Skip to content

💃 Free CC0 mannequin pose reference library for AI image tools (Nano Banana, ChatGPT, ComfyUI, Midjourney) — 250+ poses × 5 angles in line art, names in ja/zh/en/ko

4.4 KB
Raw Blame

Contributing to Shisei

Thanks for helping. Shisei is a pose reference library, so most changes are judged by one question: does the catalog still describe exactly the images in the repository, in all four languages? This file covers how to propose a change and what "done" means here. Coding agents (Claude Code, Codex, …) also follow AGENTS.md.

Before you start

  • Typos, translations, gallery bugs: open a pull request directly, or an issue if you are not sure.
  • New poses or a new series: open an issue first (there is a "Pose request" template) so we can agree on the pose, its category and its four names. IDs are stable and never reused.
  • Security issues: do not open a public issue; see SECURITY.md.

The one hard rule: no outside images

Every image in this repository is made for the project. Do not add, trace, re-encode or derive images from other pose sites or collections (including ai-pose-gallery and its mirrors), even if their front-end code is open source: their terms forbid redistributing the images. A pull request that adds an image must say how it was made.

Figures are mannequins or AI-generated adults. Nothing that reads as a minor, a real person or explicit content is accepted.

Setup

Requirements: Python 3.9+ for the tools, cwebp 1.6+ (brew install webp) only to rebuild image tiers. The gallery has no runtime dependencies and no build step beyond build_site.py.

python3 -I tools/verify.py                                   # catalog ↔ files, known gaps, orphans
python3 -I tools/build_site.py --link-images                 # writes site/ with symlinked images
python3 -m http.server 8000 -d site                          # http://localhost:8000
python3 tools/build_images.py std=~/path/to/std-source-set …  # (re)build w300 + w600 + w900 for views that passed QC

What a pull request needs

  1. tools/verify.py passes. Never weaken it to make it pass. A view that could not be made is null in the catalog, listed in gaps (missing), and explained in docs/GENERATION.md — never filled with a mirrored, flipped or low-quality drawing.
  2. Catalog entries are complete: class_* and name_* in ja / zh / en / ko and an English desc_en. Japanese is the original naming; change all four names together and keep them short (動作・手の位置).
  3. Images: only the w300, w600 and w900 tiers are committed, built by tools/build_images.py (metadata stripped), and only for views with a QC pass in data/qc.json (docs/qc/). Full-size originals ship in the Release zip. Image rebuilds go in their own commit.
  4. Web changes follow docs/DESIGN.md, include a screenshot, work at 390×844 without horizontal scroll, and add every user-visible string to web/i18n.js in all four languages. Check with node --check web/app.js web/i18n.js.
  5. A CHANGELOG line under [Unreleased] for every user-visible change.
  6. Code style: comments in English, explaining why. No runtime dependencies in the gallery.

Commit messages

Use Conventional Commits in English:

feat(data): add 6 squatting poses

Explain why the change is needed and how it was verified.

Scopes: data, images, web, tools, docs, ci. Keep the subject under 72 characters, one logical change per commit. Pull requests are squash-merged, so the PR title becomes the commit subject.

AI-assisted contributions

AI-assisted work is welcome; much of this repository was made with coding agents and image models. You are responsible for everything you submit: look at every image, run the checks above, and make sure it follows the rules on this page. Credit the tool with a Co-Authored-By: trailer if you like.

Licence

Code and documentation are licensed under the Apache License 2.0; images and the project's own data are dedicated to the public domain under CC0 1.0. By submitting a contribution you agree that code and docs are licensed under Apache-2.0 (§5) and that images and data are released under CC0; there is no separate CLA. Make sure you have the right to submit it. The pose list follows ai-pose-gallery's catalog and is credited, not licensed (see NOTICE).

Code of conduct

Everyone taking part follows the Code of Conduct.