 tests/test_public_workflow_integrity.py
5d487395fb78c3d9ef98d0af037059ca87a92a28e65285023c234f4151c8b093 tests/test_release_metadata.py 6e31f0891e592138a911c4ffffda9ebe5b5db6350318c4d1bfdf3c5e1f149678 tests/test_release_metadata.py
5f6caf1814052798f20571ebb863c098bec23d8f2e091fc1f30ce1ef9c58675a tests/test_review_identity_lifecycle.py 5f6caf1814052798f20571ebb863c098bec23d8f2e091fc1f30ce1ef9c58675a tests/test_review_identity_lifecycle.py
5f7d3e01bc78fa076ecb17f8532b6d3f3137545da4028032ad6769407d80be07 tests/test_review_provenance_reroll.py 5f7d3e01bc78fa076ecb17f8532b6d3f3137545da4028032ad6769407d80be07 tests/test_review_provenance_reroll.py
d5e255a5fdd06ea18e70f4077f855ed2f02664f91caf79b66bfbdf7df44ad07b tests/test_review_rehydration.py d5e255a5fdd06ea18e70f4077f855ed2f02664f91caf79b66bfbdf7df44ad07b tests/test_review_rehydration.py
@@ -217,11 +217,11 @@ b42bcc0983147462d96c6b71cc4a38e846995dafa6eebefc3ae0d9bb36caf359 tests/test_v2_
0b5b20a8318862d167f135720e0489b8d4f93406d67c6646e052e35b03dcb876 tests/test_v2_session.py 0b5b20a8318862d167f135720e0489b8d4f93406d67c6646e052e35b03dcb876 tests/test_v2_session.py
b5c320279fd38d2038e39951a3d016aa9b80741523d2816e01ceda1bdecb71b8 tests/test_v2_terminal_merge.py b5c320279fd38d2038e39951a3d016aa9b80741523d2816e01ceda1bdecb71b8 tests/test_v2_terminal_merge.py
e22486ac62c53132c275614e4e161c78e6dce09dd0d1a7cf64f641a3b7052b99 tests/test_v2_widget_migration.py e22486ac62c53132c275614e4e161c78e6dce09dd0d1a7cf64f641a3b7052b99 tests/test_v2_widget_migration.py
dff09049cbcc59896dad32d36cf3fdef2e6a67c6a8ff3150a33d784d8ca165c5 tests/test_v217_stable_facade.py 7b4edd7d09e382926f94f391f5a58fdee4204affbf6a72c225313ea0a5c9d9f6 tests/test_v217_stable_facade.py
473abac48a57aa5f297c6e141a94de17ddcb9424a8744bae1c46e6a6a734a0ab tests/test_v3_hardening.py 473abac48a57aa5f297c6e141a94de17ddcb9424a8744bae1c46e6a6a734a0ab tests/test_v3_hardening.py
9242f74663ccb04020ad2fb1199990c84433c9b4c5cf0560e93416c7b15e1913 tests/test_v3_physical_decode_groups.py 9242f74663ccb04020ad2fb1199990c84433c9b4c5cf0560e93416c7b15e1913 tests/test_v3_physical_decode_groups.py
35b26c8cade7acb2eec35375d7dfa9a4a112f67a5cec6a842d61c69039de0d84 tests/test_v3_video_seam.py 35b26c8cade7acb2eec35375d7dfa9a4a112f67a5cec6a842d61c69039de0d84 tests/test_v3_video_seam.py
42ac5b9337fce10a236f140d000e620ac45079ca1aaafcb8be731cfc6e0d858e tests/test_v31_run_storage.py 1c84723b0c14f3169760fe1ae95d230416f6e947ada7d97ee890a2da33e9c988 tests/test_v31_run_storage.py
3af78ab85c8e19dee8d0a98bfd11a2c0e88c30d87256da1f9699c49a19a684e9 tests/test_v32_reference.py 3af78ab85c8e19dee8d0a98bfd11a2c0e88c30d87256da1f9699c49a19a684e9 tests/test_v32_reference.py
43aef01fc587c9f9e87baf312eb4d588975aaaacab161a36bfcb638a4c11e31c tests/test_v34_driving_audio.py 43aef01fc587c9f9e87baf312eb4d588975aaaacab161a36bfcb638a4c11e31c tests/test_v34_driving_audio.py
aa1f36601489c360d2f1f989c10c5424bd85bd2115df2427afe696758eefd239 tests/test_v34_hybrid_reference.py aa1f36601489c360d2f1f989c10c5424bd85bd2115df2427afe696758eefd239 tests/test_v34_hybrid_reference.py
@@ -252,10 +252,10 @@ e813b5d21af2963a325862acd344b33caa124d0cd3940a940c58118cf27742e0 tests/test_v38
c9577e9b2d430c9435a287bf679a920020192d51f780e02b47f286e2e4b6a615 tests/test_v38_easy_mode.py c9577e9b2d430c9435a287bf679a920020192d51f780e02b47f286e2e4b6a615 tests/test_v38_easy_mode.py
bb5a1530c890d357dc48ea74039340d746e876e8f85e06d4d9a19369f8ff5961 tests/test_v38_hover_help.py bb5a1530c890d357dc48ea74039340d746e876e8f85e06d4d9a19369f8ff5961 tests/test_v38_hover_help.py
98eb59407741dd187dc22d94b4f433f341294d05cc38106fde15b580b1465001 tests/test_v38_memory_action_policy.py 98eb59407741dd187dc22d94b4f433f341294d05cc38106fde15b580b1465001 tests/test_v38_memory_action_policy.py
8edaae5f1ca93925e9269c9a4447c97edf5958395f08659e61a09e8d7efa088a tests/test_v38_multi_reference_audio.py 8fb637f1d61389a86ef545adf35a518a5ae44322655a38fe598fd3b3c8e0a4a9 tests/test_v38_multi_reference_audio.py
623d528a02b6712cba28c374ba3ad979bc84e2ac5318640692589a04ba6ab15f tests/test_v38_open_integration_contract.py 623d528a02b6712cba28c374ba3ad979bc84e2ac5318640692589a04ba6ab15f tests/test_v38_open_integration_contract.py
12337a491fa67b38b32aaeb16e0b209393960dbc46db2a8bf968b8c3902855ef tests/test_v38_packed_row_planner.py 12337a491fa67b38b32aaeb16e0b209393960dbc46db2a8bf968b8c3902855ef tests/test_v38_packed_row_planner.py
35c7a49eb6ff0159bb13ce0f731f10a6b6eca0972bd9a2debbffe650a0dc4867 tests/test_v38_public_surface_cleanup.py 94eb249f407bf6858d936721d4a8752c5a95be6fa00dfc7f828b944e53ab1bc7 tests/test_v38_public_surface_cleanup.py
7aebd3668e04233f34c2d94e75d4708cadd64d6882cf77ec59afa0a394dcb006 tests/test_v38_public_workflow.py 7aebd3668e04233f34c2d94e75d4708cadd64d6882cf77ec59afa0a394dcb006 tests/test_v38_public_workflow.py
14ff6561a127287d0a20887790a978612ef9d291f1d3be7d5e85b37b2c74760e tests/test_v38_ref_encode_cache.py 14ff6561a127287d0a20887790a978612ef9d291f1d3be7d5e85b37b2c74760e tests/test_v38_ref_encode_cache.py
8b376f45261d045b584f02b4d14cd70c2d6269b9c814d4e62b5c612f6e30e36f tests/test_v38_refine_target.py 8b376f45261d045b584f02b4d14cd70c2d6269b9c814d4e62b5c612f6e30e36f tests/test_v38_refine_target.py
@@ -264,7 +264,7 @@ aa72293fe5ba6f4499e1ec95aace3b2e94813fa35cf987a767e77bbf2ae525b1 tests/test_v38
2ac395009e49f9554af8d1a251d93fc5d5e7070bf80f5a0b7b3f7e00cde15077 tests/test_v38_review_cache.py 2ac395009e49f9554af8d1a251d93fc5d5e7070bf80f5a0b7b3f7e00cde15077 tests/test_v38_review_cache.py
72a0e3cb334049b66dd36f52c826931067c242fa63557933fd667138b38d6e8d tests/test_v38_review_control.py 72a0e3cb334049b66dd36f52c826931067c242fa63557933fd667138b38d6e8d tests/test_v38_review_control.py
f62daa235117a3041c66ebb17adedb13c0452c3a2fc0acdd51006d7aa0078a9b tests/test_v38_review_execution_cap.py f62daa235117a3041c66ebb17adedb13c0452c3a2fc0acdd51006d7aa0078a9b tests/test_v38_review_execution_cap.py
615b51cafb1441fe49486381d9cd30f53b570eec10d35b785bc13629664d84ac tests/test_v38_review_run_storage.py 3a5e298bc9b0d46e9560e397576352117d6353dd03977ad54fe41e4881c58287 tests/test_v38_review_run_storage.py
2d8b74a56da4c15509ef9ef082f87321182d9d97ca6c417d5a6c055d67a4ee8e tests/test_v38_review_ui.py 2d8b74a56da4c15509ef9ef082f87321182d9d97ca6c417d5a6c055d67a4ee8e tests/test_v38_review_ui.py
75697036f361a9b8ea74f69b15e408670f686c8c0e8a6c1d48fe059cc837d12a tests/test_v38_selective_second_pass_node.py 75697036f361a9b8ea74f69b15e408670f686c8c0e8a6c1d48fe059cc837d12a tests/test_v38_selective_second_pass_node.py
8026a241301dfc4de518649305458a538ef17a0b1567b113877a2af377738528 tests/test_v38_temporal_window_refine.py 8026a241301dfc4de518649305458a538ef17a0b1567b113877a2af377738528 tests/test_v38_temporal_window_refine.py
@@ -353,10 +353,10 @@ be0d8585fa57e0ccd6d02577bbc7299629e5bfe20e186747f571fccc98b67d7e v2/seam_types.
3ea23dbce6fcd217b6e81235855479b9b16fed7a21693cf156f9014555c4fe13 v2/session.py 3ea23dbce6fcd217b6e81235855479b9b16fed7a21693cf156f9014555c4fe13 v2/session.py
78c830cdf49034c6e3b2cd3da5914d5cfc5638eb63ec6dbd1bab13c50d6d4caf v3/__init__.py 78c830cdf49034c6e3b2cd3da5914d5cfc5638eb63ec6dbd1bab13c50d6d4caf v3/__init__.py
ea2922567df90436dd6fa21696833a49dd0cbfeee142a5fedc3a7c303824d473 v3/adaptive_continuity.py ea2922567df90436dd6fa21696833a49dd0cbfeee142a5fedc3a7c303824d473 v3/adaptive_continuity.py
b3dc61503d35d9ef0a0dc7fed1c9b2fdc00fcb2072d4e39f64ae0826f9ec7dca v3/assembly_v35.py c03f9f6307b5e3a658fe6d94b01b48bbf5726737e1a8979d6d6f374965f2dc50 v3/assembly_v35.py
7815230e0a9540bc2edeaf818e940b7a1fdd817fd217ba8a4f9cb21aaf6a308e v3/assembly.py 7815230e0a9540bc2edeaf818e940b7a1fdd817fd217ba8a4f9cb21aaf6a308e v3/assembly.py
3f62fdb35e804f42d7cbf0828b4e87e69a2b99246c3f7467fe4fa698f1d803c9 v3/conditioning_bridge_nodes.py 3f62fdb35e804f42d7cbf0828b4e87e69a2b99246c3f7467fe4fa698f1d803c9 v3/conditioning_bridge_nodes.py
e3a6f537a8709c7d05e4df95f1ff40c9129e7845e9ea23097d64953b4770e06d v3/driving_nodes.py fa7a355e52fbcfe1f53b16fa1c37af5d839215b49412adb41c90129ae4e9d9f7 v3/driving_nodes.py
147a1d3fc5e09b323a4c3d13d30e432410389bae52b58330462375b9f307b791 v3/easy_audio_nodes.py 147a1d3fc5e09b323a4c3d13d30e432410389bae52b58330462375b9f307b791 v3/easy_audio_nodes.py
64b3e54b42e5156642f9957ccc374fd190d9fc7145637984e702534fb1bf80b4 v3/easy_image_nodes.py 64b3e54b42e5156642f9957ccc374fd190d9fc7145637984e702534fb1bf80b4 v3/easy_image_nodes.py
ac859cb1a6e23b8ceae300d818c24fe1346b649a0d663017a9b3ac64aa96fca3 v3/easy_nodes.py ac859cb1a6e23b8ceae300d818c24fe1346b649a0d663017a9b3ac64aa96fca3 v3/easy_nodes.py
@@ -369,10 +369,10 @@ cb53e1b104415032ec79b7e7aa1c99848e63da3eedaa251eeba102970e4172b5 v3/hires_fix_n
91a868d2e392b018b01d0396209849c146b0d560ceb109fcdd87dea7f3d786aa v3/memory_action_policy.py 91a868d2e392b018b01d0396209849c146b0d560ceb109fcdd87dea7f3d786aa v3/memory_action_policy.py
85477badd85281cfb77582fdd5b34228a9f6b68376f1b7743af2ca7d4b55aa36 v3/memory_attribution.py 85477badd85281cfb77582fdd5b34228a9f6b68376f1b7743af2ca7d4b55aa36 v3/memory_attribution.py
4a48f74584d8ba3187e1accc31ae5acd3a2ffb9a491bb4f5ca43322db418e770 v3/node_facade.py 4a48f74584d8ba3187e1accc31ae5acd3a2ffb9a491bb4f5ca43322db418e770 v3/node_facade.py
7fe71802319ad7659c90839f960238d0fd054662af31a4d320806c426e755ecb v3/nodes.py f2779a7453a84dec8b993cf044e69ec90da6fe86940a73181234b5000f5a8fa4 v3/nodes.py
a3411c7a295308d21623551cf4aeca5bef511eb5f063f31ac84032e3bdf4e16d v3/packed_row_planner.py a3411c7a295308d21623551cf4aeca5bef511eb5f063f31ac84032e3bdf4e16d v3/packed_row_planner.py
b7c4a35a5e73f528492bc6946b97b687c412f1d3f5f976c19d45c6dc20288c9a v3/pixel_vae_resize.py b7c4a35a5e73f528492bc6946b97b687c412f1d3f5f976c19d45c6dc20288c9a v3/pixel_vae_resize.py
b139843ccfc826937c80f118920bbee2609c37d45ef88291f80f491744fd5e76 v3/plan.py 7e8681bc1ca1633389e1af1ab14100338d23a047f96c59c19151acbbfae5c108 v3/plan.py
0e3dd69d6d2bec6d9491c2fb806c25f0c62629b1eb5f9e4f77ae8d27a5b19cfc v3/planning_types.py 0e3dd69d6d2bec6d9491c2fb806c25f0c62629b1eb5f9e4f77ae8d27a5b19cfc v3/planning_types.py
ec71e3965f445756d54d005c0e27bea520e26dbac44e1ee3d3867bfe2da100b0 v3/ref_encode_cache.py ec71e3965f445756d54d005c0e27bea520e26dbac44e1ee3d3867bfe2da100b0 v3/ref_encode_cache.py
8b9c1cadf76e2c2425d6bf8a5783ced593a75f83a6b3f219a0f870b4c177e058 v3/reference_effective_plan.py 8b9c1cadf76e2c2425d6bf8a5783ced593a75f83a6b3f219a0f870b4c177e058 v3/reference_effective_plan.py
@@ -382,7 +382,7 @@ b64f4d3d9547ac00f8edb2155109476df5b8baea730fb94e8b5bfca6d11564e6 v3/reference_i
1366da221237853eddc9e4b724d4c28917868ea32607dd086361a84df052cfde v3/reference_runtime.py 1366da221237853eddc9e4b724d4c28917868ea32607dd086361a84df052cfde v3/reference_runtime.py
4fb950bbdc7cdb55365fde373ee190af6cdf99f7ce4c54ef73b94f0aca4e1fda v3/reference_storage_contract.py 4fb950bbdc7cdb55365fde373ee190af6cdf99f7ce4c54ef73b94f0aca4e1fda v3/reference_storage_contract.py
057921bf2b05e88c2cbc08c16e2ac483aea905b8a144fbadf4a21d6ae8252427 v3/reference_video.py 057921bf2b05e88c2cbc08c16e2ac483aea905b8a144fbadf4a21d6ae8252427 v3/reference_video.py
b191009d686680c6eec1fdba15b76b7c9f40f11f331faaf7b69ef784e99f4989 v3/refine_context.py 8d6cd0d7f507fd6c2d33667a6ea205e131d47571444abc94d529a98748a52517 v3/refine_context.py
519586b1b3f48c73d329f7a2aaa8a1bfc6fe31ac2ecb64f2b1fdf9c19026001a v3/refine_sampling.py 519586b1b3f48c73d329f7a2aaa8a1bfc6fe31ac2ecb64f2b1fdf9c19026001a v3/refine_sampling.py
43f33f989eb70ca87b916fc33267732a2be3dd4581fc72af17c53f840b53ca49 v3/refine_schedule.py 43f33f989eb70ca87b916fc33267732a2be3dd4581fc72af17c53f840b53ca49 v3/refine_schedule.py
afc149dc347b4c1502ddf32ea4229e633799cd34d2b2c6f98bf8a47910a9ef77 v3/refine_scope.py afc149dc347b4c1502ddf32ea4229e633799cd34d2b2c6f98bf8a47910a9ef77 v3/refine_scope.py
@@ -394,13 +394,13 @@ ae909b43ad17934c85881d3b0bea81e7b45c215eeb4d1f9f6a9ea72faf633ddf v3/reliability
be6610fb9427546d9485a17015cfbdc8786715e6a0967f3e4c1dee35d8768385 v3/runtime_coordinator.py be6610fb9427546d9485a17015cfbdc8786715e6a0967f3e4c1dee35d8768385 v3/runtime_coordinator.py
d0a004aa3030c8adaa5ea788f9acc758fcb4a556f203f9a3f63fa4e4b5772063 v3/sampling_engine.py d0a004aa3030c8adaa5ea788f9acc758fcb4a556f203f9a3f63fa4e4b5772063 v3/sampling_engine.py
452022107cea233aa84d5366ec4943ac5aadbd5fcf786113f1d542a641eac8f7 v3/second_pass_nodes.py 452022107cea233aa84d5366ec4943ac5aadbd5fcf786113f1d542a641eac8f7 v3/second_pass_nodes.py
aa4afff697bd9b8a8505e6cc7d94fcef7aabf9d93ed123ec226fad1960807e77 v3/second_pass.py 97fcb293e0714c3a74b89f08e11c21dce45d25a3010948be1af9dc059d288187 v3/second_pass.py
6e71e944313339f14e14e11cec682593159ca5542aa5d38095371da781d60224 v3/targeted_refine_sampling.py 6e71e944313339f14e14e11cec682593159ca5542aa5d38095371da781d60224 v3/targeted_refine_sampling.py
5a044bdd70c546e4446e0c26af92aae88049ffa2e0e2d01b52033a83a3152d0e v3/targeted_second_pass.py 5a044bdd70c546e4446e0c26af92aae88049ffa2e0e2d01b52033a83a3152d0e v3/targeted_second_pass.py
26b13614ba0a5fd8b192a106c9503b2f4b6fe7f9268a0fcb0f544979a49cab34 v3/video_adapter_nodes.py 26b13614ba0a5fd8b192a106c9503b2f4b6fe7f9268a0fcb0f544979a49cab34 v3/video_adapter_nodes.py
4178dcfa442abdb06ce276e05bd7bcbaa6da28469ba5e57d3da3c663de4e4a9d v3/video_seam_v35.py 4178dcfa442abdb06ce276e05bd7bcbaa6da28469ba5e57d3da3c663de4e4a9d v3/video_seam_v35.py
2f07ace1462644f0bee553827b3571535de4d0c2d9ef847bd9e0ca73a4eed5cb v3/video_seam.py 2f07ace1462644f0bee553827b3571535de4d0c2d9ef847bd9e0ca73a4eed5cb v3/video_seam.py
0d08a173819d63d969694f14e4a1d59ccbec357f37342e00a4db03e1144c2e02 version.py f1c9df5eb9f6c75bd0e505fb84731951fcd9ef1ab5d7f93a0c72f05488fa05fb version.py
b935ad3bb729948bb3890ce967b0ecbef737a5e51ddfc47a0060fc20aa78dcda video_reference_modes.py b935ad3bb729948bb3890ce967b0ecbef737a5e51ddfc47a0060fc20aa78dcda video_reference_modes.py
685df8e0fa23e3663ab31da1e4470cd19bdaa0dc16362cb606e874d9d6153c2c web/decode_cache_helper.js 685df8e0fa23e3663ab31da1e4470cd19bdaa0dc16362cb606e874d9d6153c2c web/decode_cache_helper.js
fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypass_toggle.js fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypass_toggle.js
@@ -408,7 +408,7 @@ fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypas
9db0aa94f634578a962d5db49e5c70356040a59df968e3a2e531b327a623081f web/easy_load_image.js 9db0aa94f634578a962d5db49e5c70356040a59df968e3a2e531b327a623081f web/easy_load_image.js
35090816ba7470a7255eb85d003b7e79f8a0065982dacf98a18f41c8ebff655d web/easy_load_video.js 35090816ba7470a7255eb85d003b7e79f8a0065982dacf98a18f41c8ebff655d web/easy_load_video.js
0c68d13a1f8beb7431a32b471f8b0612c42fcc4bf72dd97611a025d158008662 web/easy_v38.js 0c68d13a1f8beb7431a32b471f8b0612c42fcc4bf72dd97611a025d158008662 web/easy_v38.js
c3e2e606af4e5f3e38d64b1338baedea8cadf26060da2a8618e18bf73f220f24 web/project_id.js fbceb922aa3a5ef54886f0ca4c1a07c516ec9fadca04080049ef11f15d584668 web/project_id.js
ae59273f163a116ece418957c557c4fd8c3285e3c60814679b9361d2e6e6fa02 web/reference_audio_ui.js ae59273f163a116ece418957c557c4fd8c3285e3c60814679b9361d2e6e6fa02 web/reference_audio_ui.js
46c9f75c9462bf26300332448f4431cca801f38964ca0a49bc2c50a88a7cc40f web/reference_image_ui.js 46c9f75c9462bf26300332448f4431cca801f38964ca0a49bc2c50a88a7cc40f web/reference_image_ui.js
e9d7753f313ebbe3d13e477f5b4e4767a735b0619b52ba54bcc5218f3eb7eb7c web/reference_images_v39.js e9d7753f313ebbe3d13e477f5b4e4767a735b0619b52ba54bcc5218f3eb7eb7c web/reference_images_v39.js
@@ -417,3 +417,13 @@ bf41bf08dd0504eb753fa5781e6c2b8c1285758ad4bb7309975bb9231d5f8e41 web/video_refe
a9f7b67bd65a9beda622aa5fdc5f8a2ecae1b58b18690d91fb0107a16a3e384e write-h3-v39-reference-prompts/references/simple-3x5.txt a9f7b67bd65a9beda622aa5fdc5f8a2ecae1b58b18690d91fb0107a16a3e384e write-h3-v39-reference-prompts/references/simple-3x5.txt
81d7b4f4da797339b46911bc96d0a3736d70df29e092715c6684bb1786908593 write-h3-v39-reference-prompts/references/structured.md 81d7b4f4da797339b46911bc96d0a3736d70df29e092715c6684bb1786908593 write-h3-v39-reference-prompts/references/structured.md
3dfd0e7042070f62b9e78239c0e75d82a7df32098db10832465f3049b7f0b38f write-h3-v39-reference-prompts/SKILL.md 3dfd0e7042070f62b9e78239c0e75d82a7df32098db10832465f3049b7f0b38f write-h3-v39-reference-prompts/SKILL.md
7fd122ff08232e114a6ef32171217ea6324e980eb7257d560c466b8177a3d3ec audio_compat.py
82efab13baacb9a229671eda5f4a2d24bb839b160b2bfb40591df51a5ddf80a4 tests/fixtures/v39_sampling_v5/project.json
8b8ca1b0c802d024c6915d903747d8722bea25628369997ab008ff896462b96a tests/fixtures/v39_sampling_v5/revisions/d8352287737286e5/chunks/rev-d8352287737286e5-txn-008f52b93a88461481ead25f91627af0-chunk-0002.safetensors
26e771cffc57f171f4cc4c6859876350bdf717d1e8213fc005343006a34f002f tests/fixtures/v39_sampling_v5/revisions/d8352287737286e5/chunks/rev-d8352287737286e5-txn-9bb0a584b7344342b114bd31b175225c-chunk-0001.safetensors
7bf0cf46f049e072a68e71cf8bbab10aef695f0931b16842a890c084021f5a11 tests/fixtures/v39_sampling_v5/revisions/d8352287737286e5/manifest.json
2e09cde80073085ac6dda5cdaf50ea9d76f1c9db15cd1b444e13e50cea8fe210 tests/test_v39_stability_r1.py
d924698cdaea82dd7212d356e368cf9701231c6a4d1f9794f1e28e7b36125124 tests/test_v39_stability_r2.py
d2e59eae0a724626b0a4e361921643dd31cb3612ae1f0d1b8b355c9d42c06174 tests/test_v39_stability_r3.py
1dc932ec0ee825a8ad95509824d225c6bd091344229da7e1ff361b7512f190d9 tests/test_v39_stability_r4_r5.py
319258798ca198312cf684f369990924657d1af168a148ad34303af31c06a350 tests/test_v39_review_refine_scope.py
+23 -1938
View File
File diff suppressed because it is too large Load Diff
+35 -6
View File
@@ -1,4 +1,33 @@
# ComfyUI-H3-Continuum 3.9.0 — V3.9 # ComfyUI-H3-Continuum 3.9.1 — V3.9
## 3.9.1 — what changed (2026-10-04)
This patch fixes Second Pass conditioning, Take reuse and Review audio:
- **Second Pass:** preserve audio-only/mixed keyframes and inherit the verified conditioning and Reference assignments of the actual Review output, including a single middle chunk.
- **Take reuse:** include sampler closures and MODEL CFG/wrappers in compatibility checks to prevent reuse under different generation settings.
- **Review Driving Audio:** cut the source PCM once at the physical group's natural time, with the same start position for Exact ON/OFF.
- **Audio resampling:** prefer ComfyUI Core's standard API, with the legacy fallback for older Core versions. The Reference Encode Cache limitation for direct VAE weight changes is documented below.
**Updating from 3.9.0:** restart ComfyUI and refresh the browser after updating. Existing v5 Takes and history remain readable and are not deleted, but cannot be reused as prefixes for new v6 generation. Start a new Full Video or choose **Start again from Chunk 1**. Public node IDs, widget order and official workflows remain unchanged; keep using the **V3.9** nodes.
The repaired code passed **1,696 CPU tests / 1 skipped**, targeted Review Second Pass GPU checks and an integrated generation/reuse check. Browser acceptance remains pending. Optional latent-upscale doubled outlines are tracked separately in [Issue #27](https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum/issues/27); they are not a required gate for these fixes. This is a main-source update, not a new GitHub Release or Registry publication.
