Stringers
HRuiCcc/RuiC-2D · HRuiCcc

Copied a Japanese dev's idea from a month earlier, 9 days, 160 stars: RuiC-2D and the Chinese "new GitHub repo reposter" accounts

What it is A free tool that runs on your own computer: drop in a character portrait and get a 2D avatar that blinks, lip-syncs to audio, and can be added to OBS for live streaming.

Why it took off Less than a day after the repo was created, a Douyin reposter published a video. About a dozen alt accounts followed, many tagged #豆包工作, building a long tail. The author's previous repo had just been pushed by the same accounts, and the demo character rode the "Big Fat Fish" meme.

CHow you could follow it Copy the playbook with a different topic: watch overseas open-source tools that just blew up and have a high barrier to entry, ship a Chinese version with no setup friction within two or three weeks, pair it with a live demo page and a 30-second clip, and reach out to Chinese reposter accounts. Don't build a third single-image-to-avatar editor.

Grade C: The playbook is clear and copyable, but the topic already has an original and several similar tools. It reached 9 stars in 160 days on its own, while the original had over 2,000 at the same time. Part of those stars came from the author's earlier followers, so copying the approach would yield little. Chatter in small circles
Data (updated daily) · Repo created 2026-11-06 · 282 stars · 7 days +165
GitHub · stars
010020030011-0811-1111-1411-1711-20282
Data 2026-11-08 → 2026-11-20 (13 days, 13 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Drop in one character portrait and get a 2D character that blinks, lip-syncs, and can be added to OBS for live streaming. The idea came from a Japanese developer who built it a month earlier, and the original already has over 2,000 stars. A Chinese indie dev turned it into a local, Chinese-language version that doesn't require an AI coding assistant and ships with a "Big Fat Fish" sample character. It got 160 stars in 9 days. The stars didn't come mainly from the product. They came from a group of small accounts on Douyin and Xiaohongshu that specialize in reposting new GitHub repos. Below: how that path worked, where its ceiling is, and how you could get the same kind of stars.

Less than a day after the repo was created, someone posted it on Douyin.

On Nov 2026 6 at 3:14 (Beijing time), HRuiCcc created the RuiC-2D repo on GitHub, and that same day added a WeChat tip QR code and a live demo page.

The next day at 1:48 noon, a Douyin account called 梦随笔至 with only 79 followers posted a video titled "2D virtual avatar world's highest mountain is here, ruic-2d tutorial," tagged #豆包 and #豆包工作, with 101 likes and 18 shares. The account usually posts nothing but GitHub new-project recommendations, and it had already featured this author's previous project back in Oct 30.

Over the next two days, several accounts ("织一束月色", "三兆兆", "小李早吖", "两清/") posted nearly identical titles: "2D avatar (world's) tallest mountain (it) has arrived", with the same #豆包工作 tag. By 11-14, Douyin had at least 9 videos about it, Xiaohongshu 2, and Bilibili 1.

What it is

A 2D avatar editor that runs on your own computer, MIT-licensed and free. The first line of the README reads: "From a single image to a 2D character that blinks, talks, and can go live."

  • Import a front-facing portrait. Person segmentation and facial layering run locally. If detection is off, click the left eye, right eye, and mouth manually to continue.
  • The head, body, eyes, mouth, tail, and hair tufts all move. You can also outline custom parts such as hands and sleeves along the contour.
  • Import voice audio to lip-sync by volume. On the timeline, double-click to add blinks or tail swings. Turn on the webcam to track your face with MediaPipe, then copy a URL and use it as an OBS browser source.
  • Export an editable character pack, transparent PNGs, and 1080 square transparent WebM.
  • pip install plus npm start and it runs. No API key needed. There's also a live demo page where you can click and hear the sample character "Big Fat Fish 1" introduce itself.
  • 11-09 also integrated LivLiv's player and 21 ready-made characters (non-commercial use only). The homepage shows a row of clickable characters on open. No new commits since.

