Open‑Source CC0 Mannequin Pose Library Released for AI Image Generators

Overview

A new GitHub repository, shisei, offers a completely free (CC0) mannequin pose reference library aimed at AI‑image tools such as Nano Banana, ChatGPT, ComfyUI, Midjourney and others. The collection contains 280 poses organized into four series, each rendered from five camera angles – front, three‑quarter, side, above and below – in clean line‑art format.

Content Details

• Series & Ratings: Four series (standard, sexy, duo, upper) with a rating system that flags suggestive poses. The “sexy” and “duo” series are blurred by default in the web gallery and omitted from search indexes.
• Multi‑language Metadata: Pose names and categories are provided in Japanese, Chinese, English and Korean, making the catalog searchable across language barriers.
• Additional Assets: For 270 sets, the line art is accompanied by clay renders, depth maps and normal maps (ControlNet‑compatible). Photo renders exist for a subset of standard poses (e.g., P001‑P022, P050).
• Quality Control: Every view undergoes a documented QC process (see `docs/qc/`). Views that contain errors – such as unintended facial features, clothing, background elements, anatomical mistakes, or depictions that could be interpreted as minors – are removed and logged in `data/qc.json`.
• Version & Size: The current release is v0.4.1 (≈ 145 MB for poses, plus ≈ 150 MB for a costume library). Downloadable ZIP files contain the 900 px‑tier images that passed QC, along with license files and SHA‑256 checksums.

Licensing and Usage

All pose images and associated data are released under CC0 1.0 – they can be used, modified and redistributed commercially or non‑commercially without attribution. The code and documentation are under the Apache 2.0 license. The pose list itself is credited to the public catalog of *ai‑pose‑gallery* and is not covered by a license, but the list is shared for reference only.

How to Access and Verify

1. Visit the repository: https://github.com/wquguru/shisei
2. Download the desired ZIP from the *Releases* page (line‑art zip is sufficient for most users).
3. Cross‑check the downloaded files against the catalog (`data/catalog.json` or `catalog.csv`). The JSON lists each pose ID, series, rating, multilingual names and the file name for every angle.
4. Validate integrity using the provided `.sha256` files.
5. Inspect QC status by opening `data/qc.json` – it explains why any view is missing or flagged.

Practical Implications for Creators

• Consistent Multi‑Angle Prompts: By feeding the same pose from different angles to an image model, creators can keep a figure stable across a series of shots (e.g., storyboard panels or product mock‑ups).
• No Attribution Burden: CC0 status removes the need for credit lines, simplifying commercial pipelines.
• Safe‑for‑Work Controls: Suggestive poses are hidden by default, allowing teams to enforce content policies while still having the assets available when needed.
• Extensible Workflow: The accompanying depth and normal maps enable advanced ControlNet or diffusion‑based lighting control.

Contribution and Future Plans

The project welcomes pull‑requests and issue reports. Contributors should read `CONTRIBUTING.md` and follow the coding guidelines in `AGENTS.md`. Security concerns are to be reported privately as described in `SECURITY.md`. The maintainers note that the catalog IDs are stable and will not change in future releases.

--- *All information is taken from the repository’s README, documentation files and release notes. No independent verification beyond the public repo has been performed.*