## 3.9.1 repair details
Second Pass preserves audio-only keyframes and adapts only the video component of mixed keyframes. Review Driving Audio uses the retained natural frame range of the physical decode group, measured before projection. Exact ON/OFF share that PCM origin. Completely empty Review intervals receive silence for the selected video duration; partially short audio stays short with Exact OFF.
New generation uses Sampling Contract v6 and Graph Contract v4, including sampler code/closures and MODEL CFG/wrapper settings. Saved v5 Takes, raw tensors and provenance remain readable, but cannot be reused in v6 generation. Start a new Full Video or use Review's **Start again from Chunk 1** to create a new revision. Old Takes remain intact. Unobservable external settings still allow fresh generation with reuse disabled; incompatible explicit Take/Retry requests are not replaced with a different Take.
Reference/Driving resampling prefers Core's standard API and uses legacy TorchAudio only when that API is absent. Preserved output PCM keeps its original waveform and sample rate.
**Reference Encode Cache limitation:** direct weight mutation inside the same VAE object is not detected automatically. Call `v3.ref_encode_cache.clear_ref_encode_cache()` afterwards or reload into a new VAE object. Unchanged VAE/input pairs continue to hit the cache. Decode Cache Helper reset affects a separate cache and does not clear Reference Encode Cache.
This repair stays within V3.9: public node IDs, socket/widget order, official workflows, Run Storage v3 and State/Session formats are retained. LoRA Plan is not included. CPU checks and browser/GPU acceptance are separate.
V3.9 also projects captured conditioning onto the actual Review output groups before Second Pass. A partial prefix or a single middle chunk can inherit its own conditioning and Reference assignments when the context validates. Full-sequence completion and saved Takes are unchanged; missing or mismatched context keeps the diagnostic prompt-only fallback.
The repaired runtime passed **1696 CPU tests, 1 skipped**, and three targeted GPU cases at **512×512 / 24 fps**: a two-group partial prefix, Chunk 2 alone, and complete three-group reuse. All groups inherited verified conditioning; saved FLAC matched the corresponding source PCM exactly and existing Takes/raw files were preserved. This validates the targeted repair. Browser acceptance remains pending; optional latent-upscale double outlines are a separate investigation in Issue #27. Historical V3.8 releases and tags remain available unchanged.
![H3 Continuum V3.9: fixed Reference Image slots, per-chunk assignments, and one Sampler input](docs/images/v39/v39-feature-summary.png) ![H3 Continuum V3.9: fixed Reference Image slots, per-chunk assignments, and one Sampler input](docs/images/v39/v39-feature-summary.png)
@@ -46,9 +75,9 @@ The selected V3.9 template has 32 nodes. Image loaders 1–6 are wired to the ne
In the helper, `All chunks` uses connected images throughout; `Per chunk` provides an image-by-chunk table. The template starts in `Per chunk` with all assignments off: after enabling a loader, tick the chunks that should use it. Slot numbers never shift: Image 1 is `@R1` and Image 9 is `@R9`, even when intervening slots are empty. In Sequence Prompt, describe the action for each chunk and use `@R` tags only for active References; their actual `<Picture N>` numbers are resolved per group. Unused tags warn without stopping generation. Start with short chunks and pre-resized images: source images may consume RAM before the Sampler's Reference Image Size setting applies. In the helper, `All chunks` uses connected images throughout; `Per chunk` provides an image-by-chunk table. The template starts in `Per chunk` with all assignments off: after enabling a loader, tick the chunks that should use it. Slot numbers never shift: Image 1 is `@R1` and Image 9 is `@R9`, even when intervening slots are empty. In Sequence Prompt, describe the action for each chunk and use `@R` tags only for active References; their actual `<Picture N>` numbers are resolved per group. Unused tags warn without stopping generation. Start with short chunks and pre-resized images: source images may consume RAM before the Sampler's Reference Image Size setting applies.
V3.9.0 is the current source version on `main`. This source update is **not** a new GitHub Release or ComfyUI Registry publication; existing Releases and tags are left in place. Configured 1024×1024 Reference-only GPU/API runs succeeded, but the unchanged default template and browser save/reload remain separate acceptance gates. For a reproducible older environment, use the matching historical Release/tag and workflow; the retained V3.8 Sampler on current `main` specifically supports V3.8X2 workflows. V3.9.1 is the current source version on `main`. This source update is **not** a new GitHub Release or ComfyUI Registry publication; existing Releases and tags are left in place. Configured 1024×1024 Reference-only GPU/API runs succeeded, but the unchanged default template and browser save/reload remain separate acceptance gates. For a reproducible older environment, use the matching historical Release/tag and workflow; the retained V3.8 Sampler on current `main` specifically supports V3.8X2 workflows.
`pyproject.toml` declares ComfyUI `>=0.32.0`; earlier real-generation/GPU validation used ComfyUI `0.34.2`, and the scoped Fixed cache comparison above used `0.38.0`. The minimum declaration does not establish V3.9 runtime acceptance on 0.32.0. The latest recorded Windows CPU run is `1569 passed / 1 skipped / 0 failed`; the earlier separate Linux rerun report records `1534 passed / 3 skipped / 0 failed` and does not include the new cache or external INT tests. Neither replaces browser save/reload or GPU acceptance for the unchanged official template. `pyproject.toml` declares ComfyUI `>=0.32.0`; earlier real-generation/GPU validation used ComfyUI `0.34.2`, and the scoped Fixed cache comparison above used `0.38.0`. The minimum declaration does not establish V3.9 runtime acceptance on 0.32.0. The latest recorded Windows CPU run is `1696 passed / 1 skipped / 0 failed`; the earlier separate Linux rerun report records `1534 passed / 3 skipped / 0 failed` and does not include the new cache or external INT tests. Neither replaces browser save/reload or GPU acceptance for the unchanged official template.
## V3.9 Reference Images ## V3.9 Reference Images
@@ -84,7 +113,7 @@ The [official V3.9 workflow](examples/workflows/MiniMax_H3_Continuum_V39.json) i
**Compatibility:** Existing V3.8X2 workflows still use the V3.8 Sampler and old Reference Images 4–9 helper. V3.9 needs its new helper and connection. Open the new V3.9 workflow instead of only changing the Sampler type in an old graph: older V3.9 prototype workflows with direct 1–3 connections and Sampler-side routing widgets need manual rewiring, and their saved widget positions may differ. Accepted V3.8 Runs/Takes are not automatically migrated. Keep a copy of the previous workflow and use a new Run Name for V3.9. **Compatibility:** Existing V3.8X2 workflows still use the V3.8 Sampler and old Reference Images 4–9 helper. V3.9 needs its new helper and connection. Open the new V3.9 workflow instead of only changing the Sampler type in an old graph: older V3.9 prototype workflows with direct 1–3 connections and Sampler-side routing widgets need manual rewiring, and their saved widget positions may differ. Accepted V3.8 Runs/Takes are not automatically migrated. Keep a copy of the previous workflow and use a new Run Name for V3.9.
The earlier RR-R6B browser PASS covered the old V3.9 controls. The new helper has configured GPU/API execution evidence; its browser save/reload and the unchanged default template's full acceptance are still open. CPU/JS and manifest results are separate; no V3.9 Release/Registry publication is implied. Earlier browser checks covered the previous V3.9 controls. The new helper has GPU/API execution evidence; its browser save/reload and the unchanged default template's full acceptance are still open. CPU/JS and manifest results are separate; no V3.9 Release/Registry publication is implied.
**Version boundary:** V3.8X2 package 3.8.3 and V3.9.0 are separate workflows. This update does not create a V3.8X2 or V3.9 GitHub Release, tag, maintenance branch, or Registry version; it does not remove any older Release. The existing [`v3.8.0` tag](https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum/tree/v3.8.0) remains the reproducible V3.8.0 baseline. **Version boundary:** V3.8X2 package 3.8.3 and V3.9.0 are separate workflows. This update does not create a V3.8X2 or V3.9 GitHub Release, tag, maintenance branch, or Registry version; it does not remove any older Release. The existing [`v3.8.0` tag](https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum/tree/v3.8.0) remains the reproducible V3.8.0 baseline.
@@ -823,7 +852,7 @@ Prompt/CLIP figures measure only the conditioning subphase, not total generation
The measured Sage-only production baselines on the tested RTX 5060 Ti 16 GB / 64 GB system were 168.069 seconds for 1×5-second 576×576 T2VA and 379.765 seconds for 3×5-second 640×640 FL2VA Long Terminal Merge. These are configuration-specific baselines, not universal speed guarantees. Sampling remained the dominant cost; Continuum Assemble + Seam stayed below 1%. The measured Sage-only production baselines on the tested RTX 5060 Ti 16 GB / 64 GB system were 168.069 seconds for 1×5-second 576×576 T2VA and 379.765 seconds for 3×5-second 640×640 FL2VA Long Terminal Merge. These are configuration-specific baselines, not universal speed guarantees. Sampling remained the dominant cost; Continuum Assemble + Seam stayed below 1%.
> **V3.8.0 is the historical release baseline.** V3.8X2 remains a separate package-3.8.3 workflow, while V3.9.0 source is available on `main`. Historical implementation modules remain in source because V3.8X2 reuses them internally. The current registration has ten V3.8X2 IDs plus two V3.9 IDs. Use the matching historical Release/tag for workflows requiring other IDs. Still Image Guide remains Experimental. > **V3.8.0 is the historical release baseline.** V3.8X2 remains a separate package-3.8.3 workflow, while V3.9 source is available on `main`. Historical implementation modules remain in source because V3.8X2 reuses them internally. The current registration has ten V3.8X2 IDs plus two V3.9 IDs. Use the matching historical Release/tag for workflows requiring other IDs. Still Image Guide remains Experimental.
## V3.5.1 Reference Audio & Compatibility Update ## V3.5.1 Reference Audio & Compatibility Update
@@ -1443,7 +1472,7 @@ Validation results apply to the tested local source and environment. They do not
## Limits ## Limits
- Second Pass / `refine_context` is not supported while a Review sequence is partial. Finish Review before starting Second Pass. - V3.8 Second Pass / `refine_context` is not supported while a Review sequence is partial. V3.9 supports validated output-group conditioning for partial Review as described above.
- The 16GB GPU Gate passed with small VRAM headroom; memory use depends on the complete environment and workflow. - The 16GB GPU Gate passed with small VRAM headroom; memory use depends on the complete environment and workflow.
- Continuation does not guarantee frame-perfect identity or motion. - Continuation does not guarantee frame-perfect identity or motion.
- Video Guide Frames guides H3; it does not reproduce every source frame. - Video Guide Frames guides H3; it does not reproduce every source frame.
+35 -6
View File
@@ -1,9 +1,38 @@
# ComfyUI-H3-Continuum 3.9.0 — V3.9 # ComfyUI-H3-Continuum 3.9.1 — V3.9
## 3.9.1で変わった点(2026-10-04)
Second Pass、Takeの再利用、Review音声に関するバグを修正しました。
- **Second Pass**:音声のみ/複合キーフレームを保持し、途中Reviewや中間Chunk単独でも、その出力区間のconditioningとReference割当を照合して引き継ぎます。
- **Takeの再利用**:Samplerのclosure、MODELのCFG/wrapperも互換性判定に含め、異なる生成設定のTakeを誤って再利用しないようにしました。
- **ReviewのDriving Audio**:物理groupの自然時間に合わせて元PCMを1回だけ切り出します。Exact ON/OFFで開始位置を揃えます。
- **音声resample**:ComfyUI Core標準APIを優先し、古いCoreでは従来方式へ戻します。同じVAE内部の重みを直接変更する場合のReference Encode Cacheの制約も下記に明記しました。
**3.9.0から更新する場合**:更新後にComfyUIを再起動し、ブラウザーも更新してください。旧v5 Takeと履歴は読めて、削除されませんが、新しいv6生成の前半として再利用できません。新規Full Video、または **Start again from Chunk 1** で開始してください。公開Node ID・widget順・公式Workflowは維持し、引き続き **V3.9ノード** を使います。
修正済みコードは **CPU 1,696成功・1スキップ**、対象を絞ったReview Second PassのGPU検証、生成/Take再利用の統合GPU検証に合格しています。ブラウザー受入は未完了です。任意のLATENT拡大で出る二重輪郭は別課題の [Issue #27](https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum/issues/27) として管理し、今回の修正の必須Gateには含めません。今回もmainソースの更新で、GitHub Release/Registry公開ではありません。
![H3 Continuum V3.9:固定の参照画像番号、チャンク別割り当て、Samplerへの1本の接続](docs/images/v39/v39-feature-summary.png) ![H3 Continuum V3.9:固定の参照画像番号、チャンク別割り当て、Samplerへの1本の接続](docs/images/v39/v39-feature-summary.png)
*現在の`main`はV3.9です。Image 1~9の固定番号、チャンク別のReference選択、V3.9 Samplerへの1本の接続が新しい操作です。* *現在の`main`はV3.9です。Image 1~9の固定番号、チャンク別のReference選択、V3.9 Samplerへの1本の接続が新しい操作です。*
## 3.9.1の修正詳細
Second Passで音声のみのキーフレームを保持し、複合キーフレームでは映像だけを空間適応します。ReviewのDriving Audioは、区間選択前の物理decode groupの自然時間から元PCMを切り出します。Exact ON/OFFの開始位置は共通です。完全に空のReview音声だけは選択映像の尺に合う無音を補い、部分的に短い音声はExact OFFで引き延ばしません。
新しい生成はSampling Contract v6/Graph Contract v4を使用します。Samplerのコード・closureとMODELのCFG/wrapper等の設定も再利用判定へ含めます。旧v5のTake・raw・生成履歴は読めますが、v6生成へ継ぎ足せません。新規Full Video、またはReviewの「Start again from Chunk 1」で新しいrevisionを作成してください。旧Takeは残ります。署名を観測できない外部設定でも新規生成は可能ですが、自動再利用は無効になります。明示した不適合TakeやRetryを別Takeへ置き換えません。
Reference/DrivingのresampleはCore標準APIを優先し、そのAPIがない旧Coreだけ従来のTorchAudioを使います。最終出力へ渡す元PCMの波形とsample rateは保持します。
**Reference Encode Cacheの制約**:同じVAEオブジェクトの内部重みを直接書き換えた場合、その変更を自動検知しません。変更後は`v3.ref_encode_cache.clear_ref_encode_cache()`を呼ぶか、新しいVAEオブジェクトへ再ロードしてください。通常の同一VAE・同一入力ではcache HITを維持します。Decode Cache Helperのresetは別のcacheを対象とし、Reference Encode Cacheを消しません。
この修正はV3.9内の変更です。公開Node ID・socket・widget順、公式Workflow、Run Storage v3、State/Session形式を維持し、LoRA Planは追加しません。CPU検証と実ブラウザー・GPU受入は別です。
V3.9では、Second Passへ渡す条件を実際のReview出力区間へ対応させます。途中までの出力や中間Chunk単独でも、条件の照合に成功すれば、その区間自身のconditioningとReference割当を継承します。全体の完了状態や保存済みTakeは変更しません。条件が不足・不一致の場合は、従来の診断付きprompt-only fallbackを維持します。
修正済みランタイムは**CPU 1696成功・1スキップ**、**512×512・24fpsのGPU 3ケース**(途中2区間、Chunk 2単独、完了済み3区間の再利用)で合格しました。全区間の条件継承を確認し、保存FLACと元PCMの対応区間は完全一致、既存Takeとrawファイルも保持しています。合格対象は修正対象の動作です。ブラウザー受入は未完了で、任意のLATENT拡大時の二重輪郭はIssue #27で別途調査します。過去のV3.8 Release/tagはそのまま利用できます。
## V3.9の基本操作を動画で見る ## V3.9の基本操作を動画で見る
[H3 Continuum V3.9 — Getting Started: Chunks, Prompts & Reference Images](https://youtu.be/AKJxBXaiG6Q)(12分26秒) [H3 Continuum V3.9 — Getting Started: Chunks, Prompts & Reference Images](https://youtu.be/AKJxBXaiG6Q)(12分26秒)
@@ -46,9 +75,9 @@ ComfyUIの **Templates → ComfyUI-H3-Continuum** には、今回選んだ[V3.9
ヘルパーの `All chunks` は接続画像を全チャンクに、`Per chunk` は画像×チャンクの表で指定します。このテンプレートは`Per chunk`で全チェックOFFから始まるため、Loaderを有効にした後、使用したいチャンクにもチェックしてください。空欄があっても番号はずれず、Image 1は常に`@R1`、Image 9は`@R9`です。Sequence Promptにはチャンクごとの動作を短く書き、そのチャンクで使う画像の`@R`タグを付けます。内部の`<Picture N>`番号はgroupごとに自動変換されます。使わないタグは警告しますが生成を止めません。まず短いチャンクと縮小済み画像で試してください。入力元画像はSamplerのReference Image Size設定より前にRAMを使うことがあります。 ヘルパーの `All chunks` は接続画像を全チャンクに、`Per chunk` は画像×チャンクの表で指定します。このテンプレートは`Per chunk`で全チェックOFFから始まるため、Loaderを有効にした後、使用したいチャンクにもチェックしてください。空欄があっても番号はずれず、Image 1は常に`@R1`、Image 9は`@R9`です。Sequence Promptにはチャンクごとの動作を短く書き、そのチャンクで使う画像の`@R`タグを付けます。内部の`<Picture N>`番号はgroupごとに自動変換されます。使わないタグは警告しますが生成を止めません。まず短いチャンクと縮小済み画像で試してください。入力元画像はSamplerのReference Image Size設定より前にRAMを使うことがあります。
V3.9.0は`main`の現行ソースです。今回のmain更新は新しいGitHub ReleaseやComfyUI Registry公開ではなく、過去のReleaseとtagは残します。設定済み1024×1024のReference-only GPU/APIテストは通っていますが、無改変の公式テンプレート初期設定とブラウザー保存・再読込は別の受入項目です。古い環境を厳密に再現するには対応する過去のRelease/tagとWorkflowを使用してください。現行mainにもV3.8 Samplerを残し、V3.8X2 Workflowを使えるようにしています。 V3.9.1は`main`の現行ソースです。今回のmain更新は新しいGitHub ReleaseやComfyUI Registry公開ではなく、過去のReleaseとtagは残します。設定済み1024×1024のReference-only GPU/APIテストは通っていますが、無改変の公式テンプレート初期設定とブラウザー保存・再読込は別の受入項目です。古い環境を厳密に再現するには対応する過去のRelease/tagとWorkflowを使用してください。現行mainにもV3.8 Samplerを残し、V3.8X2 Workflowを使えるようにしています。
`pyproject.toml`の宣言上はComfyUI `>=0.32.0`ですが、以前の実生成・GPU検証はComfyUI `0.34.2`、上記の限定したFixedキャッシュ比較は`0.38.0`で行いました。最低宣言版でV3.9の実機動作を確認済みという意味ではありません。最新の記録済みWindows CPU試験は`1569 passed / 1 skipped / 0 failed`、以前の別環境Linux再実行報告は`1534 passed / 3 skipped / 0 failed`で、新しいキャッシュ/外部INT試験は含みません。これらは無改変の公式テンプレートの実ブラウザー保存・再読込やGPU受入を代替しません。 `pyproject.toml`の宣言上はComfyUI `>=0.32.0`ですが、以前の実生成・GPU検証はComfyUI `0.34.2`、上記の限定したFixedキャッシュ比較は`0.38.0`で行いました。最低宣言版でV3.9の実機動作を確認済みという意味ではありません。最新の記録済みWindows CPU試験は`1696 passed / 1 skipped / 0 failed`、以前の別環境Linux再実行報告は`1534 passed / 3 skipped / 0 failed`で、新しいキャッシュ/外部INT試験は含みません。これらは無改変の公式テンプレートの実ブラウザー保存・再読込やGPU受入を代替しません。
## V3.9 Reference Images ## V3.9 Reference Images
@@ -84,7 +113,7 @@ Promptは実行を止めません。明示的なTimelineでも解析できなけ
**互換性:** V3.8X2の既存Workflowは旧V3.8 Samplerと旧4~9ヘルパーで引き続き利用します。V3.9では新ヘルパーへの接続が必要です。旧Workflow上でSamplerの型だけ変えず、上記のV3.9専用Workflowを開いてください。以前のV3.9試作WorkflowにあったSampler直結1~3やSampler側の割り当て欄は配線と保存widgetの位置が異なります。V3.8の途中Run/Takeは自動移行しません。元Workflowを保存し、V3.9では別Run Nameを使ってください。 **互換性:** V3.8X2の既存Workflowは旧V3.8 Samplerと旧4~9ヘルパーで引き続き利用します。V3.9では新ヘルパーへの接続が必要です。旧Workflow上でSamplerの型だけ変えず、上記のV3.9専用Workflowを開いてください。以前のV3.9試作WorkflowにあったSampler直結1~3やSampler側の割り当て欄は配線と保存widgetの位置が異なります。V3.8の途中Run/Takeは自動移行しません。元Workflowを保存し、V3.9では別Run Nameを使ってください。
旧V3.9操作系のRR-R6BブラウザーPASSは、新しいヘルパーのブラウザー保存・再読込PASSを意味しません。新ヘルパーは設定済みGPU/API実行の証拠がありますが、公式テンプレート初期設定の受入は別です。CPU/JS・Manifestも別に判定し、V3.9 GitHub Release/Registry公開はまだ行っていません。 以前のV3.9操作系のブラウザー確認は、新しいヘルパーの保存・再読込確認とは別です。新ヘルパーは設定済みGPU/API実行の証拠がありますが、公式テンプレート初期設定の受入は別です。CPU/JS・Manifestも別に判定し、V3.9 GitHub Release/Registry公開はまだ行っていません。
**バージョン境界:** V3.8X2(package 3.8.3)とV3.9.0は別Workflowです。今回、新しいtag・GitHub Release・保守ブランチ・Registry版は作成せず、既存Releaseも削除しません。旧V3.8.0は既存の`v3.8.0` tagから取得できます。 **バージョン境界:** V3.8X2(package 3.8.3)とV3.9.0は別Workflowです。今回、新しいtag・GitHub Release・保守ブランチ・Registry版は作成せず、既存Releaseも削除しません。旧V3.8.0は既存の`v3.8.0` tagから取得できます。
@@ -323,7 +352,7 @@ Long Terminal Mergeのpairは分割せず、1つのReview単位として扱い
Driving AudioはReview/Smart Regenerateと併用できます。Continuum Image/Audio/Video Loaderは任意入力向けのnative bypassに対応します。 Driving AudioはReview/Smart Regenerateと併用できます。Continuum Image/Audio/Video Loaderは任意入力向けのnative bypassに対応します。
> **Reviewの制限:** Review途中のpartial sequenceでは、Second Pass/`refine_context`を正式対応範囲に含めません。Reviewを完了してsequenceを確定してからSecond Passを実行してください。 > **V3.8 Reviewの制限:** Review途中のpartial sequenceでは、Second Pass/`refine_context`を正式対応範囲に含めません。Reviewを完了してsequenceを確定してからSecond Passを実行してください。V3.9では、上記のとおり検証済みの出力区間条件を継承できます。
### Take・Branch・安全な継続 ### Take・Branch・安全な継続
@@ -414,7 +443,7 @@ Prompt/CLIPの数値はconditioning区間だけで、総生成時間ではあり
RTX 5060 Ti 16 GB/RAM 64 GBの検証環境で測定したSage-only Production baselineは、576×576 T2VA 1×5秒が168.069秒、640×640 FL2VA Long Terminal Merge 3×5秒が379.765秒です。環境・設定固有の測定値であり、すべての環境に対する速度保証ではありません。Samplingが最大コストで、Continuum Assemble + Seamは1%未満でした。 RTX 5060 Ti 16 GB/RAM 64 GBの検証環境で測定したSage-only Production baselineは、576×576 T2VA 1×5秒が168.069秒、640×640 FL2VA Long Terminal Merge 3×5秒が379.765秒です。環境・設定固有の測定値であり、すべての環境に対する速度保証ではありません。Samplingが最大コストで、Continuum Assemble + Seamは1%未満でした。
**V3.8.0はhistorical release baselineです。V3.8X2はpackage 3.8.3の別Workflowとして維持し、V3.9.0はmainの現行ソースです。** V3.8X2が内部利用する旧module/classはsourceへ維持します。現在の登録面はV3.8X2の10 IDとV3.9追加の2 IDです。一部は旧IDを維持しますが、それ以外のIDを必要とする旧保存Workflowは対応するhistorical Release/tagを使用してください。Still Image Guideは引き続きExperimentalです。 **V3.8.0はhistorical release baselineです。V3.8X2はpackage 3.8.3の別Workflowとして維持し、V3.9はmainの現行ソースです。** V3.8X2が内部利用する旧module/classはsourceへ維持します。現在の登録面はV3.8X2の10 IDとV3.9追加の2 IDです。一部は旧IDを維持しますが、それ以外のIDを必要とする旧保存Workflowは対応するhistorical Release/tagを使用してください。Still Image Guideは引き続きExperimentalです。
## V3.5.1 Reference Audio/互換性更新 ## V3.5.1 Reference Audio/互換性更新
+19 -18
View File
@@ -1,7 +1,7 @@
dca65fb1393e67f59c6f81f9411e3e745eb8f1097e9a2adf7ffc6f4e3901ca9f __init__.py dca65fb1393e67f59c6f81f9411e3e745eb8f1097e9a2adf7ffc6f4e3901ca9f __init__.py
54ba2a58388c42063d24335aa2737e86bd4c15fe9632dcc9ffba18c328e04641 .comfyignore 54ba2a58388c42063d24335aa2737e86bd4c15fe9632dcc9ffba18c328e04641 .comfyignore
3e7a1ec3380be2a222c8a3e9b8a6bb184d3a760478957f48186438f15235bdba branch_provenance.py 3e7a1ec3380be2a222c8a3e9b8a6bb184d3a760478957f48186438f15235bdba branch_provenance.py
593ec5bafb8e506cc7b326bdc7e3f024a142569f24f8d39c1aaa3b5acfd0beca CHANGELOG.md 16f886cd4b296290880e549a3f4683308246dd1fd93e479c3bbd6ab29307beda CHANGELOG.md
c1ed3a0342730450958fb62fd21854a876991f3a4aefb24e5383b0c73cd2312d compatibility.py c1ed3a0342730450958fb62fd21854a876991f3a4aefb24e5383b0c73cd2312d compatibility.py
4d62508f900af9f93ec320afb900c27cc9a300d1e0c0a94714258498c5ede5fc conditioning.py 4d62508f900af9f93ec320afb900c27cc9a300d1e0c0a94714258498c5ede5fc conditioning.py
294251c40b1be2e296c5e5f627b68091d087bdb39179e3a0c3cebd2ee8a344f5 constants.py 294251c40b1be2e296c5e5f627b68091d087bdb39179e3a0c3cebd2ee8a344f5 constants.py
@@ -22,7 +22,7 @@ b30fe4640359bebfcdeeb94dbeec94cb346c8e3646509e1dcdb87d7021ba04ff docs/N2B_BRANC
717577efdec5e61600504b77bc2a099ef022c4fe2333e936c3310c7182d8f2a0 docs/V38_OPEN_INTEGRATION_CONTRACT.md 717577efdec5e61600504b77bc2a099ef022c4fe2333e936c3310c7182d8f2a0 docs/V38_OPEN_INTEGRATION_CONTRACT.md
3496636f645b39336fd95a4f21a562810c56f8e1d92fa6eeb1e4b0786ad11b37 docs/V38_RELEASE_AND_MIGRATION.md 3496636f645b39336fd95a4f21a562810c56f8e1d92fa6eeb1e4b0786ad11b37 docs/V38_RELEASE_AND_MIGRATION.md
4d17d74107e78404b190fec9494b443249e5aff4f58670c50d54dfdfb064334b docs/V39_REFERENCE_INLINE.md 4d17d74107e78404b190fec9494b443249e5aff4f58670c50d54dfdfb064334b docs/V39_REFERENCE_INLINE.md
cedd7b07901d6065079ef37380f0062d535a39c98c8f614e1b26582f549d7183 driving_audio.py a8c56379abe23b98c8ceed6ea5e60a02ecc3c6d9636232d7371615f4eb306e81 driving_audio.py
1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/MiniMax_H3_Continuum_V38X2.json 1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/MiniMax_H3_Continuum_V38X2.json
40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/MiniMax_H3_Continuum_V39.json 40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/MiniMax_H3_Continuum_V39.json
1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/workflows/MiniMax_H3_Continuum_V38X2.json 1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/workflows/MiniMax_H3_Continuum_V38X2.json
@@ -33,29 +33,29 @@ f3d8dd0f0872afdd97fe24e58b5cb74190d03e3583a66a5d013fee41f6894cda examples/workf
65689fd800a60ab6939bea020d630616b03a453cf6d48022cd1361992a765abc examples/workflows/MiniMax_H3_Continuum_V39_Reference_Inline.zip 65689fd800a60ab6939bea020d630616b03a453cf6d48022cd1361992a765abc examples/workflows/MiniMax_H3_Continuum_V39_Reference_Inline.zip
40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/workflows/MiniMax_H3_Continuum_V39.json 40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/workflows/MiniMax_H3_Continuum_V39.json
f69b734b5bf19fb6db78bbe4737db0bef69c2a6cef2710f28767882ef35ec847 examples/workflows/MiniMax_H3_Continuum_V39.zip f69b734b5bf19fb6db78bbe4737db0bef69c2a6cef2710f28767882ef35ec847 examples/workflows/MiniMax_H3_Continuum_V39.zip
7861fac4c8fe82372c5459aa574d35ea003bdb4fd846d567bd891c2e38571b3d graph_contract.py a344da581b3c2d77558a460d2b5256e6c9a63b5008c98b52999ceda2e0b2b767 graph_contract.py
bf0556b076dbaca67c3405a7d86c333ca7fdf3d4381529b0a9edc906519943e9 guide_timeline.py bf0556b076dbaca67c3405a7d86c333ca7fdf3d4381529b0a9edc906519943e9 guide_timeline.py
1473c384082700f422f2cf9141e0d012d87d2b43ac8a8a433edc7172e6ca7bf6 hardening.py 1473c384082700f422f2cf9141e0d012d87d2b43ac8a8a433edc7172e6ca7bf6 hardening.py
9fca2d9c04be387160dee84e5ee9ce538cb910e79e356b9236e6d2240ea6d4ee INSTALL_JA.txt 199f946243bed42786319517f26e628612877df61ce36d55552485155b5df44b INSTALL_JA.txt
23bddc5a41d2e997f4fbf789c7128e32f4f197f1405b2aef1c76794f57482446 install_windows.bat 23bddc5a41d2e997f4fbf789c7128e32f4f197f1405b2aef1c76794f57482446 install_windows.bat
12bf572a65d57bbc9a72b459f29691c2057768f4fb7a74f9db87f4775e995bda install_windows.ps1 12bf572a65d57bbc9a72b459f29691c2057768f4fb7a74f9db87f4775e995bda install_windows.ps1
139546ec1ff459945aa89b50f74b5ee69f62a0e07bbddfcae06936091a409a83 layout_adapter.py 139546ec1ff459945aa89b50f74b5ee69f62a0e07bbddfcae06936091a409a83 layout_adapter.py
c44abbd789091bbead01250dbd5453e60539d2665c69f162335fdbd2936019c2 LICENSE c44abbd789091bbead01250dbd5453e60539d2665c69f162335fdbd2936019c2 LICENSE
d3e213005670e94dee33a240349a4d245cf9a3b6602099f11ed30cf5a32e2c21 media.py d3e213005670e94dee33a240349a4d245cf9a3b6602099f11ed30cf5a32e2c21 media.py
8ce5a7c255088fb6038925a93b71b87972bf2f02e450af86da5c317e95506c41 metadata.ini ad0835f8fb6d171d8ff8d7295fb48291abc93ef02fd46a2e283265c15b93366a metadata.ini
d7e613e2cc24110213b477bdd1d390234a41f2af3b1ed006a809066d71d658eb model_patch.py d7e613e2cc24110213b477bdd1d390234a41f2af3b1ed006a809066d71d658eb model_patch.py
ecb65a5dbf9e1683c5a3cadd82daaea3a90a2b776f6ebabfbe8b07aa93177422 nodes.py ecb65a5dbf9e1683c5a3cadd82daaea3a90a2b776f6ebabfbe8b07aa93177422 nodes.py
99405e4357a143b0f5d0b374c9c0d2089435a7055bea2f3d1216dbbad25e4339 PACKAGE_INFO_JA.txt 99405e4357a143b0f5d0b374c9c0d2089435a7055bea2f3d1216dbbad25e4339 PACKAGE_INFO_JA.txt
ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt
b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py
667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml 1047d05b0aedc405e3a36ee571a8939111d69bb5565e09a3e1caa3c40bf33170 pyproject.toml
47f5e57dc6c84b0103d89d6549e87ad45ab4c263cadb36f21a952843a65204a7 README_JA.md cea43f63792a28b9d19035b3ee64ff482cf0357aec9edd6fab34eb84956b2973 README_JA.md
b89e0cc5348a8dbab8781a1b57c99c4c42f47598efe218cc087a77bc58419839 README.md 828fc5d51539c3bb5c4bddaa741b25fcda3faf2c1dec89ed545a63247b7b02a1 README.md
44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py 9539f5cb9fb695841073fa58d9246d5a7410bc372f6cd340378185004dfaf978 reference_audio.py
5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py 5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py
caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py
467edb2c188d024c2392aafe7423b3f835e16b10f570da76e2f7dda8a962c1c5 requirements.txt 467edb2c188d024c2392aafe7423b3f835e16b10f570da76e2f7dda8a962c1c5 requirements.txt
53827c94b3f588f35bb8eacb404e1bd5e233d25f4ee0fdd286efc7d8ddb52b6f run_storage.py 1845cbf7ba4d3b35afd087a15f5a4d92dc73d4858e61982f8ccf431a836a22b1 run_storage.py
6056de6c0a5322a8a51ba5a114538dca2ef67788cd1afba10fedd483b95f6afa state_io.py 6056de6c0a5322a8a51ba5a114538dca2ef67788cd1afba10fedd483b95f6afa state_io.py
3a359a0467725cc292ccb4c86cf0c3d30bb6a114d01f5345792b09ea3d45441d state.py 3a359a0467725cc292ccb4c86cf0c3d30bb6a114d01f5345792b09ea3d45441d state.py
43ec4023dc06c941e5d7a04db4b8a89fb58d744ceb986f6eb9bf534d4fdcb87b temporal.py 43ec4023dc06c941e5d7a04db4b8a89fb58d744ceb986f6eb9bf534d4fdcb87b temporal.py
@@ -81,10 +81,10 @@ be0d8585fa57e0ccd6d02577bbc7299629e5bfe20e186747f571fccc98b67d7e v2/seam_types.
