Someone else's free image gallery, and the Chinese version got 6,000 stars 9 hours later: how ai-pose-gallery-zh won the race

In October 27, a promotional account in Japan's AI beauty circles posted 1,439 sets of "AI beauty pose" blank-model images on X for free. That afternoon, a Chinese developer with barely a thousand followers moved the whole site's images and Japanese tags into GitHub, translated everything into Chinese, and deployed it. Ten days later the repo had 6,246 stars and 1,588 forks. All the Chinese, English and ComfyUI versions of the same images together don't come close. It won on speed, and only on speed. Below: what it is, why it was this one, and whether you can use the same move.

One Japanese post, and 9 hours

In the morning of October 27 (around 03:00 UTC), the Japanese promotional account ギア (@giaaibusi) published a long X post titled 「【1,439のポーズ無料配布】AI美女「ポージング」データベース爆誕!」, roughly: 1,439 poses free to download, commercial use and modification allowed. Japan's AI circles reposted it heavily, and someone replied 「これめちゃくちゃ使えます」 (this is super useful). The same day, the Chinese account 知识猫 @GeekCatX also reposted it, with a blunt caption: "Japanese netizens are the real pros at AI… the bros making explicit content… bookmark it hard."

12:24 (UTC), developer Anhao (GitHub anhao, X @anhao_ai 「安好」) created the repo ai-pose-gallery-zh. The first commit read "initialize Chinese version of AI pose gallery". An hour later he added share cards and a search engine description, and an hour after that, visit analytics. That day he posted: "1439 AI beauty pose reference images, open-sourced in Chinese. Download them directly to make videos. Essential for AI image work."

By the end of that day (10-27 23:15 UTC) the repo had 316 stars (archived on rising.dev). The curve afterward looked like this:

The single-day peak was +1,074 on 10-29; by 11-06 the daily gain had fallen to about 140. The author's base was thin: 107 GitHub followers, 1,149 X followers, and his previous best repo had only 83 stars. This project wasn't lifted by the author's own following.

What it actually is

A static web gallery plus a 1.1 GB repo. It holds all of ギア's images: pose blank models with matching depth maps, outfit library, portrait library, over 12,000 images locally. The Japanese motion categories, outfit tags and hairstyles were all translated into Chinese, so a click assembles a generation prompt. You can view it online (ai-pose-gallery.anhao.app), or fork it and deploy your own copy with the bundled Cloudflare config.

After it blew up, the author added a lot over the next three days. 10-31's v2.0 is a "three-in-one AI character compositing workbench": pick a portrait, an outfit and a pose, then generate directly through Gemini, GPT Image, Alibaba Bailian, Volcano Ark or a local ComfyUI. The same day he shipped a ComfyUI plugin, pinned with the post "completely free of cloud moderation limits", which reached 22k views. The community merged 22 PRs, someone added Agent skills, and someone asked in an issue for a switch between line art and depth maps.

But all of that came after it blew up. What made it blow up was that first day's "Chinese version of the image set".

Why suddenly people wanted these images

Two conditions came together.

First, the models changed. GPT Image, Nano Banana (Gemini's image model), Seedream, Qwen Image Edit: this generation of editing models can take one reference image and "swap it to this pose". ギア's original pitch was literally 「このポーズに変えて」 (change to this pose). Consistent, multi-angle, clean blank-model pose sets went from an illustrator's niche material to something generation workflows need.

Second, there were people who wanted it. Lots of people in the Chinese-speaking world make AI beauty images, AI short dramas and suggestive videos, but rarely say so openly. The gallery already had categories like "full-body solo, sexy" and "two-person intimate"; the reposting lines were things like "for the people making explicit stuff" and "not for minors". Many stars were "bookmarked for later" (my inference).

Why this one and not the others

Within 48 hours, the same images had more than a dozen versions:

The runner-up had a ninth of its stars. The content is the same images, and the features aren't very different. The gap comes almost entirely from two things:

• It was hours to two days early. It went live about 9 hours after ギア's post, inside the same news cycle.
• It got passed around as "that link". From 10-28, AI and reposting accounts on Chinese X passed it along one after another: Alex Zhang (@jojogh_007), 周览资源, AIExplorerTim, ShinChven, 领哥, and 10-31 from GitHubDaily. On Taiwan's Threads, @hugoyu wrote "Photographers, you have to bookmark this site." After that came Zhihu columns, CSDN ("4 days, 4500+ stars"), Sohu, a YouTube GitHub Daily episode, and reposting accounts on Facebook and Instagram. I found no evidence of paid promotion; the wording is so similar that copying from each other is more likely.

