280 mannequin poses, five angles each, and the license is just 'yes'
A CC0 pose reference library built for AI image workflows — every view human-reviewed, with depth and normal maps for ControlNet.

What it does Shisei is a pose reference library of 280 faceless mannequin poses across four series (standard, upper body, sexy, duo), each drawn from five camera angles — front, three-quarter, side, above, below. You feed a view into Nano Banana, ChatGPT, ComfyUI or Midjourney as a pose or composition reference; using several angles of the same pose keeps the figure consistent across shots. Pose names and categories come in Japanese, Chinese, English and Korean.
The interesting bit
Every single view went through human QC — failures (nipple marks, stray faces, broken anatomy, wrong camera angle, figures that read as minors) are removed from the catalog with the reason recorded in data/qc.json. Views that were never generated stay empty rather than being filled with a bad drawing, which is a rarer discipline than it sounds. There are also 944 clay, depth and normal maps pixel-aligned with the line art, ready for ControlNet Depth and Normal.
Key highlights
- 1,212 line-art views passed QC, plus 89 photo renders and 944 clay/depth/normal maps
- Images are CC0 — commercial use, no attribution, and no conditions on what you generate
- Zero-dependency static gallery with lightbox, angle filters, four UI languages and a dark theme
- Catalog (
catalog.json/catalog.csv) is the source of truth; stable IDs that never change - Downloads are tiered zips with
.sha256checksums; the line-art pack is 36.5 MB
Caveats
- The pose list itself (selection, numbering, names) comes from ai-pose-gallery’s catalog and is credited, not licensed — neither license covers it
- Two poses (P038, P052) currently show a different pose than their name and are hidden
- The sexy and duo series are suggestive adult references, blurred by default in the gallery
Verdict Anyone generating figures with AI — or drawing them the old-fashioned way — gets a genuinely free, well-QC’d reference set. If you need photographic realism or explicit content, look elsewhere (the latter is explicitly off-limits).
Frequently asked
- What is wquguru/shisei?
- A CC0 pose reference library built for AI image workflows — every view human-reviewed, with depth and normal maps for ControlNet.
- Is shisei open source?
- Yes — wquguru/shisei is open source, released under the Apache-2.0 license.
- What language is shisei written in?
- wquguru/shisei is primarily written in Python.
- How popular is shisei?
- wquguru/shisei has 502 stars on GitHub.
- Where can I find shisei?
- wquguru/shisei is on GitHub at https://github.com/wquguru/shisei.