3ea23dbce6fcd217b6e81235855479b9b16fed7a21693cf156f9014555c4fe13 v2/session.py 3ea23dbce6fcd217b6e81235855479b9b16fed7a21693cf156f9014555c4fe13 v2/session.py
78c830cdf49034c6e3b2cd3da5914d5cfc5638eb63ec6dbd1bab13c50d6d4caf v3/__init__.py 78c830cdf49034c6e3b2cd3da5914d5cfc5638eb63ec6dbd1bab13c50d6d4caf v3/__init__.py
ea2922567df90436dd6fa21696833a49dd0cbfeee142a5fedc3a7c303824d473 v3/adaptive_continuity.py ea2922567df90436dd6fa21696833a49dd0cbfeee142a5fedc3a7c303824d473 v3/adaptive_continuity.py
b3dc61503d35d9ef0a0dc7fed1c9b2fdc00fcb2072d4e39f64ae0826f9ec7dca v3/assembly_v35.py c03f9f6307b5e3a658fe6d94b01b48bbf5726737e1a8979d6d6f374965f2dc50 v3/assembly_v35.py
7815230e0a9540bc2edeaf818e940b7a1fdd817fd217ba8a4f9cb21aaf6a308e v3/assembly.py 7815230e0a9540bc2edeaf818e940b7a1fdd817fd217ba8a4f9cb21aaf6a308e v3/assembly.py
3f62fdb35e804f42d7cbf0828b4e87e69a2b99246c3f7467fe4fa698f1d803c9 v3/conditioning_bridge_nodes.py 3f62fdb35e804f42d7cbf0828b4e87e69a2b99246c3f7467fe4fa698f1d803c9 v3/conditioning_bridge_nodes.py
e3a6f537a8709c7d05e4df95f1ff40c9129e7845e9ea23097d64953b4770e06d v3/driving_nodes.py fa7a355e52fbcfe1f53b16fa1c37af5d839215b49412adb41c90129ae4e9d9f7 v3/driving_nodes.py
147a1d3fc5e09b323a4c3d13d30e432410389bae52b58330462375b9f307b791 v3/easy_audio_nodes.py 147a1d3fc5e09b323a4c3d13d30e432410389bae52b58330462375b9f307b791 v3/easy_audio_nodes.py
64b3e54b42e5156642f9957ccc374fd190d9fc7145637984e702534fb1bf80b4 v3/easy_image_nodes.py 64b3e54b42e5156642f9957ccc374fd190d9fc7145637984e702534fb1bf80b4 v3/easy_image_nodes.py
ac859cb1a6e23b8ceae300d818c24fe1346b649a0d663017a9b3ac64aa96fca3 v3/easy_nodes.py ac859cb1a6e23b8ceae300d818c24fe1346b649a0d663017a9b3ac64aa96fca3 v3/easy_nodes.py
@@ -97,10 +97,10 @@ cb53e1b104415032ec79b7e7aa1c99848e63da3eedaa251eeba102970e4172b5 v3/hires_fix_n
91a868d2e392b018b01d0396209849c146b0d560ceb109fcdd87dea7f3d786aa v3/memory_action_policy.py 91a868d2e392b018b01d0396209849c146b0d560ceb109fcdd87dea7f3d786aa v3/memory_action_policy.py
85477badd85281cfb77582fdd5b34228a9f6b68376f1b7743af2ca7d4b55aa36 v3/memory_attribution.py 85477badd85281cfb77582fdd5b34228a9f6b68376f1b7743af2ca7d4b55aa36 v3/memory_attribution.py
4a48f74584d8ba3187e1accc31ae5acd3a2ffb9a491bb4f5ca43322db418e770 v3/node_facade.py 4a48f74584d8ba3187e1accc31ae5acd3a2ffb9a491bb4f5ca43322db418e770 v3/node_facade.py
7fe71802319ad7659c90839f960238d0fd054662af31a4d320806c426e755ecb v3/nodes.py f2779a7453a84dec8b993cf044e69ec90da6fe86940a73181234b5000f5a8fa4 v3/nodes.py
a3411c7a295308d21623551cf4aeca5bef511eb5f063f31ac84032e3bdf4e16d v3/packed_row_planner.py a3411c7a295308d21623551cf4aeca5bef511eb5f063f31ac84032e3bdf4e16d v3/packed_row_planner.py
b7c4a35a5e73f528492bc6946b97b687c412f1d3f5f976c19d45c6dc20288c9a v3/pixel_vae_resize.py b7c4a35a5e73f528492bc6946b97b687c412f1d3f5f976c19d45c6dc20288c9a v3/pixel_vae_resize.py
b139843ccfc826937c80f118920bbee2609c37d45ef88291f80f491744fd5e76 v3/plan.py 7e8681bc1ca1633389e1af1ab14100338d23a047f96c59c19151acbbfae5c108 v3/plan.py
0e3dd69d6d2bec6d9491c2fb806c25f0c62629b1eb5f9e4f77ae8d27a5b19cfc v3/planning_types.py 0e3dd69d6d2bec6d9491c2fb806c25f0c62629b1eb5f9e4f77ae8d27a5b19cfc v3/planning_types.py
ec71e3965f445756d54d005c0e27bea520e26dbac44e1ee3d3867bfe2da100b0 v3/ref_encode_cache.py ec71e3965f445756d54d005c0e27bea520e26dbac44e1ee3d3867bfe2da100b0 v3/ref_encode_cache.py
8b9c1cadf76e2c2425d6bf8a5783ced593a75f83a6b3f219a0f870b4c177e058 v3/reference_effective_plan.py 8b9c1cadf76e2c2425d6bf8a5783ced593a75f83a6b3f219a0f870b4c177e058 v3/reference_effective_plan.py
@@ -110,7 +110,7 @@ b64f4d3d9547ac00f8edb2155109476df5b8baea730fb94e8b5bfca6d11564e6 v3/reference_i
1366da221237853eddc9e4b724d4c28917868ea32607dd086361a84df052cfde v3/reference_runtime.py 1366da221237853eddc9e4b724d4c28917868ea32607dd086361a84df052cfde v3/reference_runtime.py
4fb950bbdc7cdb55365fde373ee190af6cdf99f7ce4c54ef73b94f0aca4e1fda v3/reference_storage_contract.py 4fb950bbdc7cdb55365fde373ee190af6cdf99f7ce4c54ef73b94f0aca4e1fda v3/reference_storage_contract.py
057921bf2b05e88c2cbc08c16e2ac483aea905b8a144fbadf4a21d6ae8252427 v3/reference_video.py 057921bf2b05e88c2cbc08c16e2ac483aea905b8a144fbadf4a21d6ae8252427 v3/reference_video.py
b191009d686680c6eec1fdba15b76b7c9f40f11f331faaf7b69ef784e99f4989 v3/refine_context.py 8d6cd0d7f507fd6c2d33667a6ea205e131d47571444abc94d529a98748a52517 v3/refine_context.py
519586b1b3f48c73d329f7a2aaa8a1bfc6fe31ac2ecb64f2b1fdf9c19026001a v3/refine_sampling.py 519586b1b3f48c73d329f7a2aaa8a1bfc6fe31ac2ecb64f2b1fdf9c19026001a v3/refine_sampling.py
43f33f989eb70ca87b916fc33267732a2be3dd4581fc72af17c53f840b53ca49 v3/refine_schedule.py 43f33f989eb70ca87b916fc33267732a2be3dd4581fc72af17c53f840b53ca49 v3/refine_schedule.py
afc149dc347b4c1502ddf32ea4229e633799cd34d2b2c6f98bf8a47910a9ef77 v3/refine_scope.py afc149dc347b4c1502ddf32ea4229e633799cd34d2b2c6f98bf8a47910a9ef77 v3/refine_scope.py
@@ -122,13 +122,13 @@ ae909b43ad17934c85881d3b0bea81e7b45c215eeb4d1f9f6a9ea72faf633ddf v3/reliability
be6610fb9427546d9485a17015cfbdc8786715e6a0967f3e4c1dee35d8768385 v3/runtime_coordinator.py be6610fb9427546d9485a17015cfbdc8786715e6a0967f3e4c1dee35d8768385 v3/runtime_coordinator.py
d0a004aa3030c8adaa5ea788f9acc758fcb4a556f203f9a3f63fa4e4b5772063 v3/sampling_engine.py d0a004aa3030c8adaa5ea788f9acc758fcb4a556f203f9a3f63fa4e4b5772063 v3/sampling_engine.py
452022107cea233aa84d5366ec4943ac5aadbd5fcf786113f1d542a641eac8f7 v3/second_pass_nodes.py 452022107cea233aa84d5366ec4943ac5aadbd5fcf786113f1d542a641eac8f7 v3/second_pass_nodes.py
aa4afff697bd9b8a8505e6cc7d94fcef7aabf9d93ed123ec226fad1960807e77 v3/second_pass.py 97fcb293e0714c3a74b89f08e11c21dce45d25a3010948be1af9dc059d288187 v3/second_pass.py
6e71e944313339f14e14e11cec682593159ca5542aa5d38095371da781d60224 v3/targeted_refine_sampling.py 6e71e944313339f14e14e11cec682593159ca5542aa5d38095371da781d60224 v3/targeted_refine_sampling.py
5a044bdd70c546e4446e0c26af92aae88049ffa2e0e2d01b52033a83a3152d0e v3/targeted_second_pass.py 5a044bdd70c546e4446e0c26af92aae88049ffa2e0e2d01b52033a83a3152d0e v3/targeted_second_pass.py
26b13614ba0a5fd8b192a106c9503b2f4b6fe7f9268a0fcb0f544979a49cab34 v3/video_adapter_nodes.py 26b13614ba0a5fd8b192a106c9503b2f4b6fe7f9268a0fcb0f544979a49cab34 v3/video_adapter_nodes.py
4178dcfa442abdb06ce276e05bd7bcbaa6da28469ba5e57d3da3c663de4e4a9d v3/video_seam_v35.py 4178dcfa442abdb06ce276e05bd7bcbaa6da28469ba5e57d3da3c663de4e4a9d v3/video_seam_v35.py
2f07ace1462644f0bee553827b3571535de4d0c2d9ef847bd9e0ca73a4eed5cb v3/video_seam.py 2f07ace1462644f0bee553827b3571535de4d0c2d9ef847bd9e0ca73a4eed5cb v3/video_seam.py
0d08a173819d63d969694f14e4a1d59ccbec357f37342e00a4db03e1144c2e02 version.py f1c9df5eb9f6c75bd0e505fb84731951fcd9ef1ab5d7f93a0c72f05488fa05fb version.py
b935ad3bb729948bb3890ce967b0ecbef737a5e51ddfc47a0060fc20aa78dcda video_reference_modes.py b935ad3bb729948bb3890ce967b0ecbef737a5e51ddfc47a0060fc20aa78dcda video_reference_modes.py
685df8e0fa23e3663ab31da1e4470cd19bdaa0dc16362cb606e874d9d6153c2c web/decode_cache_helper.js 685df8e0fa23e3663ab31da1e4470cd19bdaa0dc16362cb606e874d9d6153c2c web/decode_cache_helper.js
fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypass_toggle.js fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypass_toggle.js
@@ -136,8 +136,9 @@ fefafd4550511d80bd85ade46a0476f0cf3de13ddb94a908dff6b175f2a0e199 web/easy_bypas
9db0aa94f634578a962d5db49e5c70356040a59df968e3a2e531b327a623081f web/easy_load_image.js 9db0aa94f634578a962d5db49e5c70356040a59df968e3a2e531b327a623081f web/easy_load_image.js
35090816ba7470a7255eb85d003b7e79f8a0065982dacf98a18f41c8ebff655d web/easy_load_video.js 35090816ba7470a7255eb85d003b7e79f8a0065982dacf98a18f41c8ebff655d web/easy_load_video.js
0c68d13a1f8beb7431a32b471f8b0612c42fcc4bf72dd97611a025d158008662 web/easy_v38.js 0c68d13a1f8beb7431a32b471f8b0612c42fcc4bf72dd97611a025d158008662 web/easy_v38.js
c3e2e606af4e5f3e38d64b1338baedea8cadf26060da2a8618e18bf73f220f24 web/project_id.js fbceb922aa3a5ef54886f0ca4c1a07c516ec9fadca04080049ef11f15d584668 web/project_id.js
ae59273f163a116ece418957c557c4fd8c3285e3c60814679b9361d2e6e6fa02 web/reference_audio_ui.js ae59273f163a116ece418957c557c4fd8c3285e3c60814679b9361d2e6e6fa02 web/reference_audio_ui.js
46c9f75c9462bf26300332448f4431cca801f38964ca0a49bc2c50a88a7cc40f web/reference_image_ui.js 46c9f75c9462bf26300332448f4431cca801f38964ca0a49bc2c50a88a7cc40f web/reference_image_ui.js
e9d7753f313ebbe3d13e477f5b4e4767a735b0619b52ba54bcc5218f3eb7eb7c web/reference_images_v39.js e9d7753f313ebbe3d13e477f5b4e4767a735b0619b52ba54bcc5218f3eb7eb7c web/reference_images_v39.js
bf41bf08dd0504eb753fa5781e6c2b8c1285758ad4bb7309975bb9231d5f8e41 web/video_reference_mode_ui.js bf41bf08dd0504eb753fa5781e6c2b8c1285758ad4bb7309975bb9231d5f8e41 web/video_reference_mode_ui.js
7fd122ff08232e114a6ef32171217ea6324e980eb7257d560c466b8177a3d3ec audio_compat.py
+11 -2046
View File
File diff suppressed because it is too large Load Diff
+29
View File
@@ -0,0 +1,29 @@
"""Use Core audio DSP; retain the existing adapter for older Core versions."""
from __future__ import annotations
import importlib
def resample_audio(waveform, source_rate: int, target_rate: int):
if source_rate == target_rate:
return waveform
try:
core_audio = importlib.import_module("comfy.audio")
except ModuleNotFoundError as exc:
if exc.name not in {"comfy", "comfy.audio"}:
raise
core_audio = None
core_resample = getattr(core_audio, "resample", None)
if core_resample is not None:
# Errors from an available Core API are real errors, not API absence.
return core_resample(waveform, source_rate, target_rate)
try:
torchaudio = importlib.import_module("torchaudio")
except ModuleNotFoundError as exc:
if exc.name != "torchaudio":
raise
raise RuntimeError(
"Audio resampling requires comfy.audio.resample or legacy torchaudio"
) from exc
return torchaudio.functional.resample(waveform, source_rate, target_rate)
+2 -43
View File
@@ -4,50 +4,9 @@ Date: 2026-09-14 (Asia/Tokyo)
Phase: R0 only Phase: R0 only
Architecture changes: none Architecture changes: none
## 1. Authoritative baseline ## 1. Historical baseline
R0の正本は、開始時点の次のdirty treeである。HEADや別cloneへ置き換えない。 この文書は2026-09-14時点のR0 CPU characterizationです。現行3.9.1の受入結果ではありません。当時の実装・Workflow・既存変更を保持した状態で調査し、アーキテクチャ変更やGPU生成は行っていません。
```text
authoritative_runtime_path:
D:\StabilityMatrix\Data\Packages\ComfyUI_W\custom_nodes\ComfyUI-H3-Continuum
branch: main
HEAD: c421913bf1311050ac10a6d782ce40f9e72be852
Python: 3.13.12
Torch: 2.14.0+cu130
ComfyUI: 0.35.1
GPU initialized by R0: NO
user backend 8188 touched: NO
```
開始時点には既存のtracked変更と、次のuntrackedファイルがあった。すべてユーザー作業として保持した。
| Untracked file | SHA-256 |
|---|---|
| `examples/workflows/MiniMax_H3_Continuum_V38.json` | `9772DD1A82C20FD0D6F40AE7CD2B2F11DFB81641B86515ADB17C6555ADA7AFF3` |
| `examples/workflows/MiniMax_H3_Continuum_V38.zip` | `9B684C37EBFE116FCB5C68032B82E5D49279325548C48494F23B3309BB211924` |
| `tests/test_audit_boundary_repairs.py` | `FE23752312D62DDC8C23A3C598ED2F92F2D46E2E4460F06AEA91921659092451` |
Workflow ZIP内の唯一のentryは`MiniMax_H3_Continuum_V38.json`で、展開JSONのSHA-256は上表のJSONと一致した。
Baseline diff artifact:
```text
D:\Codex\_test_results\ComfyUI-H3-Continuum\r0-runtime-redesign-20260914\BASELINE_WORKTREE.diff
```
Snapshot:
```text
verified:
D:\Codex\_snapshots\ComfyUI-H3-Continuum\pre-r0-runtime-redesign-verified-excluding-pytest-cache-20260914_021621
files: 892
partial attempt (Copy-Item was denied by .pytest_cache ACL):
D:\Codex\_snapshots\ComfyUI-H3-Continuum\pre-r0-runtime-redesign-20260914_021557
```
Verified snapshotは`.git`とアクセス不能だった`.pytest_cache`だけを除外し、source/destination file countを一致確認した。
## 2. Baseline validation ## 2. Baseline validation
+3 -2
View File
@@ -11,9 +11,10 @@ ComfyUIを終了後、展開フォルダーで実行します。
または: または:
```powershell ```powershell
D:\StabilityMatrix\Data\Packages\ComfyUI_W\venv\Scripts\python.exe ` $ComfyRoot = "path/to/ComfyUI" # 使用中のComfyUIの場所へ置き換える
& "$ComfyRoot/venv/Scripts/python.exe" `
.\tools\verify_runtime.py ` .\tools\verify_runtime.py `
--comfy-root D:\StabilityMatrix\Data\Packages\ComfyUI_W --comfy-root $ComfyRoot
``` ```
確認項目: 確認項目:
+17 -3
View File
@@ -9,6 +9,8 @@ from typing import Any
import torch import torch
from .audio_compat import resample_audio
DRIVING_AUDIO_CONTRACT_VERSION = 1 DRIVING_AUDIO_CONTRACT_VERSION = 1
DRIVING_AUDIO_PREPROCESS_VERSION = 1 DRIVING_AUDIO_PREPROCESS_VERSION = 1
@@ -49,6 +51,20 @@ class DrivingAudioAssets:
audio_latent: torch.Tensor audio_latent: torch.Tensor
def slice_review_source_audio(audio: dict[str, Any] | None, assembly_plan):
"""Cut the selected full-source PCM once; do not alter the plan or source."""
from .v3.plan import validate_review_audio_projection
from .media import validate_audio
projection = validate_review_audio_projection(assembly_plan)
if audio is None or projection is None:
return audio
waveform, sample_rate = validate_audio(audio)
samples = int(waveform.shape[-1])
start = min(samples, round(projection["source_start_frame"] * sample_rate / projection["fps"]))
stop = min(samples, round(projection["source_stop_frame"] * sample_rate / projection["fps"]))
return {**audio, "waveform": waveform[..., start:stop], "sample_rate": sample_rate}
def prepare_driving_audio_source( def prepare_driving_audio_source(
audio: dict[str, Any] | None, audio: dict[str, Any] | None,
audio_vae: Any, audio_vae: Any,
@@ -77,9 +93,7 @@ def prepare_driving_audio_source(
resolved_rate = int(getattr(audio_vae, "audio_sample_rate", 32000)) resolved_rate = int(getattr(audio_vae, "audio_sample_rate", 32000))
resampled = effective_waveform resampled = effective_waveform
if sample_rate != resolved_rate: if sample_rate != resolved_rate:
import torchaudio resampled = resample_audio(
resampled = torchaudio.functional.resample(
effective_waveform, effective_waveform,
sample_rate, sample_rate,
resolved_rate, resolved_rate,
+3 -2
View File
@@ -7,7 +7,7 @@ import json
from typing import Any from typing import Any
GRAPH_CONTRACT_VERSION = 3 GRAPH_CONTRACT_VERSION = 4
CHECKPOINT_REF2VA = "ref2va" CHECKPOINT_REF2VA = "ref2va"
CHECKPOINT_FL2VA = "fl2va" CHECKPOINT_FL2VA = "fl2va"
@@ -93,7 +93,7 @@ def build_upstream_graph_contract(
require_audio_vae: bool = False, require_audio_vae: bool = False,
reference_audio_route: str | None = None, reference_audio_route: str | None = None,
) -> tuple[dict[str, Any], bool, list[str]]: ) -> tuple[dict[str, Any], bool, list[str]]:
"""Fingerprint MODEL/CLIP/VAE routes feeding the Continuum sampler.""" """Fingerprint MODEL/CLIP/VAE/SAMPLER routes feeding Continuum."""
if not isinstance(prompt, dict): if not isinstance(prompt, dict):
return {"version": GRAPH_CONTRACT_VERSION}, False, ["ComfyUI PROMPT graph is unavailable"] return {"version": GRAPH_CONTRACT_VERSION}, False, ["ComfyUI PROMPT graph is unavailable"]
node_id = str(unique_id) node_id = str(unique_id)
@@ -107,6 +107,7 @@ def build_upstream_graph_contract(
specs = [ specs = [
("model", "model", True), ("model", "model", True),
("clip", "clip", True), ("clip", "clip", True),
("sampler", "sampler", True),
("video_vae", "video_vae", bool(require_video_vae)), ("video_vae", "video_vae", bool(require_video_vae)),
] ]
if require_reference_audio_vae: if require_reference_audio_vae:
+1 -1
View File
@@ -1,4 +1,4 @@
[project] [project]
name = H3 Continuum name = H3 Continuum
version = 3.9.0 version = 3.9.1
description = Native MiniMax H3 masked AV continuation, Second Pass, and disk-backed assembly. description = Native MiniMax H3 masked AV continuation, Second Pass, and disk-backed assembly.
