Skip to content
Back to search

minimax-h3-colab

SkillWorks with Claude CodeGitHub

Create short MiniMax H3 first-frame or reference-to-video clips from local images through Google Colab CLI. Use when a user asks for image-guided H3 video generation or a sequential batch of clips.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/killkli--minimax-h3-colab-skill/badge.svg)](/listings/killkli--minimax-h3-colab-skill)

HTML

<a href="/listings/killkli--minimax-h3-colab-skill"><img src="/api/listings/killkli--minimax-h3-colab-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

git clone https://github.com/killkli/minimax-h3-colab-skill.git
cd minimax-h3-colab-skill
./install.sh
uv python install 3.12
uv tool install --python 3.12 google-colab-cli
colab --auth=oauth2 usage

The runner itself supports Python 3.11 and newer. The current Google Colab CLI release needs Python 3.12 or newer; `uv python install` supplies it without changing the Studio or system Python.

第一次執行 `usage` 時,Colab CLI 會引導 Google OAuth2 授權。完成授權後,可用下列命令啟動一次推論:

After OAuth2 is complete, run one inference with:

./run_colab_inference.sh \
  --image /absolute/path/reference.png \
  --prompt /absolute/path/prompt.txt \
  --output /absolute/path/result.mp4

可重複傳入 1–9 個 `--image`;順序會對應 prompt 中的 `<Picture 1>` 至 `<Picture 9>`。`prompt.txt` 會以 UTF-8 原文上傳並傳給 Notebook。

Repeat `--image` for 1–9 ordered references. Their order maps to `<Picture 1>` through `<Picture 9>`, and the UTF-8 prompt file is uploaded verbatim to the notebook.

