docs: publish V3.9 Reference Images prompt skill

This commit is contained in:
ukr8b3g-cmyk
2026-09-29 20:15:09 +09:00
parent 2b5774f28e
commit b593991f7b
8 changed files with 111 additions and 4 deletions
+6 -2
View File
@@ -122,8 +122,8 @@ ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDA
b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py
667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml 667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml
94fdc2d545b2d4077a4d70229fd22ba3ccac21d49d1097d2873f76d04e9a3c7a pytest.ini 94fdc2d545b2d4077a4d70229fd22ba3ccac21d49d1097d2873f76d04e9a3c7a pytest.ini
5d1302baa84564055c60020f158b0a6ff70cd255b6b4678be7fa03f52bbbcdd3 README_JA.md 53199468f1b5e0e37a70c05a75760af8ec79173be9caec768f3cf7dd554308fd README_JA.md
7fe3bf8d946a4ee75452757212701b4979fbc77656823b8901d0ed6f4120a1d1 README.md 9af0e573be17c93362d2ee0d32953d4fff9c55b767f61ba97ba786960aec62ae README.md
44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py 44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py
5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py 5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py
caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py
@@ -408,3 +408,7 @@ f69b734b5bf19fb6db78bbe4737db0bef69c2a6cef2710f28767882ef35ec847 examples/workf
37107bbefa1a8c92abfc286b873efddd9a5feda2d0455e97149cc35a611cabfe tests/test_v39_official_workflow.py 37107bbefa1a8c92abfc286b873efddd9a5feda2d0455e97149cc35a611cabfe tests/test_v39_official_workflow.py
40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/MiniMax_H3_Continuum_V39.json 40199d2500174ba51c040d6983f88d74c4bfbaf19191e728818e6b969b869b0f examples/MiniMax_H3_Continuum_V39.json
1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/MiniMax_H3_Continuum_V38X2.json 1b2212b7511fac0b63dfa6e9f006cd8644c72d08c6945ab0e4ed0e722c97a730 examples/MiniMax_H3_Continuum_V38X2.json
2bcf455b4ece8b1861e327f6f9f8a79170e7c77470155aacada52e18438543dd write-h3-v39-reference-prompts.zip
3dfd0e7042070f62b9e78239c0e75d82a7df32098db10832465f3049b7f0b38f write-h3-v39-reference-prompts/SKILL.md
a9f7b67bd65a9beda622aa5fdc5f8a2ecae1b58b18690d91fb0107a16a3e384e write-h3-v39-reference-prompts/references/simple-3x5.txt
81d7b4f4da797339b46911bc96d0a3736d70df29e092715c6684bb1786908593 write-h3-v39-reference-prompts/references/structured.md
+1
View File
@@ -176,6 +176,7 @@ Download the current V3.8X2 workflow: [JSON](examples/workflows/MiniMax_H3_Conti
## Prompt and skill downloads ## Prompt and skill downloads
- [V3.9 Reference Images prompt skill](write-h3-v39-reference-prompts/SKILL.md) | [ZIP](write-h3-v39-reference-prompts.zip) — for fixed `@R1`–`@R9` image slots and per-chunk assignments. Use it with the V3.9 Reference Images node; it is separate from the general Continuum prompt skill below.
- [LLM system prompt ZIP](H3-Continuum-LLM-System-Prompt-v1.zip) — system instructions and reference material for Continuum prompt authoring. - [LLM system prompt ZIP](H3-Continuum-LLM-System-Prompt-v1.zip) — system instructions and reference material for Continuum prompt authoring.
- [Continuum prompt skill ZIP](H3-Continuum-Skill-v1.zip) — general chunk-aware prompt authoring for Codex and other compatible agents. - [Continuum prompt skill ZIP](H3-Continuum-Skill-v1.zip) — general chunk-aware prompt authoring for Codex and other compatible agents.
- [Continuum Dance Director skill ZIP](minimax-h3-continuum-dance-director.zip) — long-form choreography prompts for one primary dancer, with body and camera continuity across chunks. - [Continuum Dance Director skill ZIP](minimax-h3-continuum-dance-director.zip) — long-form choreography prompts for one primary dancer, with body and camera continuity across chunks.
