wquguru/shisei

Project overview

Shisei is an original, CC0-licensed pose reference library built for image-generation and figure-drawing workflows. Each pose is rendered as a faceless mannequin from five camera angles — front, three-quarter, side, above and below — so the same pose can be reused consistently across shots. The catalog currently lists 280 poses across four series: standard full body, upper body, a suggestive "sexy" series and a "duo" series, with names and categories in Japanese, Chinese, English and Korean.

Line art is the primary format. Most views also come as clay, depth and normal maps aligned with the line art for ControlNet-style use, and 89 photo renders exist for part of the standard series. Views are quality-checked; rejected views are removed from the catalog rather than replaced with a poor drawing, and data/qc.json records the reason. Standard poses also carry an English pose description, and the catalog records why a view is empty.

Quick start

If you only want images, download the release pack rather than cloning the full repository. The recommended line-art pack for all series contains 280 poses and 1301 images (line art plus photos) at the 900 px tier, with licenses and checksums alongside. For everything, use the complete pack with line art, photos, clay, depth and normal maps; separate clay, depth and normal zips can be fetched when only one format is needed. You do not need to run any code to use these files.

To use a pose, upload one or more angles of the same pose as the pose or composition reference in your image tool, then supply face, clothing and background through your prompt or other references. Multiple angles of one pose help keep the figure consistent across a set of images.

To browse or build the gallery locally, clone the repository and run from the project root:

python3 tools/build_site.py --link-images
python3 -m http.server 8000 -d site

The first command writes site/ and symlinks images for a fast preview; the second serves the gallery at http://localhost:8000. A public deployment build is available with python3 -I tools/build_site.py --public --site https://shisei.indaloop.app. The hosted gallery is also directly available at shisei.indaloop.app.

Using the downloads

Release zips contain the 900 px tier of views that passed QC, with the license files and a .sha256 checksum next to each. Original full-size images (1344 × 1792, photos 1120 × 2240, redrawn sets 900 × 1200) are only in the originals pack. The repository itself holds web tiers: roughly 145 MB of poses plus about 150 MB of costume library.

The suggestive series (sexy and duo, plus some upper-body categories) are hidden or blurred by default in the gallery and excluded from previews, share cards and the sitemap. Adults-only intent is stated for the photo renders, which are AI-generated rather than photographs of real people.

Constraints

Images under images/ and the release packs are dedicated to the public domain under CC0 1.0: they can be used, modified and redistributed commercially without attribution, and generated outputs carry no conditions from the reference images. Code and documentation use Apache-2.0. The pose list — selection, numbering, names and categories — follows ai-pose-gallery's catalog and is credited rather than covered by either license, so check the NOTICE if you plan to redistribute the catalog.

The figures are mannequins or AI-generated adults; the project asks that they not be used to depict real people or minors, or for explicit content. Not every pose has all five views — missing views are left empty. Two standard poses (P038 and P052) are known to show a different pose from their name and remain hidden, and the P062 below view is still open on the roadmap.

HotRepo's take

The project entered HotRepo's new-project list on 2026-11-02 as a Python entry, with 52 stars and 10 forks gained in the roughly 24-hour window ending that day. The collected star curve shows steady, near-continuous growth from 128 to 270 between 2026-10-30 and 2026-11-02, without a single large jump. That pattern is consistent with steady interest rather than one viral spike, though the observation window is too short to say more. No GitHub daily rank has been recorded for the repository, so the available evidence is the growth trace alone.