The idea isn't its own.

The README says: "Base runtime and layering code reference Mesh Avatar Studio," and it can export the compatible .mavatar format.

Mesh Avatar Studio was open-sourced in Oct 4 by Japanese developer shinshin86. As of 11-14 it has 2,366 stars and 252 forks, with READMEs in Chinese, English, and Japanese. On 11-10 and 11-11 it gained 53 and 228 stars in a single day, faster than RuiC-2D. A 11-13 Xiaohongshu post titled "turn a portrait into a movable avatar" got 941 likes and 1,715 saves, and watching the original clip shows it's about Mesh Avatar, not RuiC-2D.

The biggest difference is the barrier to entry. Mesh Avatar requires installing Node 22, then having Claude Code or Codex split the image and build the character following the instructions, then fine-tuning in the editor. RuiC-2D uses a local Python pipeline that layers the image automatically, has a Chinese UI, needs no AI coding assistant, and adds a timeline, lip-sync to audio, WebM export, and custom parts.

This space is quite crowded:

ProjectLaunchedScaleHow it differs from RuiC-2D
shinshin86/mesh-avatar-studio2026-10-042,366 starOriginal; characters are built with an AI coding assistant
livliv-dev/player + livliv.aiOpen-source player 2026-11-041,550 stars, 11-07 gained 712 in one dayOverseas web product where AI builds the character online; RuiC-2D bundles its player directly
shitagaki-lab/see-through2026-035,670 starAcademic project that layers single-image anime characters and outputs layers only, with no animation
psd2live, still2rig-psd, rigby2d, etc.2026-07~10Dozens to 1,090 starsStarts from layered PSD files, or is a skill pack for AI assistants
Inochi2D, PNGTuber-Remix, etc.2020~2024A few hundred to 1,867 starsEstablished open-source alternatives to Live2D that require drawing your own layered assets
Live2D Cubism + VTube StudioYearsIndustry standardRequires layered PSD files and professional rigging; a custom skin set costs thousands to tens of thousands of yuan

(GitHub,11-14)

Where the stars came from

Daily stars (GitHub, from 11-08):

DatestarAdded that day
11-0840—
11-0974+34
11-1084+10
11-1197+13
11-12105+8
11-13117+12
11-14144+27
11-15162+18

No explosive growth. It never made GitHub Trending. It's a long tail stacked up by small accounts.

Douyin reposter accounts were the main channel. 11-08 and 11-09 saw 5 videos concentrated over two days, with 11-09 being the biggest day, +34. Reach was modest. The top one was a comparison by "念昔-吃口大模型" (2,951 followers), 11-09 with the title "live2d sees a technical revolution? Who's hot, who's not?" It got 234 likes and 294 saves. "未来注释员" (5,744 followers) got 76 likes the same day. The rest got a few dozen likes each.

Two Xiaohongshu posts. "企鹅加德" (122 followers) posted "let you turn a portrait into a virtual avatar" at noon 11-07, less than 21 hours after the repo was created. It got 42 likes and 49 saves. The account runs a dedicated "XX's tallest mountain has arrived" series. "阿甘探AI" (10k followers) posted "one image becomes a 2D avatar #大肥鱼" 11-08, with 19 likes and 70 saves.

One small Bilibili post. "牧野v" 11-11 posted a tutorial with the same title, getting 189 views. Tagged #百度搭子 #办公就用dumate.

The author's name is doing the work (inference, strong evidence). HRuiCcc's GitHub account was registered only in Aug 22, opened 12 repos in three months, and has about 3,550 total stars. The previous repo, RuiC-PoseAtlas, was created 10-28 and gained two or three hundred stars a day from 10-30 to 11-01. It now has 1,586 stars and was even featured by a product-manager community. 梦随笔至 and 念昔, which promoted RuiC-2D, had both promoted PoseAtlas in Oct 30. On the day RuiC-2D launched, PoseAtlas was still gaining 76 stars a day. The video hadn't gained traction yet, but the new repo already had 40 stars, most likely from followers of his profile.

