The runner-up, 19 hours later: a repo that "repackages" free Japanese assets, reached 1278 stars in 9 days

A Japanese creator of AI beauty images released three reference libraries for free on X, and Chinese developers rushed to make "Chinese versions" overnight. RuiC-PoseAtlas wasn't the first, and the assets aren't its own, but by bundling three libraries in three languages with originals you can take as a full package, it placed second among dozens of identical repos. What it can teach you is a way to earn stars, not a spot you can still squeeze into.

A Japanese tweet, and a repo created on a Wednesday afternoon

Around Oct 26, a Japanese X account called ギア (@giaaibusi) posted: "AI美女「ポージング」データベース無料公開!1,439枚、無料、商用OK" (AI beauty pose database released free! 1439 images, free, commercial use OK). Later it turned out this was the alt account of AI beauty creator あぎ (@agi_aibusi).

The images show 288 poses, each in five views (front, three-quarter, side, top-down, and bottom-up), gray faceless mannequins, used as skeleton references for AI image generation or drawing. The post spread fast in Chinese and Japanese AI image circles, with comments in Chinese, Korean, and English. Agi later said "I don't know why it got such a big reaction overseas." In October, Chinese AI account Mu Feng (@mr_mofun) relayed it in Chinese: "Free for commercial use! 1439 AI portrait pose reference library is here" (27).

That same night, a little after 8 p.m. (Beijing time), the first Chinese version, anhao/ai-pose-gallery-zh, was created on GitHub. The next afternoon at 3 o'clock, HRuiCcc created RuiC-PoseAtlas, mirroring all 1439 original and thumbnail images into the repo within 7 minutes. GitHub Pages went live 20 minutes later.

Around Oct 30, あぎ posted on X: "The database I released on a small account blew up, and then someone made a Chinese and Korean version without permission and is publishing it as an open-source project. Lol w"

What it actually is

A pure static website: HTML + CSS + vanilla JS, no framework, no dependencies, three pages:

All three sets come from the original creator's three Japanese sites. The README is candid: "This project only handles data organization, filling in English names, trilingual tag alignment, and frontend rewrite." What it added itself: language switching and search in Chinese, English, and Japanese, arrow keys to switch angles, shareable links to a specific angle, and a validation script that automatically checks images and translations before submission. And it packaged the original images right into the repo: the repo is 817 MB, so one clone gets you the full pose and outfit originals. The portrait originals at 4.5 GB don't fit, so they link directly to the original site. At the top of the README are three full-pack Quark Drive downloads (poses 659 MB, outfits 1.38 GB, faces 4.86 GB), with a WeChat tip code and WeChat ID further down.

README 开头有一行「面向成年创作者,部分内容不适合未成年人」。这不是客套:台湾媒体动區動趨 11-01 的报道说,原作者的库主要做给 AI 美女和成人内容市场,姿势库约三分之二偏成人向。

How the stars grew

(GitHub daily stars / forks)

The biggest jump was +292 in a single day on 10-31, right after the outfit and portrait libraries launched. It still gains about +70 a day now, slowly tapering off. The fork-to-star ratio is 24%, far above typical projects. Many people came to take the assets.

Where people came from, by share:

1. Spillover from the original creator's viral post (confirmed). This is the main driver. Chinese users who saw the Japanese post first went looking for a "Chinese version", and each time the original creator released a new library, this project followed within a day, and each wave brought another wave of stars.
2. Following GitHub breadcrumbs (inferred). The repo was only 9 days old and couldn't get onto GitHub Trending, so people likely searched "pose gallery" or "姿势库", or found it after seeing the anhao version. A dozen forks were deployed directly as their own mirror sites, and someone built a ComfyUI node on top of it.
3. Auto-aggregator sites and tutorial accounts (confirmed). hotrepo.ai, gitstars.io, the Telegram channel Code Stars, YouTube Shorts @LookOnThisGit, and several Chinese tutorial sites wrote about it within 10-29 days. These sites specialize in fast-rising repos, so being listed brings traffic back in turn.
4. Media (confirmed, but late). Product manager site 人人都是产品经理 ran a feature piece 11-03, "Pick the pose, pick the outfit, set the face", largely built around the README line "Pose sets the skeleton, costume sets the look, portrait sets the face." By then the peak had already passed.