+1
View File
@@ -169,6 +169,7 @@ V3.8X2 Workflow:[JSON](examples/workflows/MiniMax_H3_Continuum_V38X2.json)/[
## プロンプト・スキルのダウンロード ## プロンプト・スキルのダウンロード
- [V3.9 Reference Images用プロンプトスキル本文](write-h3-v39-reference-prompts/SKILL.md)/[ZIP](write-h3-v39-reference-prompts.zip):固定の`@R1`~`@R9`とチャンク別の画像割り当てに対応します。V3.9のReference Imagesノード用で、下記の汎用Continuumスキルとは別です。
- [LLM用システムプロンプトZIP](H3-Continuum-LLM-System-Prompt-v1.zip):Continuum向けプロンプト作成のシステム指示と詳細資料。 - [LLM用システムプロンプトZIP](H3-Continuum-LLM-System-Prompt-v1.zip):Continuum向けプロンプト作成のシステム指示と詳細資料。
- [Continuum専用プロンプトスキルZIP](H3-Continuum-Skill-v1.zip):Codexなどで使う、チャンク構成に対応した汎用プロンプト作成スキル。 - [Continuum専用プロンプトスキルZIP](H3-Continuum-Skill-v1.zip):Codexなどで使う、チャンク構成に対応した汎用プロンプト作成スキル。
- [Continuum Dance DirectorスキルZIP](minimax-h3-continuum-dance-director.zip):主役のダンサー1人を対象に、チャンク間の身体動作とカメラの連続性を考慮した長尺ダンス用プロンプトを作成するスキル。 - [Continuum Dance DirectorスキルZIP](minimax-h3-continuum-dance-director.zip):主役のダンサー1人を対象に、チャンク間の身体動作とカメラの連続性を考慮した長尺ダンス用プロンプトを作成するスキル。
+2 -2
View File
@@ -35,8 +35,8 @@ ecb65a5dbf9e1683c5a3cadd82daaea3a90a2b776f6ebabfbe8b07aa93177422 nodes.py
ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt ea64b8e0aa0dda828047c9c12818be8554429774049b9e8c0482c84b9c4f4071 PACKAGE_VALIDATION.txt
b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py b470819f16fc179654dc1600721a63f0a5a14e777f4c004d574b499da6886ba5 public_api.py
667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml 667f2646e82362d472d91a7cf5dc9a86c95f6637f3bc8c672dd516800b1a598c pyproject.toml
5d1302baa84564055c60020f158b0a6ff70cd255b6b4678be7fa03f52bbbcdd3 README_JA.md 53199468f1b5e0e37a70c05a75760af8ec79173be9caec768f3cf7dd554308fd README_JA.md
7fe3bf8d946a4ee75452757212701b4979fbc77656823b8901d0ed6f4120a1d1 README.md 9af0e573be17c93362d2ee0d32953d4fff9c55b767f61ba97ba786960aec62ae README.md
44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py 44127adcfa30ee84bff9cd4b2d0c695355209550e76982d374122008728aa5bc reference_audio.py
5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py 5cb9bfd6810d3a93a637451980a4cc5c1436e5b84cfd1c7a111cd192e876351b reference_video.py
caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py caa572ea65dafc86850db99f52433d360ac9c4ab36b854292a3dccd5ba32dc76 reference.py
Binary file not shown.
+51
View File
@@ -0,0 +1,51 @@
---
name: write-h3-v39-reference-prompts
description: Write Sequence Prompts for H3 Continuum V3.9 Reference Images, using fixed image slots and per-chunk assignments. Supports concise hand-written prompts and structured LLM prompts. Does not generate videos or modify workflows.
---
# H3 V3.9 reference prompts
An independent Continuum adaptation, not an official MiniMax skill. Preserve the installed `write-continuum-h3-prompts` skill; this package is separate and self-contained.
## Establish the small input contract
Use known workflow settings and images. Establish only what affects the result:
- Chunks, seconds per chunk, Reference Use, and each connected slot's selected chunks.
- What each image supplies: character identity, clothing, object, environment, or style; what to preserve and what may change.
- Intended action, camera, and start/end state of each chunk.
- Whether First/Last Image, Timeline Video or audio inputs are actually enabled.
Prefer a simple list such as `R1: red-coated woman, chunks 1/2/3; R4: navy-top woman, chunks 1/2; R9: teal-coated woman, chunks 2/3`. Do not require YAML or a large questionnaire. Ask only when missing information materially changes a reference binding or the requested story. Never infer slot numbers from attachment order or invent details of unseen images.
## Reference and routing contracts
- `@R1` to `@R9` identify the fixed sockets on **H3 Continuum Reference Images V3.9**. Current V3.9 All chunks and Per chunk both resolve active tags before encoding. This is not V3.8 syntax compatibility.
- Write `the woman in @R1`, not `@R1 walks`. The tag denotes a source image, not a person.
- `<Subject N>` denotes a continuing entity. Keep its meaning stable throughout the video; it is independent of the R number. Multiple images may describe one Subject; an image may supply an object rather than a person.
- Do not manually renumber reference assets as `<Picture N>`. Their effective Picture numbers vary with selected references and First/Last inputs. If explicitly describing First/Last frame labels, require the actual presentation mapping; do not guess.
- Cite an image only where it is connected and selected. An inactive tag is currently left unchanged with a runtime warning, not repaired automatically. Identify this before handing over text; do not add runtime execution restrictions.
- Reference selection is not a visibility schedule. Turning a reference off does not remove a carried character. Describe exits, entrances and inherited states explicitly. Not every reference needs to appear as a separate object/person.
- Use `Prompt Format = Timeline` for staged chunks. Put every `[Chunk N]` on its own line, covering each requested chunk exactly once. Put text on the next line. Do not put chunk-specific references in a shared preamble before the first header.
- A chunk is not a camera cut. Shot markers belong inside the selected chunk text. Some workflows merge logical chunks into physical groups; check the effective Reference Plan if First/Last or terminal merging is involved. Never promise independent sampling or silently change assignments to bypass a conflict.
## Choose the requested writing style
**Simple / hand-written:** Default to one short paragraph per chunk: image role + observable action + camera + end state. The next paragraph inherits that state. Avoid mandatory Subject numbering or six headings for simple requests. Use explicit source-to-target actions when the reference pose differs. Add short soundscape/music lines only when audio intent matters.
**Structured / LLM:** When requested or useful for multi-entity/attribute-transfer work, use these fields in order inside each chunk:
`subject_definitions`, `summary`, `retention_analysis`, `detailed_description`, `overall_soundscape`, `non_diegetic_music`.
Read [the structured contract and example](references/structured.md) before authoring this form.
Do not equate longer prose with greater control. Preserve requested detail but avoid contradictory actions, redundant appearance intensifiers and too many events for the interval. For continuation, describe the actual intended handoff (position, motion, camera); `Continuation of Chunk N.` is a useful cue, not a sampler command or quality guarantee.
Preserve supplied dialogue, lyrics and visible text in their original language. Do not invent an Audio/Video reference. Driving Audio, reference audio and generated soundscape are different paths; prompt text cannot connect them. For voice references, request the actual audio-label mapping and assign each line to a speaker. An image-only request does not authorize adding dialogue/music.
## Return and check
Return a short settings/assignment note outside a single copy-ready Sequence Prompt block. If prompt-only was requested, omit the note. Keep citations, analysis, audit results and YAML outside the prompt block. English scene descriptions are the default, not a forced translation of supplied speech/text.
Before returning: verify full chunk coverage, active reference tags, stable Subject meanings, attribute ownership, plausible action density, entrance/exit timing intent, and boundary-state consistency. Do not claim observed output states when no video was inspected. Do not guarantee exact timestamps, identity, seamlessness or that this wording is empirically optimal.
Use [the 3 × 5-second simple example](references/simple-3x5.txt) for syntax, not as a fixed duration or mandatory story. The reference assignments for that example are R1=1/2/3, R4=1/2, R9=2/3; all other slots off, First/Last/Video/Audio references disabled.
No skill installation, generation, downloads, workflow edits, node changes or publication are authorized by prompt authoring alone.