One more number stands out: forks are a quarter of stars, far above a typical project. The reason is that the images are in the repo, plus one-click deployment, and the original site explicitly forbids redistribution, so people fork it as a backup in case it's taken down. Some fork repos are literally named "pose reference gallery backup". Lots of forks make it look even hotter on GitHub.

Its biggest problem: the images aren't its own

The original ギア site's terms say the images can't be redistributed, reposted or resold, and bulk downloading is banned, including automated scraping with tools. A full Chinese mirror does both. On X, people have already publicly called it an unauthorized mirror of the full original image set, and a Zhihu article raised it too.

Gia isn't running a charity either. His profile reads "欲望市場 × テクノロジーの攻略専用垢" (Desire Market × Tech Strategy Account), and November sells a 48-hour limited "AI Beauty Final Form" program and runs workshops. The free pose library is his lead magnet for selling courses, so a full Chinese mirror would take that lead magnet away from him. Whether he sends a DMCA takedown isn't clear yet, but he has the motive.

The author seems to have noticed: the README now says "for Chinese translation, organization and engineering study archival only", and adult content is hidden by default in a safe mode. For the 1,320 anime outfit entries added on 11-03, the repo stores only metadata and the original site's image URLs and no longer downloads the images.

If it were you

The bottom line: this one can't be followed anymore. The same images have more than a dozen versions, the media has covered them, the window has closed, and it's built on someone else's rights. What's worth taking is the method: on the day a foreign free resource blows up in local circles, make the first Chinese version, open-source it with a live site, and push it to Chinese AI accounts. There's something to learn here, but following this specific one isn't worth it.

Who it suits: people who follow Japanese and English AI circles daily and can finish the scrape → translate → deploy chain within hours. It's even better if you have a few Chinese AI accounts willing to repost for you.

Who it doesn't suit: anyone hoping to earn stars by making yet another Chinese pose library. Second place with 726 and the single-digit numbers after it are your benchmark. Also anyone who worries about copyright disputes or doesn't want a repo that could draw a DMCA on their GitHub account.

How to do it:

• Don't make this set again. Watch for the next one: a free resource that suddenly gets heavily reposted in Japanese, Korean or English AI circles (prompt collections, reference image libraries, workflows, model packs), and build the Chinese version within a few hours of release. It wins on those few hours.
• Check the license first. If the original author said redistribution is OK, copy it in full the same way, add one-click deployment, and expect the "backup" need to amplify forks. If redistribution is forbidden, host only an index, Chinese tags and links to the original site (like it did later with the anime outfits). You'll get fewer stars, but it won't be deleted overnight.
• Add share cards and a search description on day one. It did that within the first hour.
• Once it blows up, move quickly into tooling: ComfyUI nodes, a generation workbench, Agent skills. That turns a one-time bookmark into an ongoing topic. That's how it stayed hot for a week.
• Or do the part it can't: draw or render your own pose white-model set (like Shisei does). Clean copyright, done once, and anyone can safely pass it on.

How to get traffic: your first post should say plainly what it is, "Chinese version, free, download directly." One line is enough; that's how its own post was written. After posting, leave the link under the Chinese reposts of the source tweet, then DM a few Chinese AI accounts and daily-digest accounts like GitHubDaily. Threads and Facebook groups in Taiwan were places it didn't target on purpose but still got traffic from, so go post there yourself.

Timing: this playbook depends on the same news cycle, meaning the 12 to 24 hours after the source goes viral. After 48 hours, you're the dozenth clone.

Signs you read it wrong: ① ギア or other source authors start filing DMCA takedowns and the repo gets pulled. That means the "mirror all the images" route is closed; only indexing and your own assets remain. ② The next similar overseas resource goes viral and the first Chinese version doesn't reach a few thousand stars. That means this demand for adult-oriented AI content just happened to coincide, not that the playbook itself works. ③ Models become good enough at posing from text descriptions alone that they no longer need reference images. Demand for this kind of asset library would drop with it.

Data and sources

GitHub repo page, README, commit and issue history; the daily star and fork records (starting 10-28) and live numbers as of 11-06; first-day 316 stars from the rising.dev archive; posts on X by ギア (@giaaibusi), 知识猫 (@GeekCatX), @anhao_ai, @jojogh_007, GitHubDaily and others; Threads @hugoyu; Zhihu columns, CSDN DevPress, Sohu coverage; usage terms on ギア's original site ai-pose-gallery.pages.dev (read 11-06); competitor star counts as of 11-06.