+1 -1
View File
@@ -1,6 +1,6 @@
[project] [project]
name = "comfyui-h3-continuum" name = "comfyui-h3-continuum"
version = "3.9.0" version = "3.9.1"
description = "Integrated N-chunk continuation for ComfyUI native MiniMax H3 video and audio" description = "Integrated N-chunk continuation for ComfyUI native MiniMax H3 video and audio"
readme = "README.md" readme = "README.md"
requires-python = ">=3.10" requires-python = ">=3.10"
+3 -3
View File
@@ -10,6 +10,8 @@ from typing import Any
import torch import torch
from .audio_compat import resample_audio
REFERENCE_AUDIO_CONTRACT_VERSION = 1 REFERENCE_AUDIO_CONTRACT_VERSION = 1
REFERENCE_AUDIO_PREPROCESS_VERSION = 1 REFERENCE_AUDIO_PREPROCESS_VERSION = 1
@@ -152,9 +154,7 @@ def _prepare_reference_audio_source(
resampled = waveform resampled = waveform
if source_sample_rate != resolved_rate: if source_sample_rate != resolved_rate:
try: try:
import torchaudio resampled = resample_audio(
resampled = torchaudio.functional.resample(
waveform, source_sample_rate, resolved_rate waveform, source_sample_rate, resolved_rate
).contiguous() ).contiguous()
except Exception as exc: except Exception as exc:
+141 -30
View File
@@ -5,6 +5,8 @@ from __future__ import annotations
import contextvars import contextvars
import copy import copy
import hashlib import hashlib
import functools
import inspect
import json import json
import marshal import marshal
import os import os
@@ -72,7 +74,10 @@ from .conditioning import (
) )
SAMPLING_CONTRACT_VERSION = 5 SAMPLING_CONTRACT_VERSION = 6
SAMPLING_READABLE_CONTRACT_VERSIONS = {5, 6}
RUNTIME_SIGNATURE_VERSION = 1
_OBSERVATION_MAX_ITEMS = 4096
REVIEW_CONTROL_VERSION = 1 REVIEW_CONTROL_VERSION = 1
RUN_STORAGE_OFF = "Off" RUN_STORAGE_OFF = "Off"
RUN_STORAGE_AUTO = "Save + Auto Resume" RUN_STORAGE_AUTO = "Save + Auto Resume"
@@ -450,6 +455,13 @@ def _tensor_exact(tensor: torch.Tensor) -> dict[str, Any]:
def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]: def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
try:
return _observe_value(value, depth=depth)
except Exception as exc:
return {"type": _qualified(value), "observation_error": type(exc).__name__}, False
def _observe_value(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
if depth > 10: if depth > 10:
return {"type": _qualified(value), "truncated": True}, False return {"type": _qualified(value), "truncated": True}, False
if value is None or isinstance(value, (bool, int, float, str)): if value is None or isinstance(value, (bool, int, float, str)):
@@ -462,6 +474,8 @@ def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
except Exception: except Exception:
return {"tensor_type": _qualified(value)}, False return {"tensor_type": _qualified(value)}, False
if isinstance(value, dict): if isinstance(value, dict):
if len(value) > _OBSERVATION_MAX_ITEMS:
return {"type": _qualified(value), "items": len(value), "truncated": True}, False
result: dict[str, Any] = {} result: dict[str, Any] = {}
safe = True safe = True
for key in sorted(value, key=lambda item: str(item)): for key in sorted(value, key=lambda item: str(item)):
@@ -470,6 +484,8 @@ def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
safe = safe and item_safe safe = safe and item_safe
return result, safe return result, safe
if isinstance(value, (list, tuple)): if isinstance(value, (list, tuple)):
if len(value) > _OBSERVATION_MAX_ITEMS:
return {"type": _qualified(value), "items": len(value), "truncated": True}, False
result = [] result = []
safe = True safe = True
for item in value: for item in value:
@@ -478,13 +494,24 @@ def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
safe = safe and item_safe safe = safe and item_safe
return result, safe return result, safe
if isinstance(value, (set, frozenset)): if isinstance(value, (set, frozenset)):
if len(value) > _OBSERVATION_MAX_ITEMS:
return {"type": _qualified(value), "items": len(value), "truncated": True}, False
# Set repr order varies with PYTHONHASHSEED; keep every item and its type. # Set repr order varies with PYTHONHASHSEED; keep every item and its type.
items = [_observe(item, depth=depth + 1) for item in value] items = [_observe(item, depth=depth + 1) for item in value]
return { return {
"type": _qualified(value), "type": _qualified(value),
"items": sorted((item for item, _ in items), key=_canonical), "items": sorted((item for item, _ in items), key=_canonical),
}, all(safe for _, safe in items) }, all(safe for _, safe in items)
if isinstance(value, torch.nn.Module):
# Never traverse every parameter through a wrapper's bound owner.
return {"type": _qualified(value), "module_state_unobserved": True}, False
if callable(value): if callable(value):
if isinstance(value, functools.partial):
observed, safe = _observe({
"function": value.func, "args": value.args,
"keywords": value.keywords, "attributes": vars(value),
}, depth=depth + 1)
return {"partial": observed}, safe
code = getattr(value, "__code__", None) code = getattr(value, "__code__", None)
descriptor: dict[str, Any] = {"callable": _qualified(value)} descriptor: dict[str, Any] = {"callable": _qualified(value)}
safe = True safe = True
@@ -498,9 +525,23 @@ def _observe(value: Any, *, depth: int = 0) -> tuple[Any, bool]:
closure, closure_safe = _observe(closure_values, depth=depth + 1) closure, closure_safe = _observe(closure_values, depth=depth + 1)
descriptor.update(defaults=defaults, kwdefaults=kwdefaults, closure=closure) descriptor.update(defaults=defaults, kwdefaults=kwdefaults, closure=closure)
safe = defaults_safe and kw_safe and closure_safe safe = defaults_safe and kw_safe and closure_safe
if inspect.ismethod(value):
owner, owner_safe = _observe(value.__self__, depth=depth + 1)
descriptor["bound_self"] = owner
safe = safe and owner_safe
elif hasattr(value, "__dict__"): elif hasattr(value, "__dict__"):
attributes, safe = _observe(vars(value), depth=depth + 1) attributes, attributes_safe = _observe(vars(value), depth=depth + 1)
descriptor["attributes"] = attributes descriptor["attributes"] = attributes
# Callable instances need both state and executable implementation.
# An empty __dict__ is not evidence for an opaque extension wrapper.
implementation = getattr(type(value), "__call__", None)
if inspect.isfunction(implementation):
call, call_safe = _observe(implementation, depth=depth + 1)
descriptor["implementation"] = call
slots = any(getattr(cls, "__slots__", ()) for cls in type(value).__mro__)
safe = attributes_safe and call_safe and bool(vars(value)) and not slots
else:
safe = False
else: else:
safe = False safe = False
return descriptor, safe return descriptor, safe
@@ -517,13 +558,14 @@ def _model_signature(model: Any, legacy_fingerprint: str) -> tuple[dict[str, Any
wrappers, wrappers_safe = _observe(getattr(model, "wrappers", {}) or {}) wrappers, wrappers_safe = _observe(getattr(model, "wrappers", {}) or {})
patches, patches_safe = _observe(getattr(model, "patches", {}) or {}) patches, patches_safe = _observe(getattr(model, "patches", {}) or {})
options = getattr(model, "model_options", {}) or {} options = getattr(model, "model_options", {}) or {}
transformer = options.get("transformer_options", {}) or {} if not isinstance(options, dict):
transformer_values = { return {"model_patcher": _qualified(model), "error": "model_options is not a mapping"}, False
str(key): transformer[key] observed_options = dict(options)
for key in transformer transformer = dict(observed_options.get("transformer_options", {}) or {})
if key not in {"h3_continuum_join_context"} transformer.pop("h3_continuum_join_context", None)
} if "transformer_options" in observed_options:
observed_transformer, transformer_safe = _observe(transformer_values) observed_options["transformer_options"] = transformer
observed_options, options_safe = _observe(observed_options)
weights: list[dict[str, Any]] = [] weights: list[dict[str, Any]] = []
weights_safe = True weights_safe = True
try: try:
@@ -541,31 +583,56 @@ def _model_signature(model: Any, legacy_fingerprint: str) -> tuple[dict[str, Any
"dtype": str(getattr(model, "model_dtype", lambda: "unknown")()), "dtype": str(getattr(model, "model_dtype", lambda: "unknown")()),
"model_size": int(getattr(model, "model_size", lambda: 0)() or 0), "model_size": int(getattr(model, "model_size", lambda: 0)() or 0),
"wrappers": wrappers, "wrappers": wrappers,
"transformer_options": observed_transformer, "model_options": observed_options,
"patches": patches, "patches": patches,
"weight_probe": weights, "weight_probe": weights,
"runtime_observation": { "runtime_observation": {
"wrappers_complete": bool(wrappers_safe), "wrappers_complete": bool(wrappers_safe),
"patches_complete": bool(patches_safe), "patches_complete": bool(patches_safe),
"transformer_options_complete": bool(transformer_safe), "model_options_complete": bool(options_safe),
}, },
} }
return descriptor, weights_safe and bool(weights) return descriptor, weights_safe and bool(weights) and wrappers_safe and patches_safe and options_safe
def _sampler_signature(sampler: Any) -> tuple[dict[str, Any], bool]: def _sampler_signature(sampler: Any) -> tuple[dict[str, Any], bool]:
function = getattr(sampler, "sampler_function", None) function = getattr(sampler, "sampler_function", None)
observed, safe = _observe({ observed, safe = _observe({
"function": function,
"extra_options": getattr(sampler, "extra_options", {}) or {}, "extra_options": getattr(sampler, "extra_options", {}) or {},
"inpaint_options": getattr(sampler, "inpaint_options", {}) or {}, "inpaint_options": getattr(sampler, "inpaint_options", {}) or {},
}) })
return { return {
"type": _qualified(sampler), "type": _qualified(sampler),
"function": _qualified(function) if function is not None else "unknown",
"options": observed, "options": observed,
}, function is not None and safe }, function is not None and safe
def _signature_observation(observer, *args, **kwargs):
try:
return observer(*args, **kwargs)
except Exception as exc:
return {"observation_error": type(exc).__name__}, False
def _reusable_sampling_contract(contract: dict[str, Any], *, resume_safe: bool = True) -> bool:
global_contract = contract.get("global") or {}
return bool(
resume_safe
and global_contract.get("sampling_contract_version") == SAMPLING_CONTRACT_VERSION
and global_contract.get("runtime_signature_version") == RUNTIME_SIGNATURE_VERSION
and contract.get("reuse_policy") != "disabled_unobservable_contract"
)
def _reusable_manifest(manifest: dict[str, Any]) -> bool:
return bool(
manifest.get("sampling_contract_version") == SAMPLING_CONTRACT_VERSION
and manifest.get("resume_safe") is True
and _reusable_sampling_contract(manifest.get("contract") or {})
)
def _module_parameter_dtype(module: Any, name: str, parameter: torch.Tensor) -> torch.dtype: def _module_parameter_dtype(module: Any, name: str, parameter: torch.Tensor) -> torch.dtype:
"""Observe ordinary weights at Core's declared load dtype without casting them.""" """Observe ordinary weights at Core's declared load dtype without casting them."""
# Quantized/custom Tensor representations retain their original fingerprint. # Quantized/custom Tensor representations retain their original fingerprint.
@@ -889,7 +956,7 @@ def build_sampling_contract(
if reference_storage_plan is not None and reference_contract is not None: if reference_storage_plan is not None and reference_contract is not None:
raise RunStorageError("routed Reference plan cannot use legacy Reference contract") raise RunStorageError("routed Reference plan cannot use legacy Reference contract")
last_frame_hash = _canonical_optional_hash(last_frame_hash) last_frame_hash = _canonical_optional_hash(last_frame_hash)
model_value, model_safe = _model_signature(model, model_fingerprint_value) model_value, model_safe = _signature_observation(_model_signature, model, model_fingerprint_value)
clip_value, clip_safe = _clip_signature(clip) clip_value, clip_safe = _clip_signature(clip)
has_first = str(first_frame_hash).lower() not in {"", "none", "null"} has_first = str(first_frame_hash).lower() not in {"", "none", "null"}
has_last = str(last_frame_hash).lower() not in {"", "none", "null"} has_last = str(last_frame_hash).lower() not in {"", "none", "null"}
@@ -920,7 +987,7 @@ def build_sampling_contract(
) )
else: else:
video_vae_value, video_vae_safe = None, True video_vae_value, video_vae_safe = None, True
sampler_value, sampler_safe = _sampler_signature(sampler) sampler_value, sampler_safe = _signature_observation(_sampler_signature, sampler)
if reference_audio_contract is not None: if reference_audio_contract is not None:
audio_vae_value, audio_vae_safe = _audio_vae_signature(reference_audio_vae) audio_vae_value, audio_vae_safe = _audio_vae_signature(reference_audio_vae)
else: else:
@@ -964,6 +1031,7 @@ def build_sampling_contract(
} }
global_contract = { global_contract = {
"sampling_contract_version": SAMPLING_CONTRACT_VERSION, "sampling_contract_version": SAMPLING_CONTRACT_VERSION,
"runtime_signature_version": RUNTIME_SIGNATURE_VERSION,
"conditioning_mode": conditioning_mode, "conditioning_mode": conditioning_mode,
"model": model_value, "model": model_value,
"clip": clip_value, "clip": clip_value,
@@ -1584,6 +1652,8 @@ class RunStorageController:
) )
} }
revision.update( revision.update(
sampling_contract_version=value.get("sampling_contract_version"),
generation_reusable=_reusable_manifest(value),
nonce_mode=lifecycle.get("mode"), nonce_mode=lifecycle.get("mode"),
effective_reroll_nonce=lifecycle.get("effective_nonce"), effective_reroll_nonce=lifecycle.get("effective_nonce"),
reroll_from_chunk=(value.get("contract") or {}).get("reroll_from_chunk"), reroll_from_chunk=(value.get("contract") or {}).get("reroll_from_chunk"),
@@ -1616,7 +1686,10 @@ class RunStorageController:
[catalog[revision_id] for revision_id in canonical_chain] [catalog[revision_id] for revision_id in canonical_chain]
) if canonical_chain else {}, ) if canonical_chain else {},
"group_revisions": sorted( "group_revisions": sorted(
catalog.values(), [dict(item,
sampling_contract_version=(manifests.get(str(item.get("storage_revision_id"))) or {}).get("sampling_contract_version"),
generation_reusable=_reusable_manifest(manifests.get(str(item.get("storage_revision_id"))) or {}))
for item in catalog.values()],
key=lambda item: ( key=lambda item: (
int(item["group"]["physical_group"]), int(item["group"]["physical_group"]),
str(item["revision_order"]), str(item["revision_order"]),
@@ -1663,8 +1736,10 @@ class RunStorageController:
hashes: list[str], hashes: list[str],
*, *,
current_contract: dict[str, Any] | None = None, current_contract: dict[str, Any] | None = None,
prompts: list[str] | None = None,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: ) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
stored_contract = manifest.get("contract") or {} stored_contract = manifest.get("contract") or {}
prompts = self.prompts if prompts is None else prompts
active_contract = self.contract if current_contract is None else current_contract active_contract = self.contract if current_contract is None else current_contract
stored = list(stored_contract.get("chunk_contract_hashes") or []) stored = list(stored_contract.get("chunk_contract_hashes") or [])
stored_prompt_hashes = list(stored_contract.get("prompt_hashes") or []) stored_prompt_hashes = list(stored_contract.get("prompt_hashes") or [])
@@ -1687,7 +1762,7 @@ class RunStorageController:
if position >= len(records) or int(records[position].get("sequence_index", -1)) != position: if position >= len(records) or int(records[position].get("sequence_index", -1)) != position:
break break
try: try:
entry = self._load_entry(records[position], self.prompts[position]) entry = self._load_entry(records[position], prompts[position])
except Exception as exc: except Exception as exc:
self.notes.append(f"stored chunk {position + 1} rejected: {exc}") self.notes.append(f"stored chunk {position + 1} rejected: {exc}")
break break
@@ -1715,7 +1790,9 @@ class RunStorageController:
*, *,
current_contract: dict[str, Any], current_contract: dict[str, Any],
enforce_sampling_contract: bool = False, enforce_sampling_contract: bool = False,
prompts: list[str] | None = None,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: ) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
prompts = self.prompts if prompts is None else prompts
manifests = self._read_manifests() manifests = self._read_manifests()
storage_revision_id = str(manifest.get("revision_id", "")) storage_revision_id = str(manifest.get("revision_id", ""))
if storage_revision_id not in manifests: if storage_revision_id not in manifests:
@@ -1734,7 +1811,7 @@ class RunStorageController:
expected_prompts = list(current_contract.get("prompt_hashes") or []) expected_prompts = list(current_contract.get("prompt_hashes") or [])
entries: list[dict[str, Any]] = [] entries: list[dict[str, Any]] = []
for position, record in enumerate(records): for position, record in enumerate(records):
if position >= len(self.prompts) or position >= len(expected_prompts): if position >= len(prompts) or position >= len(expected_prompts):
raise RunStorageError("provenance prefix exceeds the current prompt plan") raise RunStorageError("provenance prefix exceeds the current prompt plan")
if enforce_sampling_contract: if enforce_sampling_contract:
boundary = int(current_contract.get("reroll_from_chunk", 0)) boundary = int(current_contract.get("reroll_from_chunk", 0))
@@ -1753,7 +1830,7 @@ class RunStorageController:
or produced_hashes[position] != requested_hashes[position] or produced_hashes[position] != requested_hashes[position]
): ):
break break
entry = self._load_entry(record, self.prompts[position]) entry = self._load_entry(record, prompts[position])
if str(entry.get("prompt_hash", "")) != str(expected_prompts[position]): if str(entry.get("prompt_hash", "")) != str(expected_prompts[position]):
raise RunStorageError("provenance prefix prompt lineage is incompatible") raise RunStorageError("provenance prefix prompt lineage is incompatible")
entries.append(entry) entries.append(entry)
@@ -1761,7 +1838,7 @@ class RunStorageController:
def _compatible_plan_prefix( def _compatible_plan_prefix(
self, manifest: dict[str, Any], current_contract: dict[str, Any], self, manifest: dict[str, Any], current_contract: dict[str, Any],
*, stop_before: int = 0, *, stop_before: int = 0, resume_safe: bool = True,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: ) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
"""Certify a prefix for import, not for cross-lineage Take selection. """Certify a prefix for import, not for cross-lineage Take selection.
@@ -1769,6 +1846,8 @@ class RunStorageController:
retain its topology and actual producing per-chunk contract. New plans retain its topology and actual producing per-chunk contract. New plans
get their own raw copies below; no foreign lineage is spliced into a DAG. get their own raw copies below; no foreign lineage is spliced into a DAG.
""" """
if not _reusable_sampling_contract(current_contract, resume_safe=resume_safe) or not _reusable_manifest(manifest):
return [], []
manifests = self._read_manifests() manifests = self._read_manifests()
source_contract = manifest.get("contract") or {} source_contract = manifest.get("contract") or {}
_manifest_sampling_identity(manifest) _manifest_sampling_identity(manifest)
@@ -1794,7 +1873,7 @@ class RunStorageController:
compatible = True compatible = True
for position, record in zip(range(expected.start - 1, expected.end), group_records): for position, record in zip(range(expected.start - 1, expected.end), group_records):
producer = manifests.get(str(record.get("storage_revision_id", ""))) producer = manifests.get(str(record.get("storage_revision_id", "")))
if producer is None: if producer is None or not _reusable_manifest(producer):
compatible = False compatible = False
break break
if "reference_group_contracts" in (producer.get("contract") or {}): if "reference_group_contracts" in (producer.get("contract") or {}):
@@ -1834,10 +1913,10 @@ class RunStorageController:
return entries, accepted return entries, accepted
def _find_plan_import( def _find_plan_import(
self, contract: dict[str, Any], *, stop_before: int = 0, self, contract: dict[str, Any], *, stop_before: int = 0, resume_safe: bool = True,
) -> dict[str, Any] | None: ) -> dict[str, Any] | None:
"""Find a verified plan prefix, stopping before explicit regeneration.""" """Find a verified plan prefix, stopping before explicit regeneration."""
if stop_before == 1: if stop_before == 1 or not _reusable_sampling_contract(contract, resume_safe=resume_safe):
return None return None
manifests = self._read_manifests() manifests = self._read_manifests()
# A matching plan already owns its history. Do not reload an older plan # A matching plan already owns its history. Do not reload an older plan
@@ -1862,7 +1941,8 @@ class RunStorageController:
), reverse=True) ), reverse=True)
for manifest in ordered: for manifest in ordered:
source = manifest.get("contract") or {} source = manifest.get("contract") or {}
if (manifest.get("status") not in {"complete", "review_ready", "interrupted"} if (not _reusable_manifest(manifest)
or manifest.get("status") not in {"complete", "review_ready", "interrupted"}
or source.get("nonce_lineage_sha256") == contract.get("nonce_lineage_sha256") or source.get("nonce_lineage_sha256") == contract.get("nonce_lineage_sha256")
or _hash(source.get("global")) != _hash(contract.get("global"))): or _hash(source.get("global")) != _hash(contract.get("global"))):
continue continue
@@ -1902,6 +1982,8 @@ class RunStorageController:
def _adopt_plan_prefix(self, imported: dict[str, Any]) -> list[dict[str, Any]]: def _adopt_plan_prefix(self, imported: dict[str, Any]) -> list[dict[str, Any]]:
"""Copy verified tensors into the new plan; keep every old revision intact.""" """Copy verified tensors into the new plan; keep every old revision intact."""
if not _reusable_sampling_contract(self.contract, resume_safe=self.resume_safe) or not _reusable_manifest(imported.get("manifest") or {}):
raise RunStorageError("plan prefix has no reusable Sampling v6 identity")
self.manifest["prefix_import"] = { self.manifest["prefix_import"] = {
"version": 1, "source_revision_id": imported["revision_id"], "version": 1, "source_revision_id": imported["revision_id"],
"source_contract_sha256": imported["manifest"]["contract_sha256"], "source_contract_sha256": imported["manifest"]["contract_sha256"],
@@ -1961,8 +2043,10 @@ class RunStorageController:
def _validated_take_selection( def _validated_take_selection(
self, self,
contract: dict[str, Any], contract: dict[str, Any],
*, source_lineage: str | None = None, *, source_lineage: str | None = None, resume_safe: bool = True,
) -> dict[str, Any]: ) -> dict[str, Any]:
if not _reusable_sampling_contract(contract, resume_safe=resume_safe):
raise RunStorageError("selected Take cannot be reused: current sampling identity is unobservable or unsupported")
lineage = str(source_lineage or contract.get("nonce_lineage_sha256", "")) lineage = str(source_lineage or contract.get("nonce_lineage_sha256", ""))
manifests, catalog, _ = self._provenance_catalog( manifests, catalog, _ = self._provenance_catalog(
lineage_sha256=lineage, lineage_sha256=lineage,
@@ -2003,6 +2087,8 @@ class RunStorageController:
source_manifest = manifests.get(source_id) source_manifest = manifests.get(source_id)
if source_manifest is None: if source_manifest is None:
raise RunStorageError("selected Take source manifest is unavailable") raise RunStorageError("selected Take source manifest is unavailable")
if not _reusable_manifest(source_manifest):
raise RunStorageError("selected Take is read-only: Sampling v6 with an observable identity is required")
source_records = list(source_manifest.get("chunks") or []) source_records = list(source_manifest.get("chunks") or [])
for position in range(expected_group.start - 1, expected_group.end): for position in range(expected_group.start - 1, expected_group.end):
if position >= len(source_records): if position >= len(source_records):
@@ -2057,9 +2143,11 @@ class RunStorageController:
self, self,
manifest: dict[str, Any], manifest: dict[str, Any],
*, *,
current_contract: dict[str, Any], current_contract: dict[str, Any] | None = None,
for_reuse: bool = True,
resume_safe: bool = True,
) -> dict[str, Any]: ) -> dict[str, Any]:
"""Validate one compatible persisted state without applying policy.""" """Validate stored integrity first; optionally certify current reuse."""