@@ -0,0 +1,14 @@
[Chunk 1]
An eye-level medium-wide camera tracks alongside the red-coated woman in @R1 and the navy-top woman in @R4 as they walk from screen left to screen right through a bright covered walkway. Preserve their distinct faces, hairstyles and clothes; adapt their poses to walking. Both remain clearly visible and keep walking as the chunk ends, with open space on the right.
overall_soundscape: Footsteps and quiet walkway ambience. No dialogue.
non_diegetic_music: N/A
[Chunk 2]
Continuation of Chunk 1. Keep the same tracking camera, lighting and walking direction. As the chunk begins, the teal-coated woman in @R9 enters from screen right and joins the red-coated woman in @R1 and the navy-top woman in @R4. Preserve all three distinct faces, hairstyles and clothes. Show all three together through the middle of the chunk. Then the navy-top woman turns into a side passage and walks out of view before the chunk ends. The red-coated and teal-coated women keep walking together.
overall_soundscape: Continuous footsteps and quiet walkway ambience. No dialogue.
non_diegetic_music: N/A
[Chunk 3]
Continuation of Chunk 2. The red-coated woman in @R1 and the teal-coated woman in @R9 continue walking side by side in the same direction. Preserve their distinct faces, hairstyles and clothes, and keep the same eye-level medium-wide tracking camera and lighting. Both remain clearly visible through the end. The side passage recedes behind them as their steady walk continues.
overall_soundscape: Continuous footsteps and quiet walkway ambience. No dialogue.
non_diegetic_music: N/A
@@ -0,0 +1,36 @@
# Structured reference format
Use only when requested or useful. Each chunk contains the following English fields:
```text
[Chunk 1]
subject_definitions:
<Subject 1> is the red-coated woman in @R1.
<Subject 2> is the navy-top woman in @R4.
summary:
[reference generation] Two women walk together through a covered walkway.
retention_analysis:
<Subject 1>: partially_preserved — preserve face, hairstyle and clothing; adapt pose and gaze to walking.
<Subject 2>: partially_preserved — preserve face, hairstyle and clothing; adapt pose and gaze to walking.
detailed_description:
[Shot 1] An eye-level medium-wide camera tracks alongside both women as they walk from screen left to screen right. Both remain clearly visible. They continue walking as the chunk ends, with open space on the right for a third person to join next.
overall_soundscape:
Footsteps and quiet walkway ambience. No dialogue.
non_diegetic_music:
N/A
```
Keep Subject 1/2 meanings in later chunks. A new person from R9 can be Subject 3; do not rename her Subject 2 because she becomes Picture 2. Re-establish only the needed source binding in each selected prompt. For a continuing person without an active reference, describe inherited identity/state without an inactive @R tag.
Use retention markers with a concrete scope: partially_preserved for preserved identity/clothing with changed motion/background; attribute_transfer for applying clothing or a prop from another source; fully_preserved only when the requested content actually warrants it. These are semantic instructions, not hard masks or guaranteed constraints.
For a bag reference: `<Subject 4> is the tan bag in @R2, worn by <Subject 1>.` In the action text keep this ownership. If the character's source image already contains another bag, explicitly replace that source accessory with the selected bag, while preserving the person's identity and clothes.
In Chunk 2+, start detailed_description with `Continuation of Chunk 1.` (use the actual preceding number) plus the inherited physical state. Keep the six-field order; the continuation sentence is not a parser requirement. Use new shot markers/times only for intentional cuts. Do not invent `<Video 1>` for preceding generated chunks.
Official MiniMax Reference guide uses these six fields and recommends substantial scene detail (normally 350–500 English words in detailed_description for generation). Compact per-chunk wording here is a Continuum adaptation to test, not an officially proven better length. A short simple prompt remains valid input to Continuum.
Sources checked for the design:
- https://github.com/MiniMax-AI/MiniMax-H3/blob/main/skills/h3-prompt-writing/references/ref-en.txt
- https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md
- Local V3.9: v3/reference_images_v39.py, v3/driving_nodes.py, v3/reference_runtime.py, v3/reference_effective_plan.py, v2/prompts.py (2026-09-28). V3.9 is a local pre-release implementation, not a promise about every public package named 3.9.