Publish V3.9 source with per-chunk references and retained V3.8X2 workflows

This commit is contained in:
ukr8b3g-cmyk
2026-09-29 01:53:22 +09:00
parent 55fc2d545f
commit 0582db62ef
67 changed files with 16505 additions and 1916 deletions
+157 -139
View File
@@ -1,14 +1,32 @@
# Changelog
## 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.
- Ship the user-selected V3.9 workflow JSON unchanged, a V3.9-only workflow ZIP, and a combined V3.8X2/V3.9 workflow ZIP. The ComfyUI Templates gallery exposes both separate workflows.
- Promote package metadata and startup version to `3.9.0`. Keep historical Releases/tags untouched. This main-source commit is not a new GitHub Release or Registry publication; browser save/reload and the unchanged default template's GPU acceptance remain separate gates.
## V3.9 development history — Reference Images in-node UI (2026-09-27)
- 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.
- 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.
## 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.
- 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.
- 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.
- 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.
- The planned separate V3.8X2 tag/Release and `release/3.8x2` maintenance branch have not been created. V3.9 workflow, commit, push, Release, and Registry publication remain pending.
## Experimental on main — Unified Timeline Video (2026-09-22)
- Renamed the existing V3.8 video IMAGE input to `Timeline Video Frames` while preserving its internal key and saved links.
- Added one appended optional `Video Reference Mode`: `Repeat Reference` remains the compatibility default; `Follow Timeline` selects the source interval belonging to each physical output group.
- Follow mode excludes continuation-prefix frames from source consumption, records each physical slice in the plan, uses slice-specific Video VAE/cache identity, and continues without video reference after source exhaustion.
- Kept Sampling, Seed, SIGMAS, Audio, masks, continuation grouping, Terminal Merge ownership, Run Storage schema, and the two official V3.8X2 workflows unchanged. Added separate Follow/Repeat comparison workflows.
- Synchronized the lightweight runtime verifier and README public-surface wording with the already-exported ten-node V3.8X2 surface, including `H3ContinuumVideoAdapter`.
- Acceptance: full CPU `1424 passed / 1 skipped / 0 failed`; browser save/reload PASS; Follow and Repeat `2 x 5 s` GPU runs PASS at 704x416/24fps with synchronized 10-second audio/video and no OOM, NaN, allocation failure, or crash.
- This remains Experimental: it is soft H3 conditioning, not exact motion copying, and it does not reduce the RAM needed to hold a long decoded IMAGE batch.
- Opt-in Follow Timeline / legacy-compatible Repeat Reference on the existing IMAGE input.
- Per-physical-group reference selection; source-end fallback; distinct interval VAE keys and local prompt caches.
- Local acceptance passed: full CPU `1424 passed / 1 skipped / 0 failed`, browser save/reload, and Follow/Repeat `2 × 5 s` GPU runs at 704×416/24fps with synchronized 10-second audio/video, mode separation, and no OOM/NaN/crash. This remains Experimental and is not a frame-perfect motion-copying or long-video RAM-optimization claim.
## Unreleased — Loader persistence repair (2026-09-21)
@@ -129,136 +147,136 @@
- Preserved all V3.4/V3.5 node IDs, backend socket keys, saved-workflow compatibility, Sampling, Conditioning, Terminal Merge, Assembly, Seam, and Run Storage contracts.
## 3.5.1
- Added `H3 Continuum Conditioning Bridge V3.5`, exposing one complete Core-compatible MODEL and CONDITIONING object per Continuum physical group without flattening CONDITIONING entries.
- Added optional standalone Reference Audio conditioning to Sampler V3.5. Its two Python-defined sockets remain permanently present; the frontend no longer adds/removes them or serializes a visibility-only widget, preserving workflow widget alignment on save/reload.
- Added `H3 Continuum Conditioning Bridge V3.5`, exposing one complete Core-compatible MODEL and CONDITIONING object per Continuum physical group without flattening CONDITIONING entries.
- Added optional standalone Reference Audio conditioning to Sampler V3.5. Its two Python-defined sockets remain permanently present; the frontend no longer adds/removes them or serializes a visibility-only widget, preserving workflow widget alignment on save/reload.
- Removed the experimental Last Queued Seed override and restored standard ComfyUI `control after generate` behavior for the V3.5 sampler.
- Renamed the displayed video-guide inputs to `Video Guide Frames` and `Video Guide Size` to distinguish a video loader's IMAGE frame batch from still Reference Images. Backend input keys and saved-workflow links are unchanged.