schema = int(manifest.get("run_storage_schema_version", -1)) schema = int(manifest.get("run_storage_schema_version", -1))
if schema not in RUN_STORAGE_READABLE_SCHEMA_VERSIONS: if schema not in RUN_STORAGE_READABLE_SCHEMA_VERSIONS:
@@ -2067,13 +2155,19 @@ class RunStorageController:
stored_contract = manifest.get("contract") stored_contract = manifest.get("contract")
if not isinstance(stored_contract, dict): if not isinstance(stored_contract, dict):
raise RunStorageError("review head sampling contract is missing") raise RunStorageError("review head sampling contract is missing")
if str(stored_contract.get("nonce_lineage_sha256", "")) != str( if for_reuse and (
current_contract is None
or not _reusable_sampling_contract(current_contract, resume_safe=resume_safe)
or not _reusable_manifest(manifest)
):
raise RunStorageError("review head is read-only or its sampling identity is unobservable")
if for_reuse and str(stored_contract.get("nonce_lineage_sha256", "")) != str(
current_contract.get("nonce_lineage_sha256", "") current_contract.get("nonce_lineage_sha256", "")
): ):
raise RunStorageError("review head belongs to a different sampling lineage") raise RunStorageError("review head belongs to a different sampling lineage")
_, contract_sha256 = _manifest_sampling_identity(manifest) _, contract_sha256 = _manifest_sampling_identity(manifest)
revision_id = str(manifest.get("revision_id", "")) revision_id = str(manifest.get("revision_id", ""))
if int(manifest.get("sampling_contract_version", -1)) != SAMPLING_CONTRACT_VERSION: if int(manifest.get("sampling_contract_version", -1)) not in SAMPLING_READABLE_CONTRACT_VERSIONS:
raise RunStorageError("review head sampling contract version is incompatible") raise RunStorageError("review head sampling contract version is incompatible")
status = str(manifest.get("status", "")) status = str(manifest.get("status", ""))
@@ -2091,16 +2185,19 @@ class RunStorageController:
if len(hashes) != chunks: if len(hashes) != chunks:
raise RunStorageError("review head chunk contract is incomplete") raise RunStorageError("review head chunk contract is incomplete")
records = list(manifest.get("chunks") or []) records = list(manifest.get("chunks") or [])
stored_prompts = None if for_reuse else [""] * chunks
if schema == RUN_STORAGE_SCHEMA_VERSION and manifest.get("branch_provenance"): if schema == RUN_STORAGE_SCHEMA_VERSION and manifest.get("branch_provenance"):
entries, accepted = self._valid_provenance_prefix( entries, accepted = self._valid_provenance_prefix(
manifest, manifest,
current_contract=stored_contract, current_contract=stored_contract,
prompts=stored_prompts,
) )
else: else:
entries, accepted = self._valid_prefix( entries, accepted = self._valid_prefix(
manifest, manifest,
hashes, hashes,
current_contract=stored_contract, current_contract=stored_contract,
prompts=stored_prompts,
) )
if len(accepted) != len(records): if len(accepted) != len(records):
raise RunStorageError("review head stored prefix failed validation") raise RunStorageError("review head stored prefix failed validation")
@@ -2190,9 +2287,13 @@ class RunStorageController:
contract: dict[str, Any], contract: dict[str, Any],
*, *,
smart_regenerate_only: bool = False, smart_regenerate_only: bool = False,
resume_safe: bool = True,
) -> dict[str, Any] | None: ) -> dict[str, Any] | None:
"""Return the latest validated compatible disk head deterministically.""" """Return the latest validated compatible disk head deterministically."""
if not _reusable_sampling_contract(contract, resume_safe=resume_safe):
self.notes.append("auto-resume disabled: current sampling identity is unobservable or unsupported")
return None
if not self.revisions_root.exists(): if not self.revisions_root.exists():
return None return None
compatible_lineage = str(contract.get("nonce_lineage_sha256", "")) compatible_lineage = str(contract.get("nonce_lineage_sha256", ""))
@@ -2215,6 +2316,9 @@ class RunStorageController:
), reverse=True) ), reverse=True)
for revision_id, manifest in ordered: for revision_id, manifest in ordered:
stored_contract = manifest.get("contract") or {} stored_contract = manifest.get("contract") or {}
if not _reusable_manifest(manifest):
self.notes.append(f"review head {revision_id} is read-only: observable Sampling v6 required")
continue
if str(stored_contract.get("nonce_lineage_sha256", "")) != compatible_lineage: if str(stored_contract.get("nonce_lineage_sha256", "")) != compatible_lineage:
mismatched_contracts.append((revision_id, stored_contract)) mismatched_contracts.append((revision_id, stored_contract))
continue continue
@@ -2343,7 +2447,7 @@ class RunStorageController:
if selected_take is not None: if selected_take is not None:
source_lineage = str(selected_take.get("lineage_sha256", "")) source_lineage = str(selected_take.get("lineage_sha256", ""))
candidate = self._validated_take_selection( candidate = self._validated_take_selection(
contract, source_lineage=source_lineage, contract, source_lineage=source_lineage, resume_safe=resume_safe,
) )
if source_lineage != str(contract.get("nonce_lineage_sha256", "")): if source_lineage != str(contract.get("nonce_lineage_sha256", "")):
source_manifest = candidate.get("manifest") source_manifest = candidate.get("manifest")
@@ -2389,11 +2493,13 @@ class RunStorageController:
candidate = self.find_latest_review_head( candidate = self.find_latest_review_head(
contract, contract,
smart_regenerate_only=smart_only, smart_regenerate_only=smart_only,
resume_safe=resume_safe,
) )
if candidate is None and not smart_only and resume_safe: if candidate is None and not smart_only and resume_safe:
self._plan_import = self._find_plan_import( self._plan_import = self._find_plan_import(
contract, contract,
stop_before=self.review_manual_regenerate_from, stop_before=self.review_manual_regenerate_from,
resume_safe=resume_safe,
) )
candidate = self._plan_import candidate = self._plan_import
if candidate is None and smart_only: if candidate is None and smart_only:
@@ -2715,6 +2821,8 @@ class RunStorageController:
if review_queue: if review_queue:
if self.validated_prefix is None: if self.validated_prefix is None:
raise RunStorageError("Review Decision has no fixed ValidatedPrefix") raise RunStorageError("Review Decision has no fixed ValidatedPrefix")
if self.validated_prefix.entries and not _reusable_sampling_contract(contract, resume_safe=safe):
raise RunStorageError("Review prefix has no reusable Sampling v6 identity")
best_entries = list(self.validated_prefix.entries) best_entries = list(self.validated_prefix.entries)
best_records = list(self.validated_prefix.records) best_records = list(self.validated_prefix.records)
if self._plan_import is not None: if self._plan_import is not None:
@@ -2742,6 +2850,9 @@ class RunStorageController:
candidates.sort(key=lambda m: (len(m.get("chunks") or []), candidates.sort(key=lambda m: (len(m.get("chunks") or []),
int((m.get("branch_provenance") or {}).get("canonical_sequence", 0))), reverse=True) int((m.get("branch_provenance") or {}).get("canonical_sequence", 0))), reverse=True)
for candidate in candidates: for candidate in candidates:
if not _reusable_sampling_contract(contract, resume_safe=safe) or not _reusable_manifest(candidate):
self.notes.append(f"stored revision {candidate.get('revision_id', '')} is read-only or unobservable")
continue
if len(candidate.get("chunks") or []) <= len(best_entries): if len(candidate.get("chunks") or []) <= len(best_entries):
continue continue
if int(candidate.get("run_storage_schema_version", -1)) not in RUN_STORAGE_READABLE_SCHEMA_VERSIONS: if int(candidate.get("run_storage_schema_version", -1)) not in RUN_STORAGE_READABLE_SCHEMA_VERSIONS:
+1 -1
View File
@@ -37,7 +37,7 @@ def review_queue_results():
capture_output=True, text=True, encoding="utf-8", timeout=45, capture_output=True, text=True, encoding="utf-8", timeout=45,
) )
records = json.loads(result.stdout) records = json.loads(result.stdout)
assert len(records) == 60, "The complete queue fixture must finish, including Workflow rehydration" assert len(records) == 62, "The complete queue fixture must finish, including Workflow rehydration and v5 history"
failures = [item for item in records if not item["pass"]] failures = [item for item in records if not item["pass"]]
assert result.returncode == 0 and not failures, result.stdout + result.stderr assert result.returncode == 0 and not failures, result.stdout + result.stderr
return {item["name"]: item for item in records} return {item["name"]: item for item in records}
+99
View File
@@ -0,0 +1,99 @@
{
"active_revisions": {
"1": "f0d03f641973fb1d89c2ca4a",
"2": "d5b6382c7152f233ac608bfa"
},
"branch_provenance_version": 1,
"canonical_chain": [
"f0d03f641973fb1d89c2ca4a",
"d5b6382c7152f233ac608bfa"
],
"canonical_head_revision_id": "d5b6382c7152f233ac608bfa",
"canonical_storage_revision_id": "d8352287737286e5",
"group_revisions": [
{
"branch_provenance_version": 1,
"created_utc": "2026-10-03T15:37:36+00:00",
"generation_contract_sha256": "f5a4b7ec0041d5f51731148aae29b228608c1904aa7b3d922764697dc97daf46",
"group": {
"end": 1,
"physical_group": 1,
"start": 1
},
"lineage_sha256": "2dbba4b61863fde2a8d74a75dae4b563e7b79484a2ff678d0cbd333e504cf158",
"parent_revision_id": null,
"records": [
{
"file_sha256": "26e771cffc57f171f4cc4c6859876350bdf717d1e8213fc005343006a34f002f",
"file_size": 110080,
"filename": "rev-d8352287737286e5-txn-9bb0a584b7344342b114bd31b175225c-chunk-0001.safetensors",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"seed": 17671095936967226055,
"sequence_index": 0,
"storage_revision_id": "d8352287737286e5"
}
],
"revision_id": "f0d03f641973fb1d89c2ca4a",
"revision_order": "2026-10-03T15:37:36+00:00|f0d03f641973fb1d89c2ca4a",
"seed_identity": "d54e2bdb426c4a8217b8d7a5cf608093c8a53b6087982053d294f61e042cbf49",
"storage_revision_id": "d8352287737286e5",
"variation_nonce": 0
},
{
"branch_provenance_version": 1,
"created_utc": "2026-10-03T15:37:36+00:00",
"generation_contract_sha256": "f5a4b7ec0041d5f51731148aae29b228608c1904aa7b3d922764697dc97daf46",
"group": {
"end": 2,
"physical_group": 2,
"start": 2
},
"lineage_sha256": "2dbba4b61863fde2a8d74a75dae4b563e7b79484a2ff678d0cbd333e504cf158",
"parent_revision_id": "f0d03f641973fb1d89c2ca4a",
"records": [
{
"file_sha256": "8b8ca1b0c802d024c6915d903747d8722bea25628369997ab008ff896462b96a",
"file_size": 124928,
"filename": "rev-d8352287737286e5-txn-008f52b93a88461481ead25f91627af0-chunk-0002.safetensors",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"seed": 2167717532453298830,
"sequence_index": 1,
"storage_revision_id": "d8352287737286e5"
}
],
"revision_id": "d5b6382c7152f233ac608bfa",
"revision_order": "2026-10-03T15:37:36+00:00|d5b6382c7152f233ac608bfa",
"seed_identity": "ea9acc75bcb213362286781997985bc03df80bb8e3b3368ab02907a465286579",
"storage_revision_id": "d8352287737286e5",
"variation_nonce": 0
}
],
"revisions": [
{
"active_chain": [
"f0d03f641973fb1d89c2ca4a",
"d5b6382c7152f233ac608bfa"
],
"branch_regenerate_from": 0,
"contract_sha256": "f5a4b7ec0041d5f51731148aae29b228608c1904aa7b3d922764697dc97daf46",
"effective_reroll_nonce": 0,
"nonce_mode": "inactive",
"reroll_from_chunk": 0,
"resume_safe": true,
"review_control_version": 1,
"review_pause_reason": "review_each_chunk",
"review_unit": {
"end": 2,
"physical_group": 2,
"start": 2
},
"revision_id": "d8352287737286e5",
"sampling_revision_id": "f5a4b7ec0041d5f5",
"status": "review_ready",
"updated_utc": "2026-10-03T15:37:37+00:00"
}
],
"run_name": "run_00bb27779f215038a7eb1845979fdd7c",
"run_storage_schema_version": 3,
"updated_utc": "2026-10-03T15:37:37+00:00"
}
@@ -0,0 +1,726 @@
{
"branch_provenance": {
"active_chain": [
"f0d03f641973fb1d89c2ca4a",
"d5b6382c7152f233ac608bfa"
],
"canonical_committed_utc": "2026-10-03T15:37:37+00:00",
"canonical_sequence": 1,
"version": 1
},
"branch_regenerate_from": 0,
"chunks": [
{
"entry": {
"clip_index": 1,
"context_frames": 0,
"motion_score": 0.0,
"plan": {
"clip_index": 1,
"context_frames": 5,
"continuation": false,
"debug": false,
"height": 64,
"magic": "H3_CONTINUUM_PLAN",
"net_frames": 124,
"requested_extend_seconds": 5.0,
"schema_version": 1,
"state_capacity_frames": 39,
"total_frames": 124,
"trim_frames": 0,
"width": 64
},
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"seed": 17671095936967226055,
"sequence_index": 0
},
"file_sha256": "26e771cffc57f171f4cc4c6859876350bdf717d1e8213fc005343006a34f002f",
"file_size": 110080,
"filename": "rev-d8352287737286e5-txn-9bb0a584b7344342b114bd31b175225c-chunk-0001.safetensors",
"sequence_index": 0,
"storage_revision_id": "d8352287737286e5"
},
{
"entry": {
"clip_index": 2,
"context_frames": 22,
"motion_score": 0.0,
"plan": {
"clip_index": 2,
"context_frames": 22,
"continuation": true,
"debug": false,
"height": 64,
"magic": "H3_CONTINUUM_PLAN",
"net_frames": 119,
"requested_extend_seconds": 5.0,
"schema_version": 1,
"state_capacity_frames": 39,
"total_frames": 141,
"trim_frames": 22,
"width": 64
},
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"seed": 2167717532453298830,
"sequence_index": 1
},
"file_sha256": "8b8ca1b0c802d024c6915d903747d8722bea25628369997ab008ff896462b96a",
"file_size": 124928,
"filename": "rev-d8352287737286e5-txn-008f52b93a88461481ead25f91627af0-chunk-0002.safetensors",
"sequence_index": 1,
"storage_revision_id": "d8352287737286e5"
}
],
"contract": {
"chunk_contract_hashes": [
"a1d04437be62e79c3945c65d26bbc2cc8fdab89c661d78e8dd8da4673b9f1f15",
"b90a1eab80aaefd1db1935d6aec14050332ccc6047a7045bf52728de4f527259",
"ffdc4b800b697062ab3d2e27c9079dda7fcd2e4c2a7d83fc98fac7402ffaaf55",
"b07b8643f7706d41ae6a0b0a74c1f6771d844502ef8b006f5cf71664a0ba26f4"
],
"chunk_contracts": [
{
"chunk_number": 1,
"effective_reroll_nonce": 0,
"global_hash": "3f2ddb7a42a25548a6ec9319f859ed1a75ae0a11933056953d722e2553af357d",
"last_frame_hash": "",
"physical_group_reference_sha256": "ac4a88fbfb79802981cae2f4ce7b3befcfc5b40fa14f1cd9d03852a2e7a4678f",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"reroll_boundary": 0
},
{
"chunk_number": 2,
"effective_reroll_nonce": 0,
"global_hash": "3f2ddb7a42a25548a6ec9319f859ed1a75ae0a11933056953d722e2553af357d",
"last_frame_hash": "",
"physical_group_reference_sha256": "24824ce6a4f406415e52b1909270780063bece5f97fc3fdb2e844d40878afdbe",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"reroll_boundary": 0
},
{
"chunk_number": 3,
"effective_reroll_nonce": 0,
"global_hash": "3f2ddb7a42a25548a6ec9319f859ed1a75ae0a11933056953d722e2553af357d",
"last_frame_hash": "",
"physical_group_reference_sha256": "b9fcc63b033a0cdf334e68dad74f77ac87ff5e1f8c3b346aa5879411856f62da",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"reroll_boundary": 0
},
{
"chunk_number": 4,
"effective_reroll_nonce": 0,
"global_hash": "3f2ddb7a42a25548a6ec9319f859ed1a75ae0a11933056953d722e2553af357d",
"last_frame_hash": "6bc13118765285cba097403c4b340bac0e550599fb27a1b415ffc88fc6237daa",
"physical_group_reference_sha256": "b9fcc63b033a0cdf334e68dad74f77ac87ff5e1f8c3b346aa5879411856f62da",
"prompt_hash": "5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"reroll_boundary": 0
}
],
"chunk_count": 4,
"effective_reroll_nonce": 0,
"global": {
"actual_prefix_steps": 2,
"audio_continuity": true,
"base_seed": 1234,
"chunk_seconds": 5.0,
"clip": {
"graph_route": {
"node": {
"class_type": "CLIPLoader",
"inputs": {
"clip_name": "audit_minimax_h3.safetensors",
"type": "minimax"
},
"node_id": "clip"
},
"output": 0
},
"runtime": {
"conditioner": {
"parameter_count": 4,
"parameter_layout_sha256": "bfc7a2fa460fae4eb803fa7bedfa5a2cb8c3c793902b65c1e7796159b92d79bd",
"tensor_count": 1,
"type": "audit_providers.TinyModule",
"weight_probe": [
{
"dtype": "torch.float32",
"name": "proj.weight",
"sample_sha256": "c3a6b1f08b0b05ac05390d6c257551ffd0cdcf40496f232b52df2498f915469e",
"shape": [
2,
2
]
}
]
},
"options": {
"layer_idx": null,
"tokenizer_options": {},
"use_clip_schedule": false
},
"patcher": {
"model_type": "audit_providers.TinyModule",
"state": {
"model_options": {
"transformer_options": {}
},
"object_patches": {},
"patches": {},
"wrappers": {}
},
"type": "comfy.model_patcher.ModelPatcher"
},
"runtime_observation": {
"options_complete": true,
"patcher_complete": true
},
"tokenizer": {
"bos_token_id": null,
"core_type": "types.SimpleNamespace",
"eos_token_id": null,
"model_max_length": 1024,
"model_wrapper_type": "types.SimpleNamespace",
"name_or_path": "audit-type-only",
"pad_token_id": null,
"padding_side": null,
"truncation_side": null,
"vocab_size": 10,
"wrapper_type": "types.SimpleNamespace"
},
"wrapper_type": "audit_providers.SyntheticClip"
}
},
"conditioning_mode": "fl2va",
"continuity": "Balanced - 22 frames",
"continuum_interop_api": 1,
"driving_audio": {
"driving_audio_contract_version": 1,
"preprocess_version": 1,
"resampled_dtype": "torch.float32",
"resampled_sha256": "7cca6be8869d599159077330aaf71972b0328bf8758cd99e6d52ba5bd3466b7a",
"resampled_shape": [
1,
2,
640000
],
"resolved_vae_sample_rate": 32000,
"source_dtype": "torch.float32",
"source_sample_rate": 32000,
"source_sha256": "7cca6be8869d599159077330aaf71972b0328bf8758cd99e6d52ba5bd3466b7a",
"source_shape": [
1,
2,
640000
]
},
"driving_audio_vae": {
"graph_route": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_audio_vae.safetensors"
},
"node_id": "audio_vae"
},
"output": 0
},
"runtime": {
"audio_sample_rate": 32000,
"first_stage": {
"parameter_count": 4,
"parameter_layout_sha256": "bfc7a2fa460fae4eb803fa7bedfa5a2cb8c3c793902b65c1e7796159b92d79bd",
"tensor_count": 1,
"type": "audit_providers.TinyModule",
"weight_probe": [
{
"dtype": "torch.float32",
"name": "proj.weight",
"sample_sha256": "cfd5cd0554b977e5208296d4a63e8d2ec6f6cc9823321f2d242239432b7d472b",
"shape": [
2,
2
]
}
]
},
"latent_channels": 32,
"patcher": {
"model_type": "audit_providers.TinyModule",
"state": {
"model_options": {
"transformer_options": {}
},
"object_patches": {},
"patches": {},
"wrappers": {}
},
"type": "comfy.model_patcher.ModelPatcher"
},
"runtime_observation": {
"config_complete": true,
"patcher_complete": true
},
"vae_dtype": "torch.float32",
"wrapper_type": "audit_providers.SyntheticAudioVAE"
}
},
"execution_semantics": {
"continuation_transport": "masked_av_prefix_22_v1",
"flf_execution": "terminal_merged_10s_seed_v2",
"terminal_seed_policy": "physical_window_seed_v2"
},
"first_frame_hash": "ce95397885a93617f1e00a1e6c9a4b5a9c03411a8f2ef58762709df00b6e95d6",
"height": 64,
"model": {
"graph_route": {
"node": {
"class_type": "UNETLoader",
"inputs": {
"unet_name": "audit_minimax_h3_fl2va.safetensors",
"weight_dtype": "default"
},
"node_id": "model"
},
"output": 0
},
"runtime": {
"base": "audit_core_payload.TinyCoreH3",
"dtype": "torch.float32",
"inner": "audit_providers.TinyModule",
"legacy_fingerprint": "2acbeaa9b63a6cf12d7a1b88a1533c12cf5329df420822568eaea5a0bc59bcd4",
"model_patcher": "comfy.model_patcher.ModelPatcher",
"model_size": 16,
"patches": {},
"runtime_observation": {
"patches_complete": true,
"transformer_options_complete": true,
"wrappers_complete": true
},
"transformer_options": {},
"weight_probe": [
{
"dtype": "torch.float32",
"name": "proj.weight",
"sample_sha256": "f6bb1294da2f78cd935b01c7656280df5eaa0439e9d97bc03775825a41a508e4",
"shape": [
2,
2
]
}
],
"wrappers": {}
}
},
"reference_audio": {
"combined_hash": "a55474f5b6c557a5eaf70e78a5341a49a39fd2b0cb717b5d08396645d6e2ded8",
"count": 2,
"items": [
{
"audio_index": 1,
"reference_audio": {
"combined_hash": "64a489714dfa65dc229ad059608cd6f488ba1b2d49e815fd71403bfd100f2452",
"preprocess_version": 1,
"reference_audio_contract_version": 1,
"resampled_dtype": "torch.float32",
"resampled_sha256": "7ad1bd40edaed4e0a4601dca856191e3002a8b09b8551ec74b81365faecf0548",
"resampled_shape": [
1,
2,
3200
],
"resolved_vae_sample_rate": 32000,
"source_dtype": "torch.float32",
"source_sample_rate": 32000,
"source_sha256": "7ad1bd40edaed4e0a4601dca856191e3002a8b09b8551ec74b81365faecf0548",
"source_shape": [
1,
2,
3200
]
}
},
{
"audio_index": 2,
"reference_audio": {
"combined_hash": "5b8b8a88f1a7b2adf9005a4e9c32625b493812bb37d33837ee4d99890adafe54",
"preprocess_version": 1,
"reference_audio_contract_version": 1,
"resampled_dtype": "torch.float32",
"resampled_sha256": "69ac770debe6804e3d5762f61896bc226f266f57b8398ea3bed3dc67fde92a8a",
"resampled_shape": [
1,
2,
3200
],
"resolved_vae_sample_rate": 32000,
"source_dtype": "torch.float32",
"source_sample_rate": 32000,
"source_sha256": "69ac770debe6804e3d5762f61896bc226f266f57b8398ea3bed3dc67fde92a8a",
"source_shape": [
1,
2,
3200
]
}
}
],
"reference_audio_bundle_contract_version": 1
},
"reference_audio_vae": {
"graph_route": {
"node": {
"class_type": "H3ContinuumReferenceAudios",
"inputs": {
"reference_audio_vae": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_audio_vae.safetensors"
},
"node_id": "audio_vae"
},
"output": 0
}
},
"node_id": "audio_bundle"
},
"output": 0
},
"runtime": {
"audio_sample_rate": 32000,
"first_stage": {
"parameter_count": 4,
"parameter_layout_sha256": "bfc7a2fa460fae4eb803fa7bedfa5a2cb8c3c793902b65c1e7796159b92d79bd",
"tensor_count": 1,
"type": "audit_providers.TinyModule",
"weight_probe": [
{
"dtype": "torch.float32",
"name": "proj.weight",
"sample_sha256": "cfd5cd0554b977e5208296d4a63e8d2ec6f6cc9823321f2d242239432b7d472b",
"shape": [
2,
2
]
}
]
},
"latent_channels": 32,
"patcher": {
"model_type": "audit_providers.TinyModule",
"state": {
"model_options": {
"transformer_options": {}
},
"object_patches": {},
"patches": {},
"wrappers": {}
},
"type": "comfy.model_patcher.ModelPatcher"
},
"runtime_observation": {
"config_complete": true,
"patcher_complete": true
},
"vae_dtype": "torch.float32",
"wrapper_type": "audit_providers.SyntheticAudioVAE"
}
},
"reference_routing_contract_version": 1,
"sampler": {
"function": "torch._VariableFunctionsClass.add",
"options": {
"extra_options": {},
"inpaint_options": {}
},
"type": "types.SimpleNamespace"
},
"sampling_contract_version": 5,
"sigmas": {
"dtype": "torch.float32",
"exact_sha256": "be115d3ffbb025877d7320bd3b0b29cbe245f520f805091917a60169e7737659",
"shape": [
3
]
},
"strict_compatibility": false,
"upstream_graph": {
"routes": {
"audio_vae": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_audio_vae.safetensors"
},
"node_id": "audio_vae"
},
"output": 0
},
"clip": {
"node": {
"class_type": "CLIPLoader",
"inputs": {
"clip_name": "audit_minimax_h3.safetensors",
"type": "minimax"
},
"node_id": "clip"
},
"output": 0
},
"model": {
"node": {
"class_type": "UNETLoader",
"inputs": {
"unet_name": "audit_minimax_h3_fl2va.safetensors",
"weight_dtype": "default"
},
"node_id": "model"
},
"output": 0
},
"reference_audio_vae": {
"node": {
"class_type": "H3ContinuumReferenceAudios",
"inputs": {
"reference_audio_vae": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_audio_vae.safetensors"
},
"node_id": "audio_vae"
},
"output": 0
}
},
"node_id": "audio_bundle"
},
"output": 0
},
"video_vae": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_video_vae.safetensors"
},
"node_id": "video_vae"
},
"output": 0
}
},
"sampler_class_type": "H3ContinuumSamplerV39",
"sha256": "f7f6f31d4e90a07b040f2774b93ff2e405e0565f8443f3ff1fb5c546394f2ca3",
"version": 3
},
"video_vae": {
"graph_route": {
"node": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "audit_video_vae.safetensors"
},
"node_id": "video_vae"
},
"output": 0
},
"runtime": {
"chunked_io": false,
"downscale_ratio": 16,
"first_stage": {
"parameter_count": 4,
"parameter_layout_sha256": "bfc7a2fa460fae4eb803fa7bedfa5a2cb8c3c793902b65c1e7796159b92d79bd",
"tensor_count": 1,
"type": "audit_providers.TinyModule",
"weight_probe": [
{
"dtype": "torch.float32",
"name": "proj.weight",
"sample_sha256": "24c60715698663563d76ecaf262778712879423eb057d12b74a037150009d619",
"shape": [
2,
2
]
}
]
},
"latent_channels": 24,
"patcher": {
"model_type": "audit_providers.TinyModule",
"state": {
"model_options": {
"transformer_options": {}
},
"object_patches": {},
"patches": {},
"wrappers": {}
},
"type": "comfy.model_patcher.ModelPatcher"
},
"required": true,
"runtime_observation": {
"config_complete": true,
"patcher_complete": true
},
"upscale_ratio": 16,
"vae_dtype": "torch.float32",
"wrapper_type": "audit_providers.SyntheticVideoVAE"
}
},
"width": 64
},
"last_frame_hash": "6bc13118765285cba097403c4b340bac0e550599fb27a1b415ffc88fc6237daa",
"nonce_lifecycle": {
"effective_nonce": 0,
"lineage_sha256": "2dbba4b61863fde2a8d74a75dae4b563e7b79484a2ff678d0cbd333e504cf158",
"mode": "inactive",
"request_sha256": "55c6f2d10dbe9abae01443bac060ba1d56d583d0f80e5f3cfc586cef54a65fcd",
"requested_nonce": 0
},
"nonce_lineage_sha256": "2dbba4b61863fde2a8d74a75dae4b563e7b79484a2ff678d0cbd333e504cf158",
"nonce_request_sha256": "55c6f2d10dbe9abae01443bac060ba1d56d583d0f80e5f3cfc586cef54a65fcd",
"prompt_hashes": [
"5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07",
"5dec1ecc3f0d6f846562049c0a4968da02cbbb43113feb87a95cfd6d6691be07"
],
"prompt_mode": "Fixed — one prompt",
"reference_group_contracts": [
{
"descriptor": {
"effective_prompt_sha256": "4de5a7dfe7c4a22a4f42e26d525e5175d90a85fe6eaa1a093663f3e5bf8ad6b1",
"first_anchor": "ce95397885a93617f1e00a1e6c9a4b5a9c03411a8f2ef58762709df00b6e95d6",
"last_anchor": "6bc13118765285cba097403c4b340bac0e550599fb27a1b415ffc88fc6237daa",
"logical_chunks": [
1
],
"physical_group": 1,
"picture_order": [
"first_frame",
"last_frame",
"R1"
],
"picture_presentation_version": 1,
"reference_preprocess_version": 1,
"reference_size_mode": "Match Output",
"references": [
{
"picture_number": 3,
"prepared_sha256": "4e2f68facb219819b6b0ba6c2c48f37a64b02ae159aef772e2e5d1dea15a3e62",
"source_slot_id": "R1"
}
],
"terminal_atomic": false,
"terminal_prompt_policy": null,
"version": 1
},
"sha256": "ac4a88fbfb79802981cae2f4ce7b3befcfc5b40fa14f1cd9d03852a2e7a4678f"
},
{
"descriptor": {
"effective_prompt_sha256": "4de5a7dfe7c4a22a4f42e26d525e5175d90a85fe6eaa1a093663f3e5bf8ad6b1",
"first_anchor": "ce95397885a93617f1e00a1e6c9a4b5a9c03411a8f2ef58762709df00b6e95d6",
"last_anchor": "6bc13118765285cba097403c4b340bac0e550599fb27a1b415ffc88fc6237daa",
"logical_chunks": [
2
],
"physical_group": 2,
"picture_order": [
"first_frame",
"last_frame",
"R1"
],
"picture_presentation_version": 1,
"reference_preprocess_version": 1,
"reference_size_mode": "Match Output",
"references": [
{
"picture_number": 3,
"prepared_sha256": "4e2f68facb219819b6b0ba6c2c48f37a64b02ae159aef772e2e5d1dea15a3e62",
"source_slot_id": "R1"
}
],
"terminal_atomic": false,
"terminal_prompt_policy": null,
"version": 1
},
"sha256": "24824ce6a4f406415e52b1909270780063bece5f97fc3fdb2e844d40878afdbe"
},
{
"descriptor": {
"effective_prompt_sha256": "4de5a7dfe7c4a22a4f42e26d525e5175d90a85fe6eaa1a093663f3e5bf8ad6b1",
"first_anchor": "ce95397885a93617f1e00a1e6c9a4b5a9c03411a8f2ef58762709df00b6e95d6",
"last_anchor": "6bc13118765285cba097403c4b340bac0e550599fb27a1b415ffc88fc6237daa",
"logical_chunks": [
3,
4
],
"physical_group": 3,
"picture_order": [
"first_frame",
"last_frame",
"R1"
],
"picture_presentation_version": 1,
"reference_preprocess_version": 1,
"reference_size_mode": "Match Output",
"references": [
{
"picture_number": 3,
"prepared_sha256": "4e2f68facb219819b6b0ba6c2c48f37a64b02ae159aef772e2e5d1dea15a3e62",
"source_slot_id": "R1"
}
],
"terminal_atomic": true,
"terminal_prompt_policy": "shared_prompt_v1",
"version": 1
},
"sha256": "b9fcc63b033a0cdf334e68dad74f77ac87ff5e1f8c3b346aa5879411856f62da"
}
],
"reroll_from_chunk": 0
},
"contract_sha256": "f5a4b7ec0041d5f51731148aae29b228608c1904aa7b3d922764697dc97daf46",
"created_utc": "2026-10-03T15:37:36+00:00",
"effective_reroll_nonce": 0,
"nonce_lifecycle": {
"effective_nonce": 0,
"lineage_sha256": "2dbba4b61863fde2a8d74a75dae4b563e7b79484a2ff678d0cbd333e504cf158",
"mode": "inactive",
"request_sha256": "55c6f2d10dbe9abae01443bac060ba1d56d583d0f80e5f3cfc586cef54a65fcd",
"requested_nonce": 0
},
"reference_routing_audit": {
"requested_routes": [
[
"R1"
],
[
"R1"
],
[
"R1"
]
],
"routing_config_sha256": "a3b1b8850a2c72eaa4e7c6c4f2697bb8bd6194091b677e5267338236e298734e",
"version": 1
},
"report_summary": "H3 Continuum V3 3.9.0\nConditioning mode: Hybrid FL2VA + Reference + Continuation.\nPrompt plan: Fixed — one prompt; 4 chunks; 1 unique prompt(s); 5.000s per chunk. Auto detected Fixed.\nPrompt Preflight: OK.\nDecode: external ComfyUI Core VAE nodes; full raw AV chunks retained.\naccelerator markers not detected (informational only); Continuum APPLY_MODEL wrapper installed\nExecution: conditioning precomputed; single call-local MODEL clone per chunk; no internal VAE decode.\nDriving Audio: absolute-time guide slices enabled; original audio is preserved for final output.\nReference Audio conditioning: 2 ordered item(s); persistent across all chunks.\nV3.9 Custom Reference routing: group-local conditioning and complete-group saved-prefix verification; the verified effective Plan is available after Queue.\nFLF execution: terminal merged 10s seed v2; the final two 5-second logical chunks share one physical sample and are split back into normal chunk entries.\nresuming after accepted chunk 1\nPhysical review execution: 1 new physical group(s) completed; 2/4 logical chunks available.\nSession e03555e8: 2 chunks, 243 retained frames, 1 reused, 64x64.\nOutput: 2 raw AV latent chunk(s); connect Core VAE Decode nodes, then H3 Continuum Assemble V3.\nDecode preview: Chunk 2\nReference Plan Inspector (last Queue, runtime verified)\nMode: custom; accepted chunks: 2\nConnected: R1\nGroup 1 [C1] reused — R1=<Picture 3>; contract=ac4a88fbfb79\nGroup 2 [C2] generated — R1=<Picture 3>; contract=24824ce6a4f4\nGroup 3 [C3,4] pending — configured R1\nRun Storage: run_00bb27779f215038a7eb1845979fdd7c / revision d8352287737286e5; 1 reused, 1 generated, 4 total; resume=2; nonce=0 (inactive).\nWarning: Review is incomplete. Second Pass refinement of a partial review sequence is not yet supported.",
"resume_disabled_reasons": [],
"resume_safe": true,
"review_control_version": 1,
"review_pause_reason": "review_each_chunk",
"review_unit": {
"end": 2,
"physical_group": 2,
"start": 2
},
"revision_id": "d8352287737286e5",
"run_name": "run_00bb27779f215038a7eb1845979fdd7c",
"run_storage_schema_version": 3,
"sampling_contract_version": 5,
"sampling_revision_id": "f5a4b7ec0041d5f5",
"session_id": "e03555e805774b9eb22b7d2eae44268c",
"status": "review_ready",
"updated_utc": "2026-10-03T15:37:37+00:00"
}
+24 -2
View File
@@ -11,6 +11,7 @@ function project(n=1,total=3,id='r1',run='fixture') {
let parent=null; let parent=null;
const groups=Array.from({length:n},(_,i)=>{ const groups=Array.from({length:n},(_,i)=>{
const g={revision_id:`${id}-g${i+1}`,parent_revision_id:parent,lineage_sha256:'lineage', const g={revision_id:`${id}-g${i+1}`,parent_revision_id:parent,lineage_sha256:'lineage',
sampling_contract_version:6,generation_reusable:true,
revision_order:String(i+1),take_number:1,group:{start:i+1,end:i+1,physical_group:i+1}}; revision_order:String(i+1),take_number:1,group:{start:i+1,end:i+1,physical_group:i+1}};
parent=g.revision_id;return g; parent=g.revision_id;return g;
}); });
@@ -19,6 +20,7 @@ function project(n=1,total=3,id='r1',run='fixture') {
canonical_chain:groups.map(g=>g.revision_id), canonical_chain:groups.map(g=>g.revision_id),
active_revisions:Object.fromEntries(groups.map(g=>[String(g.group.physical_group),g.revision_id])), active_revisions:Object.fromEntries(groups.map(g=>[String(g.group.physical_group),g.revision_id])),
group_revisions:groups,revisions:[{revision_id:id,status:n===total?'complete':'review_ready', group_revisions:groups,revisions:[{revision_id:id,status:n===total?'complete':'review_ready',
sampling_contract_version:6,generation_reusable:true,
updated_utc:`2026-09-08T00:00:0${n}Z`,review_unit:{start:n,end:n,physical_group:n}}]}; updated_utc:`2026-09-08T00:00:0${n}Z`,review_unit:{start:n,end:n,physical_group:n}}]};
} }
function environment() { function environment() {
@@ -464,8 +466,8 @@ await test('mode policy and transient UI serialization remain stable',async e=>{
}); });
await test('history catalog preserves lineage eligibility and compact summary',async e=>{ await test('history catalog preserves lineage eligibility and compact summary',async e=>{
const n=e.makeNode();e.w(n,'generation_mode').value='Review Each Chunk';const p=project(3,3); const n=e.makeNode();e.w(n,'generation_mode').value='Review Each Chunk';const p=project(3,3);
p.group_revisions.push({revision_id:'old',lineage_sha256:'other',group:{start:1,end:1,physical_group:1}}, p.group_revisions.push({revision_id:'old',lineage_sha256:'other',sampling_contract_version:6,generation_reusable:true,group:{start:1,end:1,physical_group:1}},
{revision_id:'orphan',parent_revision_id:'missing',lineage_sha256:'lineage',group:{start:2,end:2,physical_group:2}}); {revision_id:'orphan',parent_revision_id:'missing',lineage_sha256:'lineage',sampling_contract_version:6,generation_reusable:true,group:{start:2,end:2,physical_group:2}});
await e.load(n,p);assert.equal(e.f.takeCatalog(n).length,3); await e.load(n,p);assert.equal(e.f.takeCatalog(n).length,3);
e.w(n,'Render History').callback();const body=e.w(n,'Render History / Takes'); e.w(n,'Render History').callback();const body=e.w(n,'Render History / Takes');
assert(body.computeSize(400)[1]<=124);const drawn=[]; assert(body.computeSize(400)[1]<=124);const drawn=[];
@@ -675,5 +677,25 @@ await test('failed graph load releases hydration gate for subsequent manual node
await n.__h3ContinuumConfigurePending?.promise;await n.__h3ContinuumHistoryRequest?.promise;await sleep(); await n.__h3ContinuumConfigurePending?.promise;await n.__h3ContinuumHistoryRequest?.promise;await sleep();
assert(e.visible(n,'Use it and continue')); assert(e.visible(n,'Use it and continue'));
}); });
await test('Sampling v5 history stays visible and actions remain read-only',async e=>{
const p=project(2);for(const item of [...p.group_revisions,...p.revisions]){
item.sampling_contract_version=5;item.generation_reusable=false;
}
const n=e.makeNode();await e.load(n,p);
assert.equal(e.f.takeCatalog(n).length,2);
assert.match(e.f.reviewStatus(n),/read-only/);
const before=[e.w(n,'take_action').value,e.w(n,'take_revision_id').value];
e.f.selectTakeAction(n,'Use This Take');
assert.deepEqual([e.w(n,'take_action').value,e.w(n,'take_revision_id').value],before);
assert(!e.visible(n,'Use it and continue'));
});
await test('archived v5 lineage remains browsable after a v6 canonical run',async e=>{
const old=project(2,'fixture',3,'old');for(const item of old.group_revisions){
item.sampling_contract_version=5;item.generation_reusable=false;item.lineage_sha256='old-lineage';
}
const current=project(1);current.group_revisions.push(...old.group_revisions);
const n=e.makeNode();await e.load(n,current);
assert.equal(e.f.takeCatalog(n).length,3);
});
console.log(JSON.stringify(results,null,2));if(results.some(r=>!r.pass))process.exitCode=1; console.log(JSON.stringify(results,null,2));if(results.some(r=>!r.pass))process.exitCode=1;
})(); })();
+2 -1
View File
@@ -25,7 +25,7 @@ def test_v3_reference_video_imports_the_shared_temporal_helper():
def test_reference_bundle_fingerprint_ignores_unused_legacy_vae(): def test_reference_bundle_fingerprint_ignores_unused_legacy_vae():
graph = { graph = {
"1": {"class_type": "H3ContinuumSamplerV38", "inputs": { "1": {"class_type": "H3ContinuumSamplerV38", "inputs": {
"model": ["2", 0], "clip": ["3", 0], "model": ["2", 0], "clip": ["3", 0], "sampler": ["8", 0],
"audio_references": ["4", 0], "reference_audio_vae": ["5", 0]}}, "audio_references": ["4", 0], "reference_audio_vae": ["5", 0]}},
"2": {"class_type": "UNETLoader", "inputs": {"unet_name": "model"}}, "2": {"class_type": "UNETLoader", "inputs": {"unet_name": "model"}},
"3": {"class_type": "CLIPLoader", "inputs": {"clip_name": "clip"}}, "3": {"class_type": "CLIPLoader", "inputs": {"clip_name": "clip"}},
@@ -34,6 +34,7 @@ def test_reference_bundle_fingerprint_ignores_unused_legacy_vae():
"5": {"class_type": "VAELoader", "inputs": {"vae_name": "unused"}}, "5": {"class_type": "VAELoader", "inputs": {"vae_name": "unused"}},
"6": {"class_type": "VAELoader", "inputs": {"vae_name": "active"}}, "6": {"class_type": "VAELoader", "inputs": {"vae_name": "active"}},
"7": {"class_type": "LoadAudio", "inputs": {"audio": "voice.wav"}}, "7": {"class_type": "LoadAudio", "inputs": {"audio": "voice.wav"}},
"8": {"class_type": "KSamplerSelect", "inputs": {"sampler_name": "euler"}},
} }
def contract(value): def contract(value):
+1 -1
View File
@@ -6,7 +6,7 @@ import tomllib
ROOT = Path(__file__).resolve().parents[1] ROOT = Path(__file__).resolve().parents[1]
EXPECTED_VERSION = "3.9.0" EXPECTED_VERSION = "3.9.1"
def test_v35_release_metadata_is_consistent(): def test_v35_release_metadata_is_consistent():
+1 -1
View File
@@ -145,4 +145,4 @@ def test_v217_result_pack_expands_legacy_outputs():
def test_v217_has_no_frontend_display_controller(): def test_v217_has_no_frontend_display_controller():
assert not (ROOT / "web" / "h3_continuum_v2.js").exists() assert not (ROOT / "web" / "h3_continuum_v2.js").exists()
version_source = (ROOT / "version.py").read_text(encoding="utf-8") version_source = (ROOT / "version.py").read_text(encoding="utf-8")
assert 'PACKAGE_VERSION = "3.9.0"' in version_source assert 'PACKAGE_VERSION = "3.9.1"' in version_source
+5 -4
View File
@@ -990,10 +990,11 @@ def _prompt_graph(model_name="minimax_h3_fl2va.safetensors"):
"class_type": "VAELoader", "class_type": "VAELoader",
"inputs": {"vae_name": "minimax_h3_video_vae_fp16.safetensors"}, "inputs": {"vae_name": "minimax_h3_video_vae_fp16.safetensors"},
}, },
"42": { "42": {
"class_type": "H3ContinuumSamplerProduction", "class_type": "H3ContinuumSamplerProduction",
"inputs": {"model": ["12", 0], "clip": ["20", 0], "video_vae": ["30", 0]}, "inputs": {"model": ["12", 0], "clip": ["20", 0], "video_vae": ["30", 0], "sampler": ["40", 0]},
}, },
"40": {"class_type": "KSamplerSelect", "inputs": {"sampler_name": "euler"}},
} }
+2
View File
@@ -261,11 +261,13 @@ def test_run_storage_graph_contract_fingerprints_helper_vae_and_sources():
"inputs": { "inputs": {
"model": ["model", 0], "model": ["model", 0],
"clip": ["clip", 0], "clip": ["clip", 0],
"sampler": ["ksampler", 0],
"audio_references": ["refs", 0], "audio_references": ["refs", 0],
}, },
}, },
"model": {"class_type": "UNETLoader", "inputs": {"name": "model.safetensors"}}, "model": {"class_type": "UNETLoader", "inputs": {"name": "model.safetensors"}},
"clip": {"class_type": "CLIPLoader", "inputs": {"name": "text.safetensors"}}, "clip": {"class_type": "CLIPLoader", "inputs": {"name": "text.safetensors"}},
"ksampler": {"class_type": "KSamplerSelect", "inputs": {"sampler_name": "euler"}},
"refs": { "refs": {
"class_type": "H3ContinuumReferenceAudios", "class_type": "H3ContinuumReferenceAudios",
"inputs": { "inputs": {
+2
View File
@@ -424,12 +424,14 @@ setTimeout(async () => {{
canonical_storage_revision_id: "storage-ready", canonical_storage_revision_id: "storage-ready",
revisions: [{{ revisions: [{{
revision_id: "storage-ready", status: "review_ready", revision_id: "storage-ready", status: "review_ready",
sampling_contract_version: 6, generation_reusable: true,
review_unit: {{ start: 1, end: 1, physical_group: 1 }}, review_unit: {{ start: 1, end: 1, physical_group: 1 }},
}}], }}],
canonical_head_revision_id: "take-1", canonical_head_revision_id: "take-1",
active_revisions: {{ "1": "take-1" }}, active_revisions: {{ "1": "take-1" }},
group_revisions: [{{ group_revisions: [{{
revision_id: "take-1", revision_order: "1", revision_id: "take-1", revision_order: "1",
sampling_contract_version: 6, generation_reusable: true,
group: {{ start: 1, end: 1, physical_group: 1 }}, group: {{ start: 1, end: 1, physical_group: 1 }},
}}], }}],
}}; }};
+2
View File
@@ -89,6 +89,7 @@ def _contract(
) -> dict: ) -> dict:
global_contract = { global_contract = {
"sampling_contract_version": SAMPLING_CONTRACT_VERSION, "sampling_contract_version": SAMPLING_CONTRACT_VERSION,
"runtime_signature_version": 1,
"conditioning_mode": "t2va", "conditioning_mode": "t2va",
"width": 96, "width": 96,
"height": 64, "height": 64,
@@ -435,6 +436,7 @@ def test_rr_r5_selected_foreign_take_requires_whole_compatible_prefix(tmp_path):
"chunks": [], "chunks": [],
} }
controller._write_manifest() controller._write_manifest()
controller.resume_safe = True
imported = controller._adopt_plan_prefix(controller._plan_import) imported = controller._adopt_plan_prefix(controller._plan_import)
assert len(imported) == 2 assert len(imported) == 2
assert all(record["storage_revision_id"] == controller.revision_id for record in imported) assert all(record["storage_revision_id"] == controller.revision_id for record in imported)
+157
View File
@@ -0,0 +1,157 @@
"""Real conditioning/Second Pass contracts for V3.9 output-scope projection."""
