anhao/ai-pose-gallery-zh
Media & DesignA Chinese-localized mirror of 1,439 AI pose reference images, bundled as a zero-dependency static gallery for artists and 3D/illustration work.
A usable choice if you want a Chinese-language pose reference library you can browse locally or self-host; it is not a replacement for the original artist's terms, and the content is adult-oriented.
The problem
An illustrator or 3D artist wants a specific body pose from a known angle, so they browse a reference gallery, filter by action category and viewpoint, then open a larger image. The original site presents this data in Japanese and does not offer bulk asset downloads, which leaves non-Japanese readers translating labels one by one and makes offline or self-hosted use awkward. This fork mirrors the records and images, translates the metadata, and ships a gallery that reads local JSON. The result is a browsable Chinese reference set with grouping, multi-angle switching and PNG download from the lightbox.
What's different
The core move is data preparation rather than a new UI idea: 1,439 records are paired with both a 300×400 JPG thumbnail and a 900×1200 PNG, 2,878 files totaling about 660 MB. The Japanese fields are translated across four systems, 26 action categories, 288 pose names and five viewing angles, with a 323-entry Japanese–Chinese table kept for proofreading. The single HTML gallery uses no framework, supports pose-name/ID search, combined category and angle filters, grouped or per-image browsing, keyboard/touch navigation and lazy-loaded originals.
The trade-off is weight and provenance. The repository carries hundreds of megabytes of PNGs plus the source data, so cloning and hosting are not lightweight even though the front end is dependency-free. The README is explicit that the images and original data belong to the original artist and that the MIT license covers only the scripts and index.html, so this is a personal study/translation archive, not a commercially licensed asset pack. It also warns that the gallery includes adult-oriented pose references and is not for minors. Deployment absolutes are hard-coded to the maintainer's domain, so a fork on another domain must update canonical, Open Graph/Twitter metadata, JSON-LD, robots.txt and sitemap.xml.
Who should try it
Good fit: a Chinese-speaking illustrator, 3D modeler or photography-reference user who wants to search poses by action and viewpoint without translating Japanese labels, or a front-end learner who wants a concrete static-gallery example with filtering and a lightbox. It is also a reasonable starting point for someone who wants to self-host a reference collection and accepts the large image payload.
Poor fit: anyone who needs a commercially licensed image library, a lightweight repository to clone casually, or content suitable for a general audience. Teams that cannot redistribute the images should treat the data and image directories as off-limits and use the repository only to study the gallery code. The documented content warning is a real constraint, not a footnote.
Requirements and limitations
Local preview requires an HTTP server because the page fetches JSON; the README suggests python3 -m http.server, and opening index.html directly will be blocked by CORS. Python 3 is needed for the download, translation and verification scripts, and the image payload is large enough that storage and transfer matter more than the code. Hosting on Cloudflare Pages is documented with a build step that copies only site assets into dist/; the README specifically warns against pointing the build output at the repository root because the .git directory can exceed the platform's per-file limit. The site fits the stated limits of files under 25 MiB and fewer than 20,000 files, but that is a deployment constraint to check if you add images.
The original Japanese data is kept unchanged alongside the Chinese version, so updates to the upstream site are not automatic; refreshing requires rerunning the downloader and regenerating translations. Translation quality is the maintainer's work and the CSV exists for correction, but the repository does not present an independent review of every translated label.
HotRepo's take
HotRepo recorded 406 total stars and a #17 position on the NEW-REPOSITORY board (新项目榜, not the overall Trending board), all languages, observed 2026-10-28. That is a single dated snapshot, and the repository was created on 2026-10-27, so persistence cannot yet be judged from HotRepo's history.
Before relying on it, check the original artist's terms and whether your use fits the repository's stated study-only scope; the gallery code itself is easy to inspect from the live demo and project instructions.
Written by AI from the project's README, updated October 28, 2026. The project's own docs are the source of truth.