The "Big Fat Fish" meme. From 10 through November, DeepSeek Harness's "blue big fat fish / whale girl" was the hottest meme in China's AI circles. TMTPost and CBNData both covered it, and one Bilibili big-fat-fish video got 1.28M views. RuiC-2D's sample character is called Big Fat Fish 1, and several reposted videos used #大肥鱼.

Why these reposter accounts were so active (speculation). The Douyin batch consistently used #豆包工作, the Bilibili post used #百度搭子, and the title templates were identical, which looks like big-tech AI products running creator incentives while small accounts batch-produce assignments from new GitHub repos. I couldn't find the campaign rules or rewards, so this is unverified.

No trace of it on X, HN, Reddit, Product Hunt, or in media. Most of the forking accounts are new Chinese accounts with number-string names, and the audience is essentially domestic.

What actually made it take off

Looking closely, four things stacked on top of each other:

  1. A fixed group of reposter accounts. On Douyin and Xiaohongshu, some people pick new GitHub repos every day and make 30 to 60-second demos. "One image becomes a moving character" is visually immediate, which suits their audience best.
  2. The author is a name they recognize. The previous repo had just taken off, so the new one got posted the next day.
  3. The barrier dropped to something ordinary users can run. The original needs an AI coding assistant, which tripped up ordinary Chinese users at the first step. A Chinese UI, two commands to start, and a demo page you can click through made it easy for reposters to screen-record.
  4. The demo rode a trend. Big Fat Fish plus 21 ready-made characters meant the video had something appealing from the first frame.

The product-level differentiators (timeline, lip-sync, WebM) are icing on the cake. The core capability was already in the original, and the original was growing faster.

The demand is real: Live2D skins are expensive and hard to learn, and "face-free voiceover streaming" and "cheap VTuber setups" have had takers in China for a long time. Bilibili's beginner Live2D tutorials routinely pull millions of views.

If it were you

Verdict: C. You learn something, but following along isn't worth much. The playbook is clear and copyable: take a hot open-source tool from overseas, make a Chinese version with no barriers and examples built on local memes, add a web demo, and domestic repost accounts will do the rest. The result also shows the ceiling: 9 days, 160 stars, while the original had over two thousand in the same period. Single-image-to-avatar already has the original, a Chinese mirror, LivLiv, and a pile of PSD rigging tools. A third version will get even fewer stars. Part of its stars came from the author's earlier followers, which newcomers don't have.

Good fit for: people who move fast and can ship a usable Chinese version within a week or two of an overseas project going viral; people willing to keep shipping, two or three repos a month, and treat their GitHub profile like an account; people already making domestic AI tool content who know a few repost accounts.

Bad fit for: people hoping one project will reach a thousand stars (on this path a single project gets one or two hundred); people who want original technical work and dislike "localizing someone else's idea"; people who only publish to GitHub and skip demo videos and a web demo.

How to do it, ranked by how worthwhile:

  1. Copy the playbook, swap the topic. Watch for tools that Japanese or Western developers have just open-sourced, gaining hundreds of stars in a week, but only available in English or Japanese and requiring a lot of setup. Ship a Chinese version within two or three weeks, lower the barriers, and credit the original in the README.
  2. Build what the original in this niche didn't. For example: better automatic layering (complex illustrations, props, and hair currently need manual fixes), filling in the back and occluded parts, and exporting directly in formats Bilibili's Live Assistant and Douyin's live companion can use. These are the spots where users actually get stuck, and they're easier to demo.
  3. Don't make a third "single image to avatar editor." The original and RuiC-2D have already played the two cards of Chinese UI and no-assistant setup.

How to get users: on the day the repo goes up, launch a web demo you can open and play with right away, with a meme character on the homepage; record a 30-second vertical video whose first line states the result ("turn one picture into a talking virtual streamer"); