AI Pose Gallery ZH: 288 Sets and 1,439 Reference Images

By Indie Kings | October 29, 2026

Updated October 29, 2026: anhao/ai-pose-gallery-zh mirror and translates a Japanese AI pose reference library into Chinese, shipping 1,439 images across 288 pose sets in a single static site. The repository was created on October 27, 2026, has 1,802 stars and one contributor, and its 659.5 megabytes of image data are explicitly excluded from the MIT license that covers the code.

Note: this project contains adult-oriented pose reference material, and some category names use sexual terminology. It is intended for adults and for art, animation and photography reference only. This article describes the software and its data pipeline and does not reproduce the material.

Image: interior of Kelvingrove Art Gallery and Museum, Glasgow, via Wikimedia Commons.

A mirror, a translation, a front end

The project began as an archiving exercise. The original site, ai-pose-gallery.pages.dev, is a reference library of human poses for AI image generation, 3D work and photography, with metadata written in Japanese and no bulk download. The Chinese mirror does three things the original does not: it stores every thumbnail and every full resolution file locally, translates the metadata, and replaces the front end with a dependency-free responsive gallery.

The numbers are specific enough to check. There are 288 pose sets and 1,439 images, which works out to roughly five angles per set. Each image exists twice, as a 300 by 400 JPEG thumbnail and a 900 by 1200 PNG original, for 2,878 files and 659.5 megabytes. The metadata covers four systems, 26 categories, 288 pose names and five viewpoints, and the translation dictionary holds 323 unique entries. The author states that the localization is complete, with no fields left in Japanese.

The front end is deliberately plain: no framework, no build step, no package manager. That is a reasonable choice for a static gallery that has to run from any folder or cheap host, and it makes the site legible to anyone who wants to fork it. The feature list reads like a competent image browser: tabs per system with counts, search by pose name, number, ID or category keyword, filters for category and viewpoint, a grouped or per-image view, a lightbox that loads the thumbnail first and then the original, download and copy-ID buttons, arrow key navigation and swipe on phones.

The hard part is the metadata

Scraping images is mechanical. Translating a controlled vocabulary is not. A pose library that mixes everyday terms with anatomical and, in this case, sexual terminology needs a consistent mapping, or search stops working: the same concept appears under three Chinese phrases and none of them match the filter. The repository's approach is a single dictionary file of unique values, deduplicated and translated once, with a script that regenerates the Chinese data set when the dictionary changes. That is the right shape for the problem, because it keeps the vocabulary reviewable in one place instead of scattered across records.

The gallery also carries more SEO work than most hobby static sites. The page sets a keyword-rich title, a separate description, keywords, author, a canonical URL pointing at the custom domain to avoid duplicate indexing with the pages.dev host, and Open Graph and Twitter card metadata with image previews. It is the kind of configuration that a marketing team usually adds and an individual maintainer often skips, and its presence suggests the mirror is meant to be found rather than kept as a private archive.

There is also a practical argument for the mirror that has nothing to do with language. Hosted galleries disappear. A site run by one artist can go offline, change its URL structure or restrict hotlinking, and a reference library that people use for daily work vanishes with it. A repository with the files checked in is a blunt but effective preservation strategy, provided the licensing allows it, which is where this project gets complicated.

What the license does not cover

The README is unusually direct about rights. The code in the scripts folder and the index page is MIT licensed. The images, thumbnails and data are not covered by that license, the repository claims no rights over the images, and it does not grant commercial use of the original works. The images remain the property of the original author, credited as Gia, and the Chinese edition is maintained by Anhao. The stated purpose is personal study, localization practice and static site work.

That distinction matters more than usual here because the repository is a full mirror rather than a set of links. Mirroring is exactly the case copyright law treats least generously, and a README disclaimer does not transfer rights. Anyone who wants to use the images commercially, or redistribute them further, needs permission from the original creator rather than from this repository. The MIT license on the code is real and useful for anyone building a similar gallery, but it stops at the img and data folders.

The content warning is the other boundary. The library includes adult material, the README asks minors not to access it and asks users not to put it to unlawful use, and it frames the whole collection as reference for art, animation and photography. That framing is consistent with how pose libraries are used in studios, and it is also the only defensible one for a public mirror. The project is best read as a localization and front end case study with a large, rights-restricted data set attached, not as a free asset pack.

Project card

FAQ

What is ai-pose-gallery-zh?
A Chinese language mirror of a Japanese pose reference library, with all 1,439 images stored in the repository and a new static gallery front end.

Is the code open source?
Yes, the scripts and index page are MIT licensed. The images and metadata are not, and the repository claims no rights over them.

Can I use the images commercially?
Not under this repository's license. Permission has to come from the original creator, who is credited as Gia.

How large is the download?
About 659.5 megabytes across 2,878 files, split between 300 by 400 thumbnails and 900 by 1200 originals.

Does it work without a server?
Yes. The gallery is plain HTML, CSS and JavaScript with no build step or dependencies, so it can be served from any static host or opened locally.

Who is it for?
Adults who need pose reference for illustration, 3D or photography. The README carries an 18+ notice and asks users not to access it if they are minors.

Bottom Line

The interesting engineering here is not the gallery, it is the vocabulary work. A 323 entry dictionary that makes 288 pose names searchable in a second language is a small, complete localization project, and the static front end is a clean example of doing a lot with no dependencies. The rights position is the caveat that decides whether you can use it: the code is MIT, the images are not, and a mirror does not change that. Read it as a case study in archiving and translation, respect the 18+ boundary, and treat the data set as someone else's work on loan.

Sources

• GitHub anhao/ai-pose-gallery-zh README, file counts, metadata coverage, license and content notice, accessed October 29, 2026
• GitHub API metadata for anhao/ai-pose-gallery-zh, stars, forks, contributors and creation date, accessed October 29, 2026
• ai-pose-gallery.anhao.app live site and ai-pose-gallery.pages.dev original, referenced by the README, accessed October 29, 2026

Related: Filee: A Drag and Drop Converter for 266 Formats With No Uploads | LinkSwift: A 30,000 Star Userscript for 9 Chinese Cloud Drives