from copy import deepcopy
import pytest
import torch
from ComfyUI_H3_Continuum_Join.v3 import driving_nodes
from ComfyUI_H3_Continuum_Join.v3 import refine_context as rc
from ComfyUI_H3_Continuum_Join.v3.reference_storage_contract import canonical_sha256
from ComfyUI_H3_Continuum_Join.v3.second_pass import run_second_pass_groups
def fixture(chunks=((1,), (2,), (3,)), captured=2, selected=(0, 1)):
groups, plans, routes = [], [], []
for index, logical in enumerate(chunks):
descriptor = {"physical_group": index + 1, "logical_chunks": list(logical),
"references": [{"source_slot_id": "R1", "picture_number": 2}]}
route = {"descriptor": descriptor, "sha256": canonical_sha256(descriptor)}
if index < captured:
groups.append(rc.make_refine_group(
conditioning=[[torch.full((1, 1, 2), float(index + 1)), {
"minimax_frame_count": 124,
"minimax_keyframes": [{"audio_latent": torch.ones(1, 32, 2, 7)}],
}]], group_id=index, logical_chunks=logical, physical_frames=124,
prompt_policy="single", physical_prompt=f"prompt {index + 1}",
source_video_shape=(1, 24, 37, 2, 3), physical_clip_index=logical[0],
context_frames=0 if index == 0 else 22,
reference_routing_group_contract=route,
))
plans.append({
"group_id": index, "logical_chunks": list(logical), "physical_frames": 124,
"physical_prompt": f"prompt {index + 1}", "prompt_policy": "single",
"trim_prefix_frames": 0 if index == 0 else 22,
"terminal_merged": len(logical) == 2, "source_width": 48, "source_height": 32,
"source_batch": 1, "latent_channels": 24, "source_latent_t": 37,
"source_latent_h": 2, "source_latent_w": 3, "source_audio_shape": [1, 8, 148],
})
routes.append({"physical_group": index + 1, "logical_chunks": list(logical),
"status": "reused" if index < captured else "pending",
"requested_slots": ["R1"], "warnings": [],
**({"contract_sha256": route["sha256"], "descriptor": descriptor}
if index < captured else {})})
context = rc.make_refine_context(groups, source_width=48, source_height=32,
conditioning_mode="t2va", complete=captured == len(chunks))
plan = {"width": 48, "height": 32,
"second_pass_contract": {"version": 1, "physical_groups": [
{**plans[index], "group_id": position} for position, index in enumerate(selected)]},
"reference_routing_v1": {"mode": "custom", "accepted_chunks": captured,
"connected_slots": ["R1"], "groups": routes}}
return context, plan
@pytest.mark.parametrize("selected", [(0,), (0, 1), (1,)])
def test_partial_views_are_complete_for_output_without_completing_sequence(selected):
source, plan = fixture(selected=selected)
view = rc.project_refine_context(source, plan)
assert source["complete"] is view["complete"] is False
assert rc.OUTPUT_SCOPE_KEY not in source
assert view[rc.OUTPUT_SCOPE_KEY]["source_group_ids"] == selected
assert rc.validate_refine_context(view, assembly_plan=plan) is view
for local_index, source_index in enumerate(selected):
assert view["groups"][local_index]["group_id"] == local_index
assert view["groups"][local_index]["physical_clip_index"] == source_index + 1
assert view["groups"][local_index]["conditioning"] is source["groups"][source_index]["conditioning"]
def test_complete_full_output_is_exact_context_passthrough():
context, plan = fixture(captured=3, selected=(0, 1, 2))
assert rc.project_refine_context(context, plan) is context
def test_terminal_pair_is_one_physical_scope():
source, plan = fixture(chunks=((1,), (2, 3), (4,)), selected=(1,))
view = rc.project_refine_context(source, plan)
assert view["groups"][0]["logical_chunks"] == (2, 3)
broken = deepcopy(plan)
broken["second_pass_contract"]["physical_groups"][0]["logical_chunks"] = [2]
assert rc.project_refine_context(source, broken) is source
def test_missing_capture_keeps_optional_fallback():
source, plan = fixture(captured=1)
assert rc.project_refine_context(source, plan) is source
@pytest.mark.parametrize("field,value", [
("physical_frames", 125), ("physical_prompt", "different"),
("trim_prefix_frames", 23), ("source_latent_t", 38), ("source_latent_w", 4),
])
def test_mismatched_partial_evidence_is_diagnostic_not_a_new_stop(field, value):
source, plan = fixture(selected=(1,))
plan["second_pass_contract"]["physical_groups"][0][field] = value
result = rc.project_refine_context(source, plan)
assert rc.OUTPUT_SCOPE_KEY not in result
assert result["complete"] is False
assert any("Output scope unavailable:" in note for note in result["notes"])
@pytest.mark.parametrize("corruption", ["hash", "duplicate", "pending"])
def test_reference_identity_cannot_be_guessed_by_local_group_index(corruption):
source, plan = fixture(selected=(1,))
routes = plan["reference_routing_v1"]["groups"]
if corruption == "hash":
routes[1]["contract_sha256"] = "0" * 64
elif corruption == "duplicate":
routes.append(deepcopy(routes[1]))
else:
routes[1].pop("descriptor")
view = rc.project_refine_context(source, plan)
assert rc.OUTPUT_SCOPE_KEY not in view
def test_projected_context_is_still_revalidated_at_second_pass_boundary():
source, plan = fixture(selected=(1,))
view = rc.project_refine_context(source, plan)
plan["second_pass_contract"]["physical_groups"][0]["physical_prompt"] = "changed"
with pytest.raises(rc.RefineContextError, match="physical_prompt"):
rc.validate_refine_context(view, assembly_plan=plan)
def test_chunk_two_actual_second_pass_uses_its_conditions_and_preserves_audio():
source, plan = fixture(selected=(1,))
view = rc.project_refine_context(source, plan)
video = {"samples": torch.zeros(1, 24, 37, 2, 3)}
audio = {"samples": torch.ones(1, 8, 148)}
calls = []
def sample(**kwargs):
calls.append(kwargs["conditioning"])
return kwargs["latent"]
_video, output_audio, updated, report = run_second_pass_groups(
model=object(), clip=object(), sampler=object(), sigmas=torch.tensor([.2, 0.]),
video_latents=[video], audio_latents=[audio], assembly_plan=plan,
refine_seed=123, refine_context=view,
encode_prompt_fn=lambda *args, **kwargs: pytest.fail("must not reencode prompt alone"),
clone_model_fn=lambda model, **kwargs: model,
latent_builder=lambda v, a: {"video": v, "audio": a},
sample_fn=sample, stream_extractor=lambda latent: (latent["video"], latent["audio"]),
)
assert float(calls[0][0][0][0, 0, 0]) == 2.0
assert output_audio[0] is audio
assert updated["second_pass_contract"]["reference_inheritance"][0]["status"] == "verified"
assert "conditioning_source=refine_context" in report
assert "prompt_only_fallback" not in report
assert source["groups"][1]["group_id"] == 1
def test_v39_boundary_projects_only_its_runtime_output(monkeypatch):
source, plan = fixture(selected=(1,))
from ComfyUI_H3_Continuum_Join.v3.nodes import _PARTIAL_REVIEW_SECOND_PASS_WARNING
old = (object(), object(), plan, _PARTIAL_REVIEW_SECOND_PASS_WARNING, object(), source)
monkeypatch.setattr(driving_nodes.H3ContinuumSamplerV38, "run", lambda *args, **kwargs: old)
result = driving_nodes.H3ContinuumSamplerV39().run(chunks=3)["result"]
assert result[:3] == old[:3]
assert result[4] is old[4]
assert result[5][rc.OUTPUT_SCOPE_KEY]["source_group_ids"] == (1,)
assert "output-scope conditioning verified" in result[3]
assert old[3] == _PARTIAL_REVIEW_SECOND_PASS_WARNING
+42
View File
@@ -0,0 +1,42 @@
import pytest
import torch
from ComfyUI_H3_Continuum_Join.v3 import refine_context as rc
def group(keyframe):
return rc.make_refine_group(
group_id=0,logical_chunks=[1],physical_frames=124,prompt_policy='single',
physical_prompt='prompt',source_video_shape=(1,24,37,2,3),
physical_clip_index=1,context_frames=0,first_image=None,last_image=None,
conditioning=[[torch.zeros(1,2,3),{'minimax_keyframes':[keyframe]}]],
)
@pytest.mark.parametrize('resize',[False,True])
@pytest.mark.parametrize('kind',['audio','video','mixed'])
def test_modalities_preserve_audio_and_frozen_source(kind,resize):
keyframe={'resolved_frame_index':0}
if kind!='video': keyframe['audio_latent']=torch.arange(32*2*7).reshape(1,32,2,7).float()
if kind!='audio': keyframe['latent']=torch.ones(1,24,1,2,3)
source=group(keyframe)
frozen=source['conditioning'][0][1]['minimax_keyframes'][0]
before={k:v.clone() for k,v in frozen.items() if torch.is_tensor(v)}
result,stats=rc.adapt_group_conditioning(source,4 if resize else 2,6 if resize else 3)
actual=result[0][1]['minimax_keyframes'][0]
assert actual is not frozen
assert actual['resolved_frame_index']==0
if kind!='video':
assert actual['audio_latent'] is frozen['audio_latent']
assert torch.equal(actual['audio_latent'],before['audio_latent'])
if kind=='audio':
assert 'latent' not in actual
assert stats['keyframes_resized']==0
else:
assert actual['latent'].shape[-2:]==((4,6) if resize else (2,3))
for k,v in before.items(): assert torch.equal(frozen[k],v)
@pytest.mark.parametrize('payload',[{}, {'latent':torch.zeros(1,2,3)},
{'audio_latent':torch.zeros(1,2)},
{'latent':torch.zeros(1,2,3),'audio_latent':torch.zeros(1,32,2,7)},
{'latent':torch.zeros(1,24,1,2,3),'audio_latent':'broken'}])
def test_invalid_payload_is_not_hidden_by_another_modality(payload):
with pytest.raises(rc.RefineConditioningAdaptationError):
rc.adapt_group_conditioning(group({'resolved_frame_index':0,**payload}),4,6)
+225
View File
@@ -0,0 +1,225 @@
from __future__ import annotations
import copy
import functools
import hashlib
import json
from pathlib import Path
import shutil
from types import SimpleNamespace
import pytest
from ComfyUI_H3_Continuum_Join import run_storage as storage
from ComfyUI_H3_Continuum_Join.graph_contract import build_upstream_graph_contract
from ComfyUI_H3_Continuum_Join.v3.review_control import (
GENERATION_MODE_REVIEW, REVIEW_ACTION_CONTINUE, REVIEW_ACTION_REGENERATE_CURRENT,
ReviewControlError,
)
from . import test_v31_run_storage as signatures
from . import test_v38_review_run_storage as review
def closure(gain):
def sample(x=None, **kwargs):
return x * gain
return sample
@pytest.mark.parametrize("option", [
"sampler_cfg_function", "sampler_pre_cfg_function", "sampler_post_cfg_function",
"model_function_wrapper", "denoise_mask_function",
])
def test_model_hook_changes_are_observed(option):
model = signatures._Model()
model.model_options[option] = closure(1)
before, safe = storage._model_signature(model, "base")
model.model_options[option] = closure(2)
after, after_safe = storage._model_signature(model, "base")
assert safe and after_safe
assert before != after
def test_cfg1_and_internal_join_context():
model = signatures._Model()
before, safe = storage._model_signature(model, "base")
model.model_options["disable_cfg1_optimization"] = True
after, safe_after = storage._model_signature(model, "base")
assert safe and safe_after and before != after
model.model_options["transformer_options"] = {"h3_continuum_join_context": object()}
join, safe_join = storage._model_signature(model, "base")
model.model_options["transformer_options"] = {}
clean, safe_clean = storage._model_signature(model, "base")
assert safe_join and safe_clean and join == clean
def test_sampler_closure_only_changes_identity_with_graph_unchanged():
graph, graph_safe, _ = build_upstream_graph_contract(
signatures._prompt_graph(), "42", require_video_vae=True,
)
model, clip, vae = signatures._Model(), signatures._Clip(), signatures._VideoVAE()
sampler = SimpleNamespace(sampler_function=closure(1), extra_options={}, inpaint_options={})
kwargs = dict(model=model, clip=clip, video_vae=vae, sampler=sampler,
upstream_graph_contract=graph, upstream_graph_safe=graph_safe)
before, safe, _ = signatures._sampling_contract(**kwargs)
sampler.sampler_function = closure(2)
after, after_safe, _ = signatures._sampling_contract(**kwargs)
assert safe and after_safe
assert before["global"]["upstream_graph"] == after["global"]["upstream_graph"]
assert storage.revision_identity(before) != storage.revision_identity(after)
same, same_safe, _ = signatures._sampling_contract(**kwargs)
assert same_safe and same == after
def test_sampler_graph_only_changes_identity():
graph = signatures._prompt_graph()
before, safe, _ = build_upstream_graph_contract(graph, "42", require_video_vae=True)
graph["40"]["inputs"]["sampler_name"] = "heun"
after, after_safe, _ = build_upstream_graph_contract(graph, "42", require_video_vae=True)
assert safe and after_safe and before != after
assert before["version"] == after["version"] == 4
assert before["routes"]["model"] == after["routes"]["model"]
def test_partial_and_bound_method_state_are_observed():
class Bound:
def __init__(self, gain):
self.gain = gain
def sample(self, value):
return value * self.gain
first, safe = storage._observe(functools.partial(Bound(1).sample, 2))
second, safe_second = storage._observe(functools.partial(Bound(3).sample, 2))
assert safe and safe_second and first != second
@pytest.mark.parametrize("location", ["wrappers", "patches", "model_options"])
def test_opaque_state_disables_reuse_without_preventing_contract(location):
class Opaque:
__slots__ = ("gain",)
def __init__(self):
self.gain = 1
def __call__(self, *args):
return self.gain
model = signatures._Model()
getattr(model, location)["external"] = Opaque()
contract, safe, reasons = signatures._sampling_contract(model=model)
assert contract["global"]["sampling_contract_version"] == 6
assert not safe and reasons
def test_empty_callable_attributes_are_not_complete():
class Opaque:
def __call__(self):
return 1
_, safe = storage._observe(Opaque())
assert not safe
def test_observation_bounds_preserve_ordinary_lora_keys_without_weight_traversal():
_, keys_safe = storage._observe({f"weight.{index}" for index in range(500)})
_, oversized_safe = storage._observe(list(range(4097)))
module = signatures._Model().model.diffusion_model
descriptor, module_safe = storage._observe(module)
assert keys_safe and not oversized_safe and not module_safe
assert descriptor["module_state_unobserved"]
def test_signature_failure_is_diagnostic_not_a_generation_stop():
model = signatures._Model()
model.model_dtype = lambda: (_ for _ in ()).throw(RuntimeError("custom dtype"))
contract, safe, reasons = signatures._sampling_contract(model=model)
assert contract and not safe and reasons
def test_core_euler_sampler_has_an_observable_runtime_signature():
from comfy.samplers import sampler_object
_, safe = storage._sampler_signature(sampler_object("euler"))
assert safe
def reader(tmp_path, contract):
controller = review._controller(tmp_path, contract)
controller.configure_review(generation_mode=GENERATION_MODE_REVIEW,
review_action=REVIEW_ACTION_CONTINUE)
return controller
@pytest.mark.parametrize("entry_point", ["head", "review", "import", "compatible", "take"])
def test_safe_false_never_adopts_a_prefix(tmp_path, entry_point):
contract = review._contract(chunks=3)
source = review._persist(tmp_path, contract, prefix=2, review_unit=(2, 2),
updated_utc="2026-10-04T00:00:00+00:00")
controller = reader(tmp_path, contract)
original = (source._manifest_path().read_bytes(), (source.run_root / "project.json").read_bytes())
if entry_point == "head":
assert controller.find_latest_review_head(contract, resume_safe=False) is None
elif entry_point == "review":
assert not controller._load_validated_review_prefix(contract, resume_safe=False).entries
elif entry_point == "import":
assert controller._find_plan_import(contract, resume_safe=False) is None
elif entry_point == "compatible":
assert controller._compatible_plan_prefix(source.manifest, contract, resume_safe=False) == ([], [])
else:
controller.take_group = 2
controller.take_revision_id = review._project(source)["canonical_head_revision_id"]
with pytest.raises(storage.RunStorageError, match="unobservable"):
controller._validated_take_selection(contract, resume_safe=False)
assert original == (source._manifest_path().read_bytes(), (source.run_root / "project.json").read_bytes())
def test_safe_false_retry_preserves_existing_take(tmp_path):
contract = review._contract(chunks=3)
source = review._persist(tmp_path, contract, prefix=2, review_unit=(2, 2),
updated_utc="2026-10-04T00:00:00+00:00")
controller = reader(tmp_path, contract)
controller.review_action = REVIEW_ACTION_REGENERATE_CURRENT
before = source._manifest_path().read_bytes()
with pytest.raises(ReviewControlError, match="could not be matched"):
controller._load_validated_review_prefix(contract, resume_safe=False)
assert source._manifest_path().read_bytes() == before
def test_identical_v6_review_and_take_still_reuse(tmp_path):
contract = review._contract(chunks=3)
source = review._persist(tmp_path, contract, prefix=2, review_unit=(2, 2),
updated_utc="2026-10-04T00:00:00+00:00")
controller = reader(tmp_path, contract)
assert len(controller._load_validated_review_prefix(contract, resume_safe=True).entries) == 2
controller.take_group = 2
controller.take_revision_id = review._project(source)["canonical_head_revision_id"]
assert len(controller._validated_take_selection(contract)["entries"]) == 2
def fixture_hashes(root):
return {str(file.relative_to(root)): hashlib.sha256(file.read_bytes()).hexdigest()
for file in root.rglob("*") if file.is_file()}
def test_actual_old_code_v5_fixture_readable_and_immutable(tmp_path):
fixture = Path(__file__).parent / "fixtures" / "v39_sampling_v5"
before = fixture_hashes(fixture)
run = tmp_path / "run"
shutil.copytree(fixture, run)
controller = storage.RunStorageController("legacy-read")
controller.run_root = run
controller.revisions_root = run / "revisions"
manifests, catalog, chains = controller._provenance_catalog()
assert len(manifests) == 1 and len(catalog) == 2 and len(chains) == 1
manifest = next(iter(manifests.values()))
assert manifest["sampling_contract_version"] == 5
assert manifest["contract"]["global"]["sampling_contract_version"] == 5
storage._manifest_sampling_identity(manifest)
head = controller._validated_review_head(manifest, for_reuse=False)
assert head["validated_prefix_count"] == 2
assert len(head["entries"]) == 2
contract = review._contract(chunks=manifest["contract"]["chunk_count"])
controller.prompts = [""] * contract["chunk_count"]
assert controller.find_latest_review_head(contract) is None
assert controller._compatible_plan_prefix(manifest, contract) == ([], [])
assert controller._find_plan_import(contract) is None
controller.take_group = 2
controller.take_revision_id = head["manifest"]["branch_provenance"]["active_chain"][-1]
with pytest.raises(storage.RunStorageError, match="read-only"):
controller._validated_take_selection(contract, source_lineage=manifest["contract"]["nonce_lineage_sha256"])
assert fixture_hashes(fixture) == before
assert fixture_hashes(run) == before
+178
View File
@@ -0,0 +1,178 @@
from __future__ import annotations
import copy
import gc
from types import SimpleNamespace
import pytest
import torch
from ComfyUI_H3_Continuum_Join import driving_audio
from ComfyUI_H3_Continuum_Join.v3 import driving_nodes
from ComfyUI_H3_Continuum_Join.v3.nodes import _apply_review_decode_scope
from ComfyUI_H3_Continuum_Join.v3.plan import (
AssemblyPlanError, REVIEW_AUDIO_PROJECTION_KEY, make_review_audio_projection,
prepare_physical_decode_entries, validate_assembly_plan,
)
from ComfyUI_H3_Continuum_Join.v3.second_pass import update_second_pass_geometry
from ComfyUI_H3_Continuum_Join.v3.file_backed_buffer import get_file_backed_image_manager
from . import test_v38_review_run_storage as review
from . import test_p23_assembly_v35 as assembly
def scope(*, chunk=2, finish=False):
contract = review._contract(chunks=4)
entries = [review._entry(position, contract) for position in range(3)]
decoded, full_plan = prepare_physical_decode_entries(
entries, chunk_seconds=5, preserve_final_frame=False, terminal_merged=False,
)
result = {"session": {"chunks": entries}, "report": "source"}
outputs = ([{"samples": entry["video"]} for entry in decoded],
[{"samples": entry["audio"]} for entry in decoded], full_plan, result)
storage = SimpleNamespace(review_generation_mode="Review Each Chunk",
review_execution=SimpleNamespace(finish_remaining=finish,
next_review_unit_start=chunk, next_review_unit_end=chunk,
partial_review=True, smart_regenerate=False))
scoped = _apply_review_decode_scope(
outputs, storage=storage, capture_refine_context=False, configured_chunks=4,
chunk_seconds=5, first_frame=None, last_frame=None, timeline_video_source=None,
)
return scoped, outputs
def finalize(plan, source, *, exact=False, backend="RAM", direct=False):
images, generated = assembly._decoded(plan, sample_rate=32000)
if not direct:
plan = dict(plan, _h3_continuum_driving_audio_v1=source)
return driving_nodes.H3ContinuumAssembleSeamV35().assemble(
images, generated, plan, exact, "Off", "Off", backend, "Off",
driving_audio=source if direct else None,
)
@pytest.fixture
def buffer_root(tmp_path, monkeypatch):
original = driving_nodes.assemble_decoded_chunks_v35
def assemble_at_test_root(**kwargs):
return original(**kwargs, backing_root=tmp_path)
monkeypatch.setattr(driving_nodes, "assemble_decoded_chunks_v35", assemble_at_test_root)
yield tmp_path
gc.collect()
get_file_backed_image_manager(tmp_path).collect_ready()
@pytest.mark.parametrize("rate", [32000, 44100, 48000])
@pytest.mark.parametrize("channels", [1, 2])
@pytest.mark.parametrize("exact", [False, True])
@pytest.mark.parametrize("backend", ["RAM", "Disk-backed", "Auto"])
def test_finalize_uses_natural_absolute_pcm_origin(rate, channels, exact, backend, buffer_root):
scoped, original = scope()
plan = scoped[2]
unit = original[2]["chunks"][1]
samples = round(20 * rate)
waveform = torch.arange(samples, dtype=torch.float32).view(1, 1, -1).repeat(1, channels, 1)
source = {"waveform": waveform, "sample_rate": rate}
frozen = waveform.clone()
start = round(unit["frame_start"] * rate / 24)
stop = round(unit["frame_stop"] * rate / 24)
assert unit["frame_start"] == 124 and unit["frame_stop"] == 243
expected = waveform[..., start:stop]
if exact:
wanted = round(plan["target_frames"] * rate / 24)
padding = max(0, wanted - expected.shape[-1])
# The unchanged V3.9 Exact rule repeats the final available PCM sample.
expected = torch.cat((expected[..., :wanted], expected[..., -1:].expand(1, channels, padding)), dim=-1)
for direct in (False, True):
images, audio, _ = finalize(plan, source, exact=exact, backend=backend, direct=direct)
assert torch.equal(audio["waveform"], expected)
assert audio["sample_rate"] == rate
assert images.shape[0] == (120 if exact else 119)
del images, audio
assert torch.equal(waveform, frozen)
assert scoped[3]["session"] is original[3]["session"]
assert len(original[3]["session"]["chunks"]) == 3
assert REVIEW_AUDIO_PROJECTION_KEY not in original[2]
assert REVIEW_AUDIO_PROJECTION_KEY not in original[3]["session"]
@pytest.mark.parametrize("length", [160000, 0])
@pytest.mark.parametrize("exact", [False, True])
def test_short_or_empty_review_audio_only_pads_when_required(length, exact, buffer_root):
plan = scope()[0][2]
waveform = torch.ones((1, 2, length), dtype=torch.float64)
source = {"waveform": waveform, "sample_rate": 32000}
images, output, report = finalize(plan, source, exact=exact)
assert output["waveform"].dtype == waveform.dtype
# Chunk 2 begins after these short sources. Use the selected output's real size.
assert output["waveform"].shape == (1, 2, round(images.shape[0] * 32000 / 24))
assert not torch.count_nonzero(output["waveform"])
assert "Empty Review interval" in report
assert torch.equal(source["waveform"], waveform)
def test_partial_short_audio_is_not_padded_with_exact_off(buffer_root):
plan = scope()[0][2]
start = round(124 * 32000 / 24)
waveform = torch.arange(start + 1000, dtype=torch.float32).view(1, 1, -1)
_, output, report = finalize(plan, {"waveform": waveform, "sample_rate": 32000})
assert torch.equal(output["waveform"], waveform[..., start:])
assert output["waveform"].shape[-1] == 1000
assert "silence added" not in report
def test_finalize_repeat_does_not_cut_the_pcm_twice(buffer_root):
plan = scope()[0][2]
source = {"waveform": torch.arange(640000, dtype=torch.float32).view(1, 1, -1), "sample_rate": 32000}
plan["_h3_continuum_driving_audio_v1"] = source
before = source["waveform"].clone()
first = finalize(plan, source)
second = finalize(plan, source)
assert torch.equal(first[1]["waveform"], second[1]["waveform"])
assert first[1]["waveform"][0, 0, 0] == round(124 * 32000 / 24)
assert torch.equal(plan["_h3_continuum_driving_audio_v1"]["waveform"], before)
def test_full_projection_keeps_source_pcm_unchanged(buffer_root):
scoped, full = scope(finish=True)
assert scoped is full
assert REVIEW_AUDIO_PROJECTION_KEY not in full[2]
source = {"waveform": torch.arange(640000, dtype=torch.float32).view(1, 1, -1), "sample_rate": 32000}
_, output, _ = finalize(full[2], source, exact=False)
assert torch.equal(output["waveform"], source["waveform"])
def test_terminal_physical_group_cannot_be_split():
plan = assembly._plan("long_terminal_3x5")
projection = make_review_audio_projection(plan, start_chunk=2, end_chunk=3)
group = plan["decode_groups"][1]
assert (projection["source_start_frame"], projection["source_stop_frame"]) == (group["frame_start"], group["frame_stop"])
with pytest.raises(AssemblyPlanError, match="split"):
make_review_audio_projection(plan, start_chunk=3, end_chunk=3)
@pytest.mark.parametrize("field,value", [
("version", 2), ("coordinate_space", "local"), ("fps", 30),
("start_chunk", 1), ("source_start_frame", -1), ("source_stop_frame", 242),
("source_stop_frame", 124),
])
def test_broken_metadata_has_no_first_audio_fallback(field, value):
plan = scope()[0][2]
plan[REVIEW_AUDIO_PROJECTION_KEY][field] = value
with pytest.raises(AssemblyPlanError, match="Review audio projection"):
driving_audio.slice_review_source_audio({"waveform": torch.ones(1, 1, 200), "sample_rate": 32000}, plan)
def test_second_pass_geometry_copy_preserves_projection_and_source():
scoped, _ = scope()
plan = scoped[2]
source = {"waveform": torch.arange(100, dtype=torch.float32).view(1, 1, -1), "sample_rate": 32000}
plan["_h3_continuum_driving_audio_v1"] = source
before = copy.deepcopy(plan[REVIEW_AUDIO_PROJECTION_KEY])
group = plan["second_pass_contract"]["physical_groups"][0]
shape = (group["source_batch"], group["latent_channels"], group["source_latent_t"],
group["source_latent_h"] * 2, group["source_latent_w"] * 2)
updated = update_second_pass_geometry(plan, [{"samples": torch.zeros(shape)}])
assert updated[REVIEW_AUDIO_PROJECTION_KEY] == before
assert torch.equal(updated["_h3_continuum_driving_audio_v1"]["waveform"], source["waveform"])
assert plan[REVIEW_AUDIO_PROJECTION_KEY] == before
validate_assembly_plan(updated)
+139
View File
@@ -0,0 +1,139 @@
from __future__ import annotations
import builtins
from types import SimpleNamespace
import pytest
import torch
from ComfyUI_H3_Continuum_Join import audio_compat, reference, reference_audio, driving_audio
from ComfyUI_H3_Continuum_Join.v3.ref_encode_cache import clear_ref_encode_cache
from . import test_v38_ref_encode_cache as cache_cases
def test_equal_rate_returns_original_without_importing_a_resampler(monkeypatch):
monkeypatch.setattr(audio_compat.importlib, "import_module", lambda *args: pytest.fail("unneeded import"))
waveform = torch.arange(200, dtype=torch.float32)
assert audio_compat.resample_audio(waveform, 32000, 32000) is waveform
@pytest.mark.parametrize("rate", [16000, 22050, 32000, 44100, 48000, 96000])
@pytest.mark.parametrize("channels", [1, 2])
def test_resample_is_exactly_core(rate, channels):
from comfy.audio import resample
waveform = torch.sin(torch.arange(rate // 10, dtype=torch.float32) * 0.0123).view(1, 1, -1).repeat(1, channels, 1)
frozen = waveform.clone()
expected = resample(waveform, rate, 32000)
actual = audio_compat.resample_audio(waveform, rate, 32000)
assert actual.shape == expected.shape
assert actual.dtype == expected.dtype
assert torch.equal(actual, expected)
assert torch.equal(waveform, frozen)
@pytest.mark.parametrize("absence", ["module", "api"])
def test_older_core_uses_the_legacy_adapter_only_if_api_absent(monkeypatch, absence):
calls = []
def legacy(waveform, source_rate, target_rate):
calls.append((source_rate, target_rate))
return waveform + 1
def modules(name):
if name == "comfy.audio":
if absence == "module":
raise ModuleNotFoundError("old Core", name="comfy.audio")
return SimpleNamespace()
assert name == "torchaudio"
return SimpleNamespace(functional=SimpleNamespace(resample=legacy))
monkeypatch.setattr(audio_compat.importlib, "import_module", modules)
waveform = torch.zeros(1, 1, 20)
assert torch.equal(audio_compat.resample_audio(waveform, 44100, 32000), waveform + 1)
assert calls == [(44100, 32000)]
@pytest.mark.parametrize("failure", ["internal_dependency", "core_computation", "both_absent"])
def test_errors_never_switch_silently_to_a_different_resampler(monkeypatch, failure):
calls = []
def core(*args):
raise ValueError("Core computation failed")
def modules(name):
calls.append(name)
if name == "comfy.audio":
if failure == "internal_dependency":
raise ModuleNotFoundError("broken scipy", name="scipy.signal")
if failure == "core_computation":
return SimpleNamespace(resample=core)
raise ModuleNotFoundError("old Core", name="comfy.audio")
raise ModuleNotFoundError("no legacy module", name="torchaudio")
monkeypatch.setattr(audio_compat.importlib, "import_module", modules)
with pytest.raises((ModuleNotFoundError, ValueError, RuntimeError), match={
"internal_dependency": "broken scipy", "core_computation": "Core computation failed",
"both_absent": "requires comfy.audio.resample",
}[failure]):
audio_compat.resample_audio(torch.zeros(1, 1, 20), 44100, 32000)
if failure != "both_absent":
assert calls == ["comfy.audio"]
@pytest.mark.parametrize("rate", [32000, 44100, 48000])
def test_reference_bundle_and_driving_do_not_require_torchaudio(rate, monkeypatch):
original_import = builtins.__import__
def no_torchaudio(name, *args, **kwargs):
if name == "torchaudio" or name.startswith("torchaudio."):
raise ModuleNotFoundError("Torchaudio deliberately absent", name="torchaudio")
return original_import(name, *args, **kwargs)
monkeypatch.setattr(builtins, "__import__", no_torchaudio)
from comfy.audio import resample
waveform = torch.sin(torch.arange(rate, dtype=torch.float32) / 10).view(1, 1, -1)
source = {"waveform": waveform, "sample_rate": rate}
vae = cache_cases.AudioVAE()
reference_source = reference_audio.prepare_reference_audio_source(source, vae)
bundle = reference_audio.H3ContinuumReferenceAudios().pack(source, None, None, vae)[0]
assert bundle is not None
assert reference_source.source_sample_rate == rate
assert torch.equal(reference_source.waveform, resample(waveform, rate, 32000))
driving = driving_audio.prepare_driving_audio_source(source, vae, target_frames=12, fps=24)
effective = waveform[..., :round(rate / 2)]
assert torch.equal(driving.source_audio["waveform"], effective)
assert driving.source_audio["sample_rate"] == rate
assert torch.equal(driving.resampled_waveform, resample(effective, rate, 32000))
assert driving.contract["preprocess_version"] == 1
same = reference_audio.prepare_reference_audio_source(source, vae)
assert same.combined_hash == reference_source.combined_hash
assert torch.equal(source["waveform"], waveform)
@pytest.mark.parametrize("kind", ["image", "audio"])
def test_direct_same_vae_weight_mutation_requires_clear_or_reload(kind):
clear_ref_encode_cache()
class WeightedImage(cache_cases.ImageVAE):
def __init__(self):
super().__init__()
self.weight = torch.tensor(1.)
def encode(self, image):
return super().encode(image) + self.weight
class WeightedAudio(cache_cases.AudioVAE):
def __init__(self):
super().__init__()
self.weight = torch.tensor(1.)
def encode(self, waveform):
return super().encode(waveform) + self.weight
vae = WeightedImage() if kind == "image" else WeightedAudio()
source = cache_cases._image_assets("stability-r5-image") if kind == "image" else cache_cases._audio_source("stability-r5-audio")
def encode(current):
if kind == "image":
return reference.encode_reference_latents_cached(current, source).latents[0]
return reference_audio.encode_reference_audio_cached(current, source).audio_latent
cold = encode(vae)
same = encode(vae)
assert vae.calls == 1 and torch.equal(cold, same)
vae.weight.add_(1)
stale = encode(vae)
assert vae.calls == 1 and torch.equal(cold, stale)
clear_ref_encode_cache()
refreshed = encode(vae)
assert vae.calls == 2 and torch.equal(refreshed, cold + 1)
reloaded = type(vae)()
reloaded.weight.fill_(3)
assert torch.equal(encode(reloaded), cold + 2)
assert reloaded.calls == 1
clear_ref_encode_cache()
+6 -2
View File
@@ -37,7 +37,7 @@ from .file_backed_buffer import (
format_auto_buffer_decision, format_auto_buffer_decision,
select_auto_buffer_backend, select_auto_buffer_backend,
) )
from .plan import FPS, validate_assembly_plan from .plan import FPS, validate_assembly_plan, REVIEW_AUDIO_PROJECTION_KEY
_COPY_BATCH_FRAMES = 8 _COPY_BATCH_FRAMES = 8
@@ -526,7 +526,9 @@ def _write_assembly(
driving_audio_report = "" driving_audio_report = ""
if final_audio_override is not None: if final_audio_override is not None:
result_audio = dict(final_audio_override) result_audio = dict(final_audio_override)
if exact_total_duration: override_waveform, _ = validate_audio(result_audio)
empty_review = (REVIEW_AUDIO_PROJECTION_KEY in plan and int(override_waveform.shape[-1]) == 0)
if exact_total_duration or empty_review:
result_audio, _ = adjust_audio_duration_direct( result_audio, _ = adjust_audio_duration_direct(
result_audio, result_audio,
target_frames=output_frames, target_frames=output_frames,
@@ -543,6 +545,8 @@ def _write_assembly(
f"sample_rate={int(selected_rate)}, samples={int(waveform.shape[-1])}; " f"sample_rate={int(selected_rate)}, samples={int(waveform.shape[-1])}; "
"generated audio and Audio Seam bypassed." "generated audio and Audio Seam bypassed."
) )
if empty_review:
driving_audio_report += " Empty Review interval: silence added for the selected video duration."
reports.append( reports.append(
f"Assembled {len(units)} physical decoded group(s) into " f"Assembled {len(units)} physical decoded group(s) into "
+17 -2
View File
@@ -16,7 +16,7 @@ from ..constants import (
normalize_continuity_mode, normalize_continuity_mode,
normalize_diagnostics_mode, normalize_diagnostics_mode,
) )
from ..driving_audio import prepare_driving_audio_source from ..driving_audio import prepare_driving_audio_source, slice_review_source_audio
from ..guide_timeline import ( from ..guide_timeline import (
GUIDE_TYPE, GUIDE_TYPE,
make_still_image_guide, make_still_image_guide,
@@ -854,6 +854,20 @@ class H3ContinuumSamplerV39(H3ContinuumSamplerV38):
plan = outputs[2].get("reference_routing_v1") plan = outputs[2].get("reference_routing_v1")
if not isinstance(plan, dict): if not isinstance(plan, dict):
raise RuntimeError("V3.9 Reference Routing has no verified output Plan") raise RuntimeError("V3.9 Reference Routing has no verified output Plan")
if len(outputs) == 6 and isinstance(outputs[5], dict):
from .refine_context import MAGIC, OUTPUT_SCOPE_KEY, project_refine_context
if outputs[5].get("magic") == MAGIC:
context = project_refine_context(outputs[5], outputs[2])
if context is not outputs[5]:
status = outputs[3]
if context.get(OUTPUT_SCOPE_KEY) is not None:
from .nodes import _PARTIAL_REVIEW_SECOND_PASS_WARNING
status = str(status).replace(
_PARTIAL_REVIEW_SECOND_PASS_WARNING,
"Second Pass: output-scope conditioning verified; "
"the full sequence may still be incomplete.",
)
outputs = (*outputs[:3], status, outputs[4], context)
return { return {
"ui": {"h3_reference_plan": [format_reference_plan(plan)]}, "ui": {"h3_reference_plan": [format_reference_plan(plan)]},
"result": outputs, "result": outputs,
@@ -1102,7 +1116,8 @@ class H3ContinuumAssembleSeamV35(H3ContinuumAssembleSeamV34):
backend = str(_singleton(buffer_backend, "buffer_backend")) backend = str(_singleton(buffer_backend, "buffer_backend"))
diagnostics_mode = str(_singleton(diagnostics, "diagnostics")) diagnostics_mode = str(_singleton(diagnostics, "diagnostics"))
preserved_audio = _copy_audio(plan.get(_DRIVING_AUDIO_PLAN_KEY)) preserved_audio = _copy_audio(plan.get(_DRIVING_AUDIO_PLAN_KEY))
selected_audio = preserved_audio or _copy_audio(driving_audio) selected_audio = preserved_audio or _copy_audio(driving_audio)
selected_audio = slice_review_source_audio(selected_audio, plan)
selected_audio_source = None selected_audio_source = None
if selected_audio is not None: if selected_audio is not None:
selected_audio_source = ( selected_audio_source = (
+5 -1
View File
@@ -23,7 +23,7 @@ from ..v2.nodes import (
validate_sparse_prompt_overrides, validate_sparse_prompt_overrides,
) )
from .assembly import H3ContinuumAssembleSeamExperimental, H3ContinuumAssembleV3 from .assembly import H3ContinuumAssembleSeamExperimental, H3ContinuumAssembleV3
from .plan import make_assembly_plan from .plan import make_assembly_plan, make_review_audio_projection, REVIEW_AUDIO_PROJECTION_KEY
from .execution_planner import ( from .execution_planner import (
PROJECTION_CURRENT_REVIEW_UNIT, PROJECTION_CURRENT_REVIEW_UNIT,
PROJECTION_FULL_ACCEPTED_PREFIX, PROJECTION_FULL_ACCEPTED_PREFIX,
@@ -188,6 +188,9 @@ def _apply_review_decode_scope(
return outputs return outputs
start = int(decision.start_chunk) start = int(decision.start_chunk)
end = int(decision.end_chunk) end = int(decision.end_chunk)
audio_projection = make_review_audio_projection(
assembly_plan, start_chunk=start, end_chunk=end,
)
selected_entries = entries[start - 1 : end] selected_entries = entries[start - 1 : end]
sequence_complete = len(entries) == int(configured_chunks) sequence_complete = len(entries) == int(configured_chunks)
@@ -220,6 +223,7 @@ def _apply_review_decode_scope(
terminal_merged=selected_terminal_merged, terminal_merged=selected_terminal_merged,
terminal_initial_pair=(start == 1), terminal_initial_pair=(start == 1),
) )
scoped_plan[REVIEW_AUDIO_PROJECTION_KEY] = audio_projection
# Projection narrows the AV decode view, not the already-verified routing # Projection narrows the AV decode view, not the already-verified routing
# decision. Keep that Queue-local plan available to the public Inspector. # decision. Keep that Queue-local plan available to the public Inspector.
route_plan = assembly_plan.get("reference_routing_v1") route_plan = assembly_plan.get("reference_routing_v1")
+61 -4
View File
@@ -12,6 +12,7 @@ ASSEMBLY_PLAN_MAGIC = "H3_CONTINUUM_ASSEMBLY_PLAN"
ASSEMBLY_PLAN_SCHEMA_VERSION = 1 ASSEMBLY_PLAN_SCHEMA_VERSION = 1
FPS = 24 FPS = 24
SECOND_PASS_CONTRACT_VERSION = 1 SECOND_PASS_CONTRACT_VERSION = 1
REVIEW_AUDIO_PROJECTION_KEY = "review_audio_projection_v1"
class AssemblyPlanError(ValueError): class AssemblyPlanError(ValueError):
@@ -131,10 +132,66 @@ _make_assembly_plan_v300 = make_assembly_plan
_validate_assembly_plan_v300 = validate_assembly_plan _validate_assembly_plan_v300 = validate_assembly_plan
def validate_assembly_plan(plan): def validate_assembly_plan(plan):
result = _validate_assembly_plan_v300(plan) result = _validate_assembly_plan_v300(plan)
_validate_assembly_plan_contract(plan) _validate_assembly_plan_contract(plan)
return result validate_review_audio_projection(plan)
return result
def validate_review_audio_projection(plan):
"""Check a scoped PCM coordinate without recomputing its absolute origin."""
if REVIEW_AUDIO_PROJECTION_KEY not in plan:
return None
metadata = plan[REVIEW_AUDIO_PROJECTION_KEY]
if not isinstance(metadata, dict):
raise AssemblyPlanError("Review audio projection must be a dictionary")
if (type(metadata.get("version")) is not int or metadata.get("version") != 1 or metadata.get("kind") != "current_review_unit"
or metadata.get("coordinate_space") != "full_source_pcm"):
raise AssemblyPlanError("Review audio projection version or coordinate space is invalid")
for field in ("fps", "start_chunk", "end_chunk", "source_start_frame", "source_stop_frame"):
if type(metadata.get(field)) is not int:
raise AssemblyPlanError(f"Review audio projection {field} must be an integer")
if metadata["fps"] != FPS or int(plan.get("fps", 0)) != FPS:
raise AssemblyPlanError("Review audio projection FPS must be 24")
start, end = metadata["start_chunk"], metadata["end_chunk"]
first, stop = metadata["source_start_frame"], metadata["source_stop_frame"]
chunks = [int(chunk["chunk_index"]) for chunk in plan["chunks"]]
units = plan.get("decode_groups") or plan["chunks"]
if (start < 1 or end < start or len(chunks) != end - start + 1
or any(index != start + position for position, index in enumerate(chunks))
or first < 0 or stop <= first
or stop - first != sum(int(unit["net_frames"]) for unit in units)):
raise AssemblyPlanError("Review audio projection physical range is inconsistent")
return metadata
def make_review_audio_projection(plan, *, start_chunk: int, end_chunk: int):
"""Read natural retained ranges before narrowing the accepted decode view."""
validate_assembly_plan(plan)
if REVIEW_AUDIO_PROJECTION_KEY in plan:
raise AssemblyPlanError("Review output cannot be projected twice")
start, end = int(start_chunk), int(end_chunk)
selected = []
covered = []
for unit in plan.get("decode_groups") or plan["chunks"]:
indices = unit.get("logical_chunk_indices") or [unit["chunk_index"]]
if not any(start <= int(index) <= end for index in indices):
continue
if not all(start <= int(index) <= end for index in indices):
raise AssemblyPlanError("Review audio projection cannot split a physical group")
first, stop = unit.get("frame_start"), unit.get("frame_stop")
if type(first) is not int or type(stop) is not int or stop - first != int(unit["net_frames"]):
raise AssemblyPlanError("Review audio projection needs retained natural frame ranges")
if selected and selected[-1][1] != first:
raise AssemblyPlanError("Review audio projection ranges are not contiguous")
selected.append((first, stop))
covered.extend(int(index) for index in indices)
if not selected or covered != list(range(start, end + 1)):
raise AssemblyPlanError("Review audio projection is outside the accepted sequence")
return dict(version=1, kind="current_review_unit", coordinate_space="full_source_pcm",
fps=FPS, start_chunk=start, end_chunk=end,
source_start_frame=selected[0][0], source_stop_frame=selected[-1][1])
def make_assembly_plan(*args, **kwargs): def make_assembly_plan(*args, **kwargs):
+75 -2
View File
@@ -21,6 +21,7 @@ from ..guide_timeline import MARK_STILL_IMAGE_GUIDE
MAGIC = "H3_CONTINUUM_REFINE_CONTEXT" MAGIC = "H3_CONTINUUM_REFINE_CONTEXT"
VERSION = 1 VERSION = 1
OUTPUT_SCOPE_KEY = "output_scope_v1"
class RefineContextError(ValueError): class RefineContextError(ValueError):
@@ -251,6 +252,49 @@ def _group_field_tuple(group: Mapping[str, Any], name: str) -> tuple[int, ...]:
) from exc ) from exc
def project_refine_context(context: dict[str, Any], assembly_plan: Mapping[str, Any]):
"""Bind a V3.9 runtime-only view to complete physical output groups.
Sequence completion and saved history are unchanged. A partial capture may
serve a smaller decode view only after the ordinary complete-group checks
pass, including the frozen Reference route. Missing or mismatched partial
evidence retains the legacy diagnostic fallback, never guessed conditions.
"""
validate_refine_context(context)
contract = assembly_plan.get("second_pass_contract")
planned = contract.get("physical_groups") if isinstance(contract, Mapping) else None
if not isinstance(planned, list) or not planned:
return context
source = context["groups"]
keys = [tuple(group["logical_chunks"]) for group in source]
targets = [tuple(group.get("logical_chunks", ())) for group in planned]
if context["complete"] and keys == targets:
return validate_refine_context(context, assembly_plan=assembly_plan)
if len(set(keys)) != len(keys) or len(set(targets)) != len(targets):
return context
lookup = dict(zip(keys, source))
if any(key not in lookup for key in targets):
return context
selected = [lookup[key] for key in targets]
view = dict(context)
view["groups"] = tuple(
{**group, "group_id": index} for index, group in enumerate(selected)
)
view[OUTPUT_SCOPE_KEY] = MappingProxyType({
"version": 1,
"source_group_ids": tuple(group["group_id"] for group in selected),
})
try:
validate_refine_context(view, assembly_plan=assembly_plan)
except RefineContextError as exc:
# These groups were previously optional/incomplete. Do not add a new
# execution stop for unavailable output-scope evidence.
fallback = dict(context)
fallback["notes"] = (*context["notes"], f"Output scope unavailable: {exc}")
return fallback
return view
def validate_refine_context( def validate_refine_context(
context: Any, assembly_plan: Mapping[str, Any] | None = None context: Any, assembly_plan: Mapping[str, Any] | None = None
) -> dict[str, Any]: ) -> dict[str, Any]:
@@ -278,6 +322,14 @@ def validate_refine_context(
notes = context.get("notes") notes = context.get("notes")
if not isinstance(notes, tuple) or any(not isinstance(note, str) for note in notes): if not isinstance(notes, tuple) or any(not isinstance(note, str) for note in notes):
raise RefineContextError("refine context notes must be an immutable string tuple") raise RefineContextError("refine context notes must be an immutable string tuple")
scope = context.get(OUTPUT_SCOPE_KEY)
if scope is not None:
ids = scope.get("source_group_ids") if isinstance(scope, Mapping) else None
if (not isinstance(scope, Mapping) or scope.get("version") != 1
or not isinstance(ids, tuple) or len(ids) != len(groups)
or any(type(value) is not int or value < 0 for value in ids)
or tuple(sorted(set(ids))) != ids or not groups):
raise RefineContextError("refine context output scope is invalid")
previous_clip_index = 0 previous_clip_index = 0
for expected_id, group in enumerate(groups): for expected_id, group in enumerate(groups):
@@ -345,9 +397,9 @@ def validate_refine_context(
# Its local group ids describe capture order, not Assembly Plan position; # Its local group ids describe capture order, not Assembly Plan position;
# validating those partial entries by index would incorrectly turn the # validating those partial entries by index would incorrectly turn the
# intended prompt-only fallback into a broken typed-contract hard error. # intended prompt-only fallback into a broken typed-contract hard error.
if not context["complete"]: if not context["complete"] and scope is None:
return context return context
if context["complete"] and len(plan_groups) != len(groups): if len(plan_groups) != len(groups):
raise RefineContextError( raise RefineContextError(
"complete refine context physical-group count differs from assembly plan" "complete refine context physical-group count differs from assembly plan"
) )
@@ -365,6 +417,15 @@ def validate_refine_context(
view_groups[group_id] if isinstance(view_groups, list) view_groups[group_id] if isinstance(view_groups, list)
and group_id < len(view_groups) else None and group_id < len(view_groups) else None
) )
if scope is not None:
# Inspector retains the full run, including pending groups.
# A Chunk-2 view must not compare against Inspector group 1.
matches = [item for item in view_groups
if isinstance(item, Mapping)
and isinstance(item.get("descriptor"), Mapping)
and tuple(item["descriptor"].get("logical_chunks", ()))
== tuple(group["logical_chunks"])] if isinstance(view_groups, list) else []
expected_route = matches[0] if len(matches) == 1 else None
captured_route = group.get("reference_routing_v1") captured_route = group.get("reference_routing_v1")
if (not isinstance(expected_route, Mapping) if (not isinstance(expected_route, Mapping)
or not isinstance(captured_route, Mapping) or not isinstance(captured_route, Mapping)
@@ -536,6 +597,18 @@ def adapt_group_conditioning(
) )
keyframe = dict(frozen_keyframe) keyframe = dict(frozen_keyframe)
latent = keyframe.get("latent") latent = keyframe.get("latent")
audio_latent = keyframe.get("audio_latent")
if audio_latent is not None and (
not torch.is_tensor(audio_latent) or audio_latent.ndim != 4
):
raise RefineConditioningAdaptationError(
f"refine group {group_id} keyframe audio latent is invalid"
)
# Core keyframes may carry audio without a spatial video latent.
# Keep its frozen audio and timing; resize only video payloads.
if latent is None and audio_latent is not None:
adapted_keyframes.append(keyframe)
continue
if not torch.is_tensor(latent) or latent.ndim != 5: if not torch.is_tensor(latent) or latent.ndim != 5:
raise RefineConditioningAdaptationError( raise RefineConditioningAdaptationError(
f"refine group {group_id} keyframe latent is invalid" f"refine group {group_id} keyframe latent is invalid"
+5 -1
View File
@@ -16,6 +16,7 @@ from ..v2.sampling import latent_from_cpu
from .plan import SECOND_PASS_CONTRACT_VERSION from .plan import SECOND_PASS_CONTRACT_VERSION
from .refine_context import ( from .refine_context import (
MAGIC as REFINE_CONTEXT_MAGIC, MAGIC as REFINE_CONTEXT_MAGIC,
OUTPUT_SCOPE_KEY,
RefineConditioningAdaptationError, RefineConditioningAdaptationError,
RefineContextError, RefineContextError,
adapt_group_conditioning, adapt_group_conditioning,
@@ -283,13 +284,16 @@ def prepare_physical_refine_groups(
f"using legacy prompt-only conditioning ({exc})." f"using legacy prompt-only conditioning ({exc})."
) )
else: else:
if bool(validated_context.get("complete")): if (bool(validated_context.get("complete"))
or validated_context.get(OUTPUT_SCOPE_KEY) is not None):
captured_groups = validated_context["groups"] captured_groups = validated_context["groups"]
else: else:
warnings.append( warnings.append(
"WARNING: refine_context capture is incomplete; using legacy " "WARNING: refine_context capture is incomplete; using legacy "
"prompt-only conditioning." "prompt-only conditioning."
) )
warnings.extend(str(note) for note in validated_context.get("notes", ())
if str(note).startswith("Output scope unavailable:"))
group_models: list[Any] = [] group_models: list[Any] = []
group_conditioning: list[list[Any]] = [] group_conditioning: list[list[Any]] = []
+1 -1
View File
@@ -1,6 +1,6 @@
"""Package version and public schema identifiers.""" """Package version and public schema identifiers."""
PACKAGE_VERSION = "3.9.0" PACKAGE_VERSION = "3.9.1"
STATE_SCHEMA_VERSION = 1 STATE_SCHEMA_VERSION = 1
PLAN_SCHEMA_VERSION = 1 PLAN_SCHEMA_VERSION = 1
PUBLIC_API_VERSION = 3 PUBLIC_API_VERSION = 3
+23 -5
View File
@@ -1536,11 +1536,13 @@ function reviewReady(node) {
// Completion ends continuation, not the user's ability to inspect/retry a Take. // Completion ends continuation, not the user's ability to inspect/retry a Take.
// Keep the backend status authoritative; never rewrite complete to review_ready. // Keep the backend status authoritative; never rewrite complete to review_ready.
if (revision?.status === "complete") return true; if (revision?.status === "complete") return true;
if (revision?.status === "review_ready" && takeReadOnly(revision)) return true;
return revision?.status === "review_ready" && reviewHasUnit(node); return revision?.status === "review_ready" && reviewHasUnit(node);
} }
function reviewHasUnit(node) { function reviewHasUnit(node) {
const revision = canonicalStorageRevision(node.__h3ContinuumTakeProject); const revision = canonicalStorageRevision(node.__h3ContinuumTakeProject);
if (revision && takeReadOnly(revision)) return false;
const unit = revision?.review_unit; const unit = revision?.review_unit;
return [unit?.physical_group, unit?.start, unit?.end].every( return [unit?.physical_group, unit?.start, unit?.end].every(
(value) => Number.isInteger(Number(value)) && Number(value) > 0, (value) => Number.isInteger(Number(value)) && Number(value) > 0,
@@ -1569,6 +1571,9 @@ function reviewStatus(node) {
if (!project) return "Queue the workflow to create the first chunk."; if (!project) return "Queue the workflow to create the first chunk.";
const revision = canonicalStorageRevision(project); const revision = canonicalStorageRevision(project);
if (!revision) return "Queue the workflow to create the first chunk."; if (!revision) return "Queue the workflow to create the first chunk.";
if (takeReadOnly(revision)) {
return "Saved history is read-only for Sampling v6\nStart from Chunk 1 with current settings. Existing Takes are kept.";
}
if (revision.status === "complete") { if (revision.status === "complete") {
const unit = revision.review_unit; const unit = revision.review_unit;
const label = reviewHasUnit(node) const label = reviewHasUnit(node)
@@ -1607,6 +1612,8 @@ function reviewStatus(node) {
} }
function selectProductionReviewAction(node, action) { function selectProductionReviewAction(node, action) {
const revision = canonicalStorageRevision(node.__h3ContinuumTakeProject);
if (revision && takeReadOnly(revision)) return;
if (activeReviewPrompt(node) || node.__h3ContinuumOutputFailure) return; if (activeReviewPrompt(node) || node.__h3ContinuumOutputFailure) return;
const from = findWidget(node, REGENERATE_WIDGET)?.value; const from = findWidget(node, REGENERATE_WIDGET)?.value;
if (reviewSettingsChanged(node) || (from && from !== "Auto" && from !== 0)) { if (reviewSettingsChanged(node) || (from && from !== "Auto" && from !== 0)) {
@@ -1645,6 +1652,10 @@ function takeGroup(item) {
const takeCatalogCache = new WeakMap(); const takeCatalogCache = new WeakMap();
function takeReadOnly(item) {
return Number(item?.sampling_contract_version) !== 6 || item?.generation_reusable !== true;
}
function takeCatalog(node) { function takeCatalog(node) {
const project = node.__h3ContinuumTakeProject; const project = node.__h3ContinuumTakeProject;
if (project && takeCatalogCache.has(project)) return takeCatalogCache.get(project); if (project && takeCatalogCache.has(project)) return takeCatalogCache.get(project);
@@ -1673,7 +1684,7 @@ function takeCatalog(node) {
if (lineageAware && !canonicalLineage) return []; if (lineageAware && !canonicalLineage) return [];
const scoped = canonicalLineage const scoped = canonicalLineage
? normalized.filter( ? normalized.filter(
({ item }) => String(item?.lineage_sha256 || "") === canonicalLineage, ({ item }) => String(item?.lineage_sha256 || "") === canonicalLineage || takeReadOnly(item),
) )
: normalized; : normalized;
const scopedById = new Map(scoped.map( const scopedById = new Map(scoped.map(
@@ -1807,6 +1818,9 @@ function takeAncestry(project, selected, catalog) {
} }
function continueFromTakeSummary(project, selected, catalog) { function continueFromTakeSummary(project, selected, catalog) {
if (takeReadOnly(selected)) {
return "Read-only history: start a new Full Video or Review from Chunk 1; saved Takes remain available.";
}
const chain = takeAncestry(project, selected, catalog); const chain = takeAncestry(project, selected, catalog);
if (!chain.length) { if (!chain.length) {
return "Continue From Here: unavailable until provenance is validated"; return "Continue From Here: unavailable until provenance is validated";
@@ -1861,9 +1875,12 @@ function takeStatus(node) {
canonicalHead canonicalHead
? `Canonical head: ${takeLabel(canonicalHead)} | ${shortRevision(canonicalHead.revision_id)}` ? `Canonical head: ${takeLabel(canonicalHead)} | ${shortRevision(canonicalHead.revision_id)}`
: "Canonical head: unavailable", : "Canonical head: unavailable",
`Compatible history: ${catalog.length} Take${catalog.length === 1 ? "" : "s"}`, `Compatible history: ${catalog.filter(item => !takeReadOnly(item)).length} Takes`,
continueFromTakeSummary(project, selected, catalog), continueFromTakeSummary(project, selected, catalog),
]; ];
if (takeReadOnly(selected)) {
lines.push(`Sampling v${selected.sampling_contract_version ?? "unknown"}: read-only; cannot reuse in v6 generation.`);
}
if (hiddenCount) { if (hiddenCount) {
lines.push( lines.push(
`${hiddenCount} incompatible or incomplete Take${hiddenCount === 1 ? "" : "s"} hidden`, `${hiddenCount} incompatible or incomplete Take${hiddenCount === 1 ? "" : "s"} hidden`,
@@ -1894,7 +1911,7 @@ function selectTakeOffset(node, offset) {
function selectTakeAction(node, action) { function selectTakeAction(node, action) {
const selected = selectedTake(node); const selected = selectedTake(node);
if (!selected || reviewSettingsChanged(node) || activeReviewPrompt(node)) return; if (!selected || takeReadOnly(selected) || reviewSettingsChanged(node) || activeReviewPrompt(node)) return;
delete node.__h3ContinuumModeSetup; delete node.__h3ContinuumModeSetup;
setExistingWidgetValue( setExistingWidgetValue(
findWidget(node, TAKE_GROUP_WIDGET), findWidget(node, TAKE_GROUP_WIDGET),
@@ -2492,7 +2509,8 @@ function configureProductionReviewUx(node) {
const edited = reviewSettingsChanged(node); const edited = reviewSettingsChanged(node);
const outputFailure = node.__h3ContinuumOutputFailure?.runName === takeRunName(node); const outputFailure = node.__h3ContinuumOutputFailure?.runName === takeRunName(node);
const busy = Boolean(activeReviewPrompt(node)); const busy = Boolean(activeReviewPrompt(node));
const canReview = !edited && !busy && !outputFailure && (!from || from === "Auto" || from === 0); const canReview = !takeReadOnly(canonicalStorageRevision(node.__h3ContinuumTakeProject))
&& !edited && !busy && !outputFailure && (!from || from === "Auto" || from === 0);
const takeCount = takeCatalog(node).length; const takeCount = takeCatalog(node).length;
const actionNames = new Set([ const actionNames = new Set([
PRODUCTION_STATUS_WIDGET, PRODUCTION_STATUS_WIDGET,
@@ -2552,7 +2570,7 @@ function configureProductionReviewUx(node) {
} }
const catalog = takeCatalog(node); const catalog = takeCatalog(node);
const hasSelection = Boolean(selectedTake(node)); const hasSelection = Boolean(selectedTake(node));
const takeActionAvailable = hasSelection && !edited && !busy; const takeActionAvailable = hasSelection && !takeReadOnly(selectedTake(node)) && !edited && !busy;
const historyToggle = transientProductionWidgets(node).find( const historyToggle = transientProductionWidgets(node).find(
(widget) => widget.name === TAKE_TOGGLE_WIDGET, (widget) => widget.name === TAKE_TOGGLE_WIDGET,
); );