- GPU-accepted Conditioning Bridge generation, Reference Audio/Image retention, and the V3.5 Second Pass/Hi-Res paths while preserving physical-group, Terminal Merge, Run Storage, and audio contracts.
## 3.5.0
- Added `H3 Continuum Sampler V3.5` with a Continuum-aware refine-context output while preserving the V3.4 sampler path.
- Added `H3 Continuum Second Pass V3.5` as the advanced bridge for externally processed H3 video latents, including physical-group and Long Terminal Merge preservation.
- Added experimental `H3 Continuum Hi-Res Fix V3.5`, using a bounded pixel/VAE resize round trip before low-denoise Second Pass sampling.
- Added `H3 Continuum Latent Resize V3.5` as an advanced utility; direct high-ratio latent interpolation is not the recommended main Hi-Res Fix path.
- Added `H3 Continuum Assemble + Seam V3.5` with Auto, RAM, and Windows-safe disk-backed video buffers while keeping audio in RAM.
- Added the recommended V3.5 template workflow with optional Hi-Res Fix, Auto assembly, shared still-image megapixel sizing, and separate Video Guide Frames sizing.
- Preserved Exact Duration, seam behavior, physical decode-group ordering, first-pass audio passthrough, and external Core VAE Decode contracts.
- Kept V3.4 nodes and saved-workflow compatibility unchanged.
- Validated a 9.49 GiB final IMAGE with hash-identical RAM/Disk-backed output and approximately 9.50 GiB lower Windows private memory in Disk-backed mode.
- GPU-accepted the integrated 1x5 576-to-1152 Main path and Hybrid/Reference 1x5 paths; documented that the 3x5 2x terminal 77T group exceeds the tested RTX 5060 Ti 16 GiB.
## 3.4.0
- Completed the V3.4 runtime path from the public sampler through V3/V2 to `run_sequence()`.
- Added Driving Audio with absolute-time guide slices while preserving the original source for final output.
- Added persistent Video Guide Frames conditioning with Efficient 0.4 MP, Balanced 0.6 MP, and Match Output modes.
- Included Driving Audio and Video Guide Frames identities in Run Storage contracts.
- Changed malformed or empty prompts to warning-and-fallback behavior instead of stopping generation.
- Removed independent compatibility, model, and upstream-node restrictions that were stricter than ComfyUI Core.
- Added V3.4 runtime, driving-audio, and context diagnostics coverage.
## 3.3.0
- Promoted Timeline Video to a stable public node with chunk-local slicing and a 0.4 MP default.
- Added `H3 Continuum Assemble + Seam` with stable Audio Seam Auto and guarded Video Seam Auto defaults.
- Kept `Auto 2` as an experimental exposure-ramp extension without widening the visible option label.
- Added native Reference Audio conditioning and retained up to three ordered Reference Images.
- Preserved stable sampler Node IDs, Run Storage compatibility, Core VAE decode, and Spectrum Interop API v1 Actual Prefix 2 behavior.
- Added consolidated runtime validation results and documented known Timeline Video and Auto 2 limitations.
## 3.2.4
- Added native MiniMax H3 Reference Audio 1 conditioning with deferred Audio VAE encoding.
- Matched Core's permissive prompt handling by warning instead of stopping on unavailable Picture or Audio tags.
- Matched Core's dynamic Reference behavior by compacting active image inputs into Picture 1 through Picture N.
- Kept Sampling, Continuation, Spectrum Interop, external Core VAE Decode, and assembly semantics unchanged.
## 3.2.3
- Formalized the third ordered Reference Image as a compatibility-preserving extension.
- Kept V3.2.2 zero, one, and two-image Reference Contract JSON unchanged.
- Added Image 3 shape, dtype, exact SHA-256, position, and preprocess metadata only for three-image runs.
- Added Golden Contract and Revision identity regression coverage for V3.2.2 saved runs.
- Kept Sampling, Continuation, Spectrum Interop, Core VAE Decode, and assembly semantics unchanged.
## 3.2.2
- Allowed Ref2VA, FL2VA, and unverified checkpoints with Reference conditioning.
- Kept checkpoint classification as diagnostics without automatic MODEL switching.
- Kept Strict Compatibility for genuinely unsafe or unsupported H3 contracts.
## 3.2.1
- Declared V3.2.1 stable after GitHub Actions, Windows Ref2VA generation, Spectrum Actual Prefix 2, chunk integrity and complete Auto Resume passed.
- Added T2VA, First/Last Frame, FL2VA and persistent two-image Reference conditioning to the production sampler.