The runner uses Colab compute units and remote GPU allocation. It does not run the H3 model on the local computer. See the language-specific README for prerequisites, batch manifests, session behavior, limits, and troubleshooting.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 57 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: minimax-h3-colab
  3. description: Create short MiniMax H3 first-frame or reference-to-video clips from local images through Google Colab CLI. Use when a user asks for image-guided H3 video generation or a sequential batch of clips.
  4. ---
  5. # MiniMax H3 on Colab
  6. Use the bundled notebook and `scripts/runner.py` to generate MiniMax H3 video from local images. Jobs default to Ref2VA; choose `first_frame` for FL2VA first-frame conditioning. The Colab CLI must be installed and authenticated with an available compute-unit balance and GPU allocation.
  7. ## Workflow
  8. 1. Inspect the requested image(s) and any supplied prompt file. `reference` accepts 1–9 ordered images mapped to `<Picture 1>` through `<Picture 9>`. `first_frame` requires exactly one image and uses FL2VA first-frame conditioning.
  9. 2. If the user supplied a prompt file, pass it through unchanged. Otherwise, for `reference`, write a complete Ref2VA prompt with `subject_definitions`, `summary`, `retention_analysis`, `detailed_description` shot blocks, `overall_soundscape`, and `non_diegetic_music`; include dialogue in its shot with `<d>[Chinese] ...</d>` when requested. Use the bundled runner's `compose_ref2va_prompt` helper for structured shot input. For `first_frame`, describe motion and shots from the supplied opening image using FL2VA prompt conventions.
  10. 3. Check Colab access and balance with `python3 scripts/runner.py usage --json`. Sufficient compute units do not guarantee that Colab will allocate the requested GPU or high-memory runtime. Report allocation failures plainly.
  11. 4. Put the jobs in a UTF-8 JSON manifest and run one batch. The runner uploads each job's images and prompt, executes the bundled notebook sequentially on the same live session, downloads each MP4, and stops the session when it created it. If reusing a session, pass `--stop-on-complete` when the user wants it stopped after the batch.
  12. 5. Confirm each output exists and report its local path. Preserve completed clips if a later job fails.
  13. Example manifest:
  14. ```json
  15. {
  16. "jobs": [
  17. {
  18. "title": "intro",
  19. "inference_mode": "reference",
  20. "reference_images": ["/absolute/path/girl.jpg"],
  21. "prompt_file": "/absolute/path/prompt.txt",
  22. "width": 1344,
  23. "height": 768,
  24. "duration_seconds": 12,
  25. "output_name": "intro"
  26. }
  27. ]
  28. }
  29. ```
  30. Omit `inference_mode` to keep the compatible `reference` default. Width and height default to 1344×768; each must be a positive multiple of 32 and their area must not exceed 1344×768 pixels. A first-frame job must have exactly one reference image.
  31. Run from this skill directory:
  32. ```bash
  33. python3 scripts/runner.py batch --manifest /absolute/path/jobs.json --gpu A100 --timeout 10800 --output-dir /absolute/path/outputs
  34. ```
  35. For a single FL2VA clip, pass `--mode first_frame` and one `--image` to `scripts/runner.py single`; optionally set `--width` and `--height`. The shell launcher accepts the same options.
  36. Use `--no-high-mem` when high-memory allocation is unavailable or not desired. For a named existing session, add `--session SESSION --stop-on-complete` if the requested workflow should end that session.
  37. On Windows hosts, run the Colab CLI and this runner inside WSL, where the CLI is installed and authenticated.
  38. ## Operational limits
  39. - Each job requires 1–9 non-empty images, a non-empty UTF-8 prompt, and a duration from 4–15 seconds. `first_frame` requires exactly one image.
  40. - Keep all jobs in one batch to reuse the same Colab session and loaded ComfyUI/model state.
  41. - Do not blindly retry a timed-out `colab exec`: the remote kernel may still be working. The runner stops the session during cleanup and records completed jobs before reporting the failure.
  42. - Do not expose OAuth credentials or runtime tokens in prompts, manifests, browser state, or logs.
  43. - The bundled `assets/MiniMax_H3_Turbo_Colab.ipynb` is the inference notebook used by the runner.
README.md · README · 42 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # MiniMax H3 Colab Skill
  2. This repository is a standalone Codex skill for creating short MiniMax H3 Ref2VA videos from local reference images through Google Colab. It includes the skill instructions, the inference notebook, a Python runner, a shell launcher, an idempotent installer, and runner tests.
  3. 本儲存庫是可獨立使用的 Codex 技能,用於透過 Google Colab 將本機參照圖片製作成短篇 MiniMax H3 Ref2VA 影片。內容包含技能指示、推論 Notebook、Python runner、Shell 啟動器、可重複執行的安裝程式,以及 runner 測試。
  4. ## Documentation
  5. - [繁體中文完整說明](README.zh-TW.md)
  6. - [Full English documentation](README.en.md)
  7. ## Quick start / 快速開始
  8. ```bash
  9. git clone https://github.com/killkli/minimax-h3-colab-skill.git
  10. cd minimax-h3-colab-skill
  11. ./install.sh
  12. uv python install 3.12
  13. uv tool install --python 3.12 google-colab-cli
  14. colab --auth=oauth2 usage
  15. ```
  16. The runner itself supports Python 3.11 and newer. The current Google Colab CLI
  17. release needs Python 3.12 or newer; `uv python install` supplies it without
  18. changing the Studio or system Python.
  19. 第一次執行 `usage` 時,Colab CLI 會引導 Google OAuth2 授權。完成授權後,可用下列命令啟動一次推論:
  20. After OAuth2 is complete, run one inference with:
  21. ```bash
  22. ./run_colab_inference.sh \
  23. --image /absolute/path/reference.png \
  24. --prompt /absolute/path/prompt.txt \
  25. --output /absolute/path/result.mp4
  26. ```
  27. 可重複傳入 1–9 個 `--image`;順序會對應 prompt 中的 `<Picture 1>` 至 `<Picture 9>`。`prompt.txt` 會以 UTF-8 原文上傳並傳給 Notebook。
  28. Repeat `--image` for 1–9 ordered references. Their order maps to `<Picture 1>` through `<Picture 9>`, and the UTF-8 prompt file is uploaded verbatim to the notebook.
  29. The runner uses Colab compute units and remote GPU allocation. It does not run the H3 model on the local computer. See the language-specific README for prerequisites, batch manifests, session behavior, limits, and troubleshooting.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.