How to Use Shisei’s 280-Pose Mannequin Reference Library for AI Image Creation

What Shisei Does

Shisei is a free mannequin pose reference library for AI image tools and other creative workflows. It contains 280 pose sets across four series: standard full body, upper body, sexy, and duo.

Each pose is drawn from up to five camera angles: front, three-quarter, side, above, and below. Using multiple views of the same pose can help keep a figure’s composition consistent across generated images. The face, clothing, and background are supplied by your prompt or other references.

Pose names and categories are available in Japanese, Chinese, English, and Korean. The project includes clean line art, 89 reviewed photo renders for selected standard poses, and a static gallery that requires no runtime dependencies.

Content note: The sexy and duo series, along with some suggestive categories, contain adult pose references. They use faceless mannequins without nudity detail and are blurred or hidden by default in the gallery.

Choose a Way to Access the Library

The quickest option is the online gallery. It provides series and category navigation, angle and source filters, pose pages, a lightbox, keyboard and swipe navigation, comparison mode, four interface languages, and a dark theme.

For local or offline use, download the recommended all-series w900 archive. This package contains 1,222 quality-controlled images at 900 pixels wide, together with the applicable licenses and notice files.

You can also select a smaller series-specific archive:

• Standard full body: 71 sets and 376 images.
• Upper body: 80 sets and 337 images.
• Sexy full body: 80 sets and 323 images.
• Duo: 49 sets and 186 images.
• Originals: the full-size source images.

Understand the Image Tiers

The repository includes three web-optimized sizes. Choose the smallest tier that meets your reference needs:

• w300: 300 pixels wide, typically about 7 KB.
• w600: 600 by 800 pixels, typically about 17 KB.
• w900: 900 by 1,200 pixels, typically about 29 KB.
• Full size: the original images, generally about 100 KB each. Photo renders and redrawn sets use some different dimensions.

The web tiers are stored under images/. Full-size originals and release archives are available from the GitHub Releases page.

Use a Pose as an AI Reference

1. Open the gallery or extract a release archive.
2. Find a pose by series, category, angle, or pose ID.
3. Choose line art or a photo render when both are available.
4. Upload the selected view to your image tool as a pose or composition reference.
5. Add your own instructions for the face, clothing, environment, lighting, and visual style.
6. For a more consistent figure, provide several angles from the same pose set instead of mixing unrelated poses.

Shisei references are designed to communicate body position and camera composition. They do not determine every visual detail of the result, so describe those details separately in your prompt or provide additional references.

Run the Gallery Locally

To work with the repository and its source files, obtain a copy of the project and make sure Python 3 is available. The gallery is a static site and has no runtime dependencies.

git clone https://github.com/wquguru/shisei.git
cd shisei

Build a local preview with images linked from the repository:

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

Open http://localhost:8000 in your browser. The --link-images option creates a fast preview by linking the existing image files rather than copying them into the generated site.

For a deployment-oriented build, provide the public site URL:

python3 -I tools/build_site.py --public --site https://shisei.indaloop.app

The public build includes images and share cards. Review the project’s design documentation before adapting the gallery.

Use Catalog Data in Your Workflow

The catalog is the source of truth for the library. Use data/catalog.json or data/catalog.csv to inspect series, ratings, categories, multilingual names, standard-series descriptions, image filenames, and empty views.

The repository layout separates responsibilities:

• data/ contains catalog files.
• images/ contains line-art and photo tiers.
• web/ contains the static gallery source.
• tools/ contains build, verification, import, review, and release utilities.
• docs/ contains design, generation, licensing, and quality-control documentation.

Use stable pose IDs such as P001, U001, S001, or D001 when building links or catalog-based tooling. IDs are intended to remain unchanged.

Validate Images and Catalog Entries

After modifying image files or catalog data, run the included verification script:

python3 tools/verify.py

This checks that the catalog and image files agree. Empty views remain empty when no acceptable drawing exists; they should not be replaced with an unreviewed image.

Shisei’s quality-control process removes views with issues such as a face or hair, clothing, background, broken anatomy, an incorrect camera angle, a figure that reads as a minor, or a drawing that does not match the named pose. Review the QC documentation before contributing or importing material.

Use Deep Links and Filters

The gallery supports direct links to a lightbox view. For example, these fragments identify a standard pose and display mode:

• #P001-diag opens a line-art view.
• #P001-diag-photo opens the photo view.
• #P001-diag-compare opens comparison mode.

Filters are represented in the query string. This example selects Japanese, the standard series, category 5, angle 1, and photo source:

?lang=ja&series=std&cat=5&angle=1&src=photo

These URLs are useful for sharing a specific reference or integrating a selected pose into documentation and internal workflows.

Licensing and Responsible Use

The images in images/, the release archives, and the project’s own data are dedicated to the public domain under CC0 1.0. They may be used, modified, and redistributed commercially or non-commercially without attribution. Images used as pose references place no conditions on the images you generate.

The code and documentation use the Apache License 2.0. The pose list, including its selection, numbering, names, and categories, follows the catalog of ai-pose-gallery and is credited separately; the project’s licenses do not cover that catalog.

The project states that its figures are mannequins or AI-generated adults and do not depict real people. Do not use the references to depict real people or minors, or for explicit content. Read the included license and notice files before redistributing a package.

Conclusion

Shisei provides a practical, multilingual set of 280 mannequin pose references with multiple camera angles, downloadable image tiers, reviewed line art, selected photo renders, and a dependency-free static gallery. Start with the online gallery or the w900 release archive, use consistent angles from one pose set, and rely on the catalog and verification tool when building a local workflow.