The author HRuiCcc's account was registered in August. Their two earlier projects each had a few hundred stars, and it has 74 followers. This surge barely relied on old fans. I found no sign of paid ads, a Hacker News post, or a big-name shoutout.

Who's competing: dozens of repos, same source material

The top one has 4.6 times its stars. The gap isn't features, it's those 19 hours. The earliest version got cited by nav sites and tutorial sites, so later searchers found it first. RuiC's three-in-one library, three languages, and full-package download put it second among the followers, but the traffic came from the material itself going viral, not these differences.

Why it broke out

• Timing. A single material-release event. Went live within 22 hours. Each time the original author added a new library, it was matched within a day.
• Merged scattered stuff into one workflow. Three Japanese sites each handled one piece. This one combined them into a single site, and added a catchy line, "Pose sets the skeleton, costume sets the look, portrait sets the face", that media and tutorial accounts could lift directly.
• Did what the original site wouldn't allow. The original banned bulk downloads and redistribution. This one put the original images into the repo and into cloud storage. That's the selling point, and also the biggest risk. The original author is only smiling about it for now. If they ask for takedown someday, the repo is gone.
• Looks like a legit open-source project. Three languages, zero dependencies, validation scripts, a tidy README. Easy to get indexed and shared.
• The demand is real. AI beauty and character image generation is already a big market, and this material is exactly what those people use every day.

If it were you

C: This wave is over, but the playbook is worth learning. There are already dozens of Chinese versions of the same material. The top one, with 6000 more stars, has locked the search entry point. The window is measured in days and has long closed. Building another one won't get more than a few stars, and it carries copyright risk on top. What's actually worth taking is the playbook: free resource that just went viral overseas → launch a multilingual aggregator site within 24 hours → follow every update the original author makes.

Who it suits: a solo builder who moves fast, already hangs out in the Japanese/Korean X circles for AI image generation, design assets, and indie games, and can take a static site from zero to live within a day. Wants stars, followers, and a project that looks good on a resume.

Who it doesn't suit: anyone who wants to build an open-source project that can be maintained long-term and grow slowly (these repos follow a spike-then-fade curve); anyone sensitive to copyright or who doesn't want to be called out by the original author; anyone who wants to chase these pose libraries right now.

How to do it:

• Watch the source, not GitHub. Whenever material, templates, or datasets marked "無料公開" or "商用OK" appear on Japanese or Korean X, immediately ask: is there a large group of users who don't read Japanese or Korean but would want to use this?
• Being first matters more than doing it well. RuiC was 19 hours behind anhao, more complete, and still only had a fifth of its stars. The target is to go live within 12 hours.
• Check licensing first. Where the original explicitly bans redistribution, build it as an "index + translation + link back to the original", and don't move the original images into the repo. It's worth asking the original author first. Getting permission is something no one else can take from you.
• Build the layer the original author didn't: multiple languages, merging several libraries, plugging into tools (ComfyUI nodes, Agent Skill, outfit-swap workflows). Some of the same-material projects already have a ComfyUI node, an Agent Skill, and a site with one-click outfit swapping. This layer is harder to drown out than plain translation.

How to get stars: on launch day, reply with the link under the original post and under the Chinese repost. Make the first screen of the README clearly explain what it is and how to use it, with one line people can quote. Once stars climb fast enough, sites like hotrepo and gitstars pick it up naturally, and tutorial accounts follow.

Deadline: each opportunity like this only has 24 to 48 hours after the original post goes out. This batch (the ギア / あぎ pose, costume, and portrait libraries) has already passed. The same author is already releasing new libraries like the depth-map version POSE-DB-02. Watching for this author's next release is the only point on this path you can still catch.

Signs you misread it: (1) The original author or platform demands takedowns and the Chinese mirror repos disappear en masse. That would mean copying the material is a dead end, and only "index + link back to the original" survives. (2) Two weeks later, RuiC or anhao is still gaining hundreds of stars a day. That would mean this isn't a one-off event but lasting demand for reference images, and it would be worth building a version with your own material and your own update schedule. (3) Someone builds an original library of the same kind (self-generated, self-licensed) and gets over a thousand stars in a week. That would mean readers want the material itself, and whoever has the clean source wins.