- Added crash-safe raw AV Run Storage, automatic resume, deterministic Revisions and partial regeneration.
- Combined ordered upstream graph fingerprints with runtime MODEL/CLIP/VAE weight probes for fail-closed reuse.
- Added per-chunk SHA-256 verification and rejected Regenerate From when Run Storage is disabled.
- Added official MiniMax H3 Sigma Shift graph-contract support.
- Updated CI dependencies, Windows runtime verification and package metadata for V3.2.1.
- Standardized project licensing as MIT.
- Kept sampling, Continuation, Spectrum Interop and external Core VAE decode semantics unchanged.
## 2.1.7
- Added the compact static `H3 Continuum Sampler` facade over the unchanged V2 execution core.
- Added Clip Overrides, Advanced, and Result pack helper nodes.
- Preserved `H3ContinuumSamplerV2` as a Legacy/Core workflow compatibility node.
- Removed the frontend display controller and all dynamic socket, proxy-widget, DOM, resize, and serialization workarounds.
- Kept State, Session, Prompt Plan, Sampling, Native Continuity, and Spectrum Interop contracts unchanged.
## 2.1.6
- Rebuilt the Sampler V2 frontend as one Vue-aware display controller.
- Removed the Individual Clip Overrides accordion; only active Clip Prompt sockets are shown.
- Kept connected out-of-range Clip Prompt sockets visible when chunks is reduced.
- Added one Advanced Settings control; connected advanced sockets stay visible while collapsed.
- Removed fixed node-height handling and all input/output add/remove UI folding.
- Kept backend socket/widget order, types, serialization, State/Session/Prompt Plan schemas, sampling, and Spectrum Interop unchanged.
- Exposed show_preview as a per-node basic control and removed the duplicate global preview setting.
## 2.0.1
- Fixed the bundled V2 example workflow widget serialization for `base_seed` with `control_after_generate`.
- Added an in-memory migration guard for the malformed 2.0.0 example where later widgets were shifted by one slot (`diagnostics` arrived as `0`).
- Preserved valid 2.0.0 workflows and all V1/V2 node identifiers and schemas.
- Added regression coverage for the legacy shifted-widget signature.
## 2.0.0
- Added production **H3 Continuum Sampler V2** for integrated N-chunk generation.
- Kept the V1 continuation algorithm and all V1 node identifiers unchanged.
- Added Fixed, List, and Timeline prompt planning plus a connectable Prompt Plan node.
- Cached each unique Qwen prompt conditioning before H3 sampling.
- Reused one externally accelerator-patched MODEL, sampler, and sigma schedule across all chunks.
- Deferred Video/Audio VAE decoding until every H3 chunk finished.
- Captured full raw AV chunk latents and next-state tails on CPU before decoding.
- Removed the second device-to-host state transfer by deriving next-state tails from committed CPU entries.
- Added conservative latent-motion Auto Context selection with 5/22/39-frame profiles.
- Added deterministic independent 64-bit per-chunk seeds and reroll nonces.
- Added V2 Session resume, accepted-prefix reuse, branch-safe reroll, and atomic safetensors/JSON persistence.
- Added exact cumulative audio sample-boundary assembly to prevent long-chain rounding drift.
- Added exact total-duration correction.
- Added single-allocation decoded IMAGE assembly and preflight RAM safety checks.
- Added Basic and Full diagnostics, including video overlap MAE/PSNR and audio correlation/boundary jump.
- Added optional V1 State entry point and V2 Session-to-V1 State conversion.
- Added V2 public API modules and regression/orchestration tests.
- Updated installer runtime verification for all ten registered nodes.
## 1.0.1
- Fixed a false strict-compatibility failure when Sol-Attn wraps MiniMax H3
`PackedLayout.__init__` with `*args` / `**kwargs` for Morton/span tracking.
- Strict checking still fails closed when the native H3 constructor genuinely
removes required positional or keyword capabilities.
- Added regression tests for the native signature, Sol-Attn forwarding wrapper,
and a genuinely incompatible constructor.
## 1.0.0
- Added **H3 Continuum Join** for initial and continuation clips.
- Added direct native video/audio latent state capture with 5/22/39-frame profiles.
- Represented continuation history as one native H3 `video` / `video_audio` reference block.
- Added Finish, Assemble, atomic State Save/Load, signed audio-grid offsets, in-place MM-RoPE adjustment, strict compatibility tests, and accelerator composition.
- Renamed the displayed video-guide inputs to `Video Guide Frames` and `Video Guide Size` to distinguish a video loader's IMAGE frame batch from still Reference Images. Backend input keys and saved-workflow links are unchanged.
- GPU-accepted Conditioning Bridge generation, Reference Audio/Image retention, and the V3.5 Second Pass/Hi-Res paths while preserving physical-group, Terminal Merge, Run Storage, and audio contracts.
## 3.5.0
- Added `H3 Continuum Sampler V3.5` with a Continuum-aware refine-context output while preserving the V3.4 sampler path.
- Added `H3 Continuum Second Pass V3.5` as the advanced bridge for externally processed H3 video latents, including physical-group and Long Terminal Merge preservation.
- Added experimental `H3 Continuum Hi-Res Fix V3.5`, using a bounded pixel/VAE resize round trip before low-denoise Second Pass sampling.
- Added `H3 Continuum Latent Resize V3.5` as an advanced utility; direct high-ratio latent interpolation is not the recommended main Hi-Res Fix path.
- Added `H3 Continuum Assemble + Seam V3.5` with Auto, RAM, and Windows-safe disk-backed video buffers while keeping audio in RAM.
- Added the recommended V3.5 template workflow with optional Hi-Res Fix, Auto assembly, shared still-image megapixel sizing, and separate Video Guide Frames sizing.
- Preserved Exact Duration, seam behavior, physical decode-group ordering, first-pass audio passthrough, and external Core VAE Decode contracts.
- Kept V3.4 nodes and saved-workflow compatibility unchanged.
- Validated a 9.49 GiB final IMAGE with hash-identical RAM/Disk-backed output and approximately 9.50 GiB lower Windows private memory in Disk-backed mode.
- GPU-accepted the integrated 1x5 576-to-1152 Main path and Hybrid/Reference 1x5 paths; documented that the 3x5 2x terminal 77T group exceeds the tested RTX 5060 Ti 16 GiB.
## 3.4.0
- Completed the V3.4 runtime path from the public sampler through V3/V2 to `run_sequence()`.
- Added Driving Audio with absolute-time guide slices while preserving the original source for final output.
- Added persistent Video Guide Frames conditioning with Efficient 0.4 MP, Balanced 0.6 MP, and Match Output modes.
- Included Driving Audio and Video Guide Frames identities in Run Storage contracts.
- Changed malformed or empty prompts to warning-and-fallback behavior instead of stopping generation.
- Removed independent compatibility, model, and upstream-node restrictions that were stricter than ComfyUI Core.
- Added V3.4 runtime, driving-audio, and context diagnostics coverage.
## 3.3.0
- Promoted Timeline Video to a stable public node with chunk-local slicing and a 0.4 MP default.
- Added `H3 Continuum Assemble + Seam` with stable Audio Seam Auto and guarded Video Seam Auto defaults.
- Kept `Auto 2` as an experimental exposure-ramp extension without widening the visible option label.
- Added native Reference Audio conditioning and retained up to three ordered Reference Images.
- Preserved stable sampler Node IDs, Run Storage compatibility, Core VAE decode, and Spectrum Interop API v1 Actual Prefix 2 behavior.
- Added consolidated runtime validation results and documented known Timeline Video and Auto 2 limitations.
## 3.2.4
- Added native MiniMax H3 Reference Audio 1 conditioning with deferred Audio VAE encoding.
- Matched Core's permissive prompt handling by warning instead of stopping on unavailable Picture or Audio tags.
- Matched Core's dynamic Reference behavior by compacting active image inputs into Picture 1 through Picture N.
- Kept Sampling, Continuation, Spectrum Interop, external Core VAE Decode, and assembly semantics unchanged.
## 3.2.3
- Formalized the third ordered Reference Image as a compatibility-preserving extension.
- Kept V3.2.2 zero, one, and two-image Reference Contract JSON unchanged.
- Added Image 3 shape, dtype, exact SHA-256, position, and preprocess metadata only for three-image runs.
- Added Golden Contract and Revision identity regression coverage for V3.2.2 saved runs.
- Kept Sampling, Continuation, Spectrum Interop, Core VAE Decode, and assembly semantics unchanged.
## 3.2.2
- Allowed Ref2VA, FL2VA, and unverified checkpoints with Reference conditioning.
- Kept checkpoint classification as diagnostics without automatic MODEL switching.
- Kept Strict Compatibility for genuinely unsafe or unsupported H3 contracts.
## 3.2.1
- Declared V3.2.1 stable after GitHub Actions, Windows Ref2VA generation, Spectrum Actual Prefix 2, chunk integrity and complete Auto Resume passed.
- Added T2VA, First/Last Frame, FL2VA and persistent two-image Reference conditioning to the production sampler.
- Added crash-safe raw AV Run Storage, automatic resume, deterministic Revisions and partial regeneration.
- Combined ordered upstream graph fingerprints with runtime MODEL/CLIP/VAE weight probes for fail-closed reuse.
- Added per-chunk SHA-256 verification and rejected Regenerate From when Run Storage is disabled.
- Added official MiniMax H3 Sigma Shift graph-contract support.
- Updated CI dependencies, Windows runtime verification and package metadata for V3.2.1.
- Standardized project licensing as MIT.
- Kept sampling, Continuation, Spectrum Interop and external Core VAE decode semantics unchanged.
## 2.1.7
- Added the compact static `H3 Continuum Sampler` facade over the unchanged V2 execution core.
- Added Clip Overrides, Advanced, and Result pack helper nodes.
- Preserved `H3ContinuumSamplerV2` as a Legacy/Core workflow compatibility node.
- Removed the frontend display controller and all dynamic socket, proxy-widget, DOM, resize, and serialization workarounds.
- Kept State, Session, Prompt Plan, Sampling, Native Continuity, and Spectrum Interop contracts unchanged.
## 2.1.6
- Rebuilt the Sampler V2 frontend as one Vue-aware display controller.
- Removed the Individual Clip Overrides accordion; only active Clip Prompt sockets are shown.
- Kept connected out-of-range Clip Prompt sockets visible when chunks is reduced.
- Added one Advanced Settings control; connected advanced sockets stay visible while collapsed.
- Removed fixed node-height handling and all input/output add/remove UI folding.
- Kept backend socket/widget order, types, serialization, State/Session/Prompt Plan schemas, sampling, and Spectrum Interop unchanged.
- Exposed show_preview as a per-node basic control and removed the duplicate global preview setting.
## 2.0.1
- Fixed the bundled V2 example workflow widget serialization for `base_seed` with `control_after_generate`.
- Added an in-memory migration guard for the malformed 2.0.0 example where later widgets were shifted by one slot (`diagnostics` arrived as `0`).
- Preserved valid 2.0.0 workflows and all V1/V2 node identifiers and schemas.
- Added regression coverage for the legacy shifted-widget signature.
## 2.0.0
- Added production **H3 Continuum Sampler V2** for integrated N-chunk generation.
- Kept the V1 continuation algorithm and all V1 node identifiers unchanged.
- Added Fixed, List, and Timeline prompt planning plus a connectable Prompt Plan node.
- Cached each unique Qwen prompt conditioning before H3 sampling.
- Reused one externally accelerator-patched MODEL, sampler, and sigma schedule across all chunks.
- Deferred Video/Audio VAE decoding until every H3 chunk finished.
- Captured full raw AV chunk latents and next-state tails on CPU before decoding.
- Removed the second device-to-host state transfer by deriving next-state tails from committed CPU entries.
- Added conservative latent-motion Auto Context selection with 5/22/39-frame profiles.
- Added deterministic independent 64-bit per-chunk seeds and reroll nonces.
- Added V2 Session resume, accepted-prefix reuse, branch-safe reroll, and atomic safetensors/JSON persistence.
- Added exact cumulative audio sample-boundary assembly to prevent long-chain rounding drift.
- Added exact total-duration correction.
- Added single-allocation decoded IMAGE assembly and preflight RAM safety checks.
- Added Basic and Full diagnostics, including video overlap MAE/PSNR and audio correlation/boundary jump.
- Added optional V1 State entry point and V2 Session-to-V1 State conversion.
- Added V2 public API modules and regression/orchestration tests.
- Updated installer runtime verification for all ten registered nodes.
## 1.0.1
- Fixed a false strict-compatibility failure when Sol-Attn wraps MiniMax H3
`PackedLayout.__init__` with `*args` / `**kwargs` for Morton/span tracking.
- Strict checking still fails closed when the native H3 constructor genuinely
removes required positional or keyword capabilities.
- Added regression tests for the native signature, Sol-Attn forwarding wrapper,
and a genuinely incompatible constructor.
## 1.0.0
- Added **H3 Continuum Join** for initial and continuation clips.
- Added direct native video/audio latent state capture with 5/22/39-frame profiles.
- Represented continuation history as one native H3 `video` / `video_audio` reference block.
- Added Finish, Assemble, atomic State Save/Load, signed audio-grid offsets, in-place MM-RoPE adjustment, strict compatibility tests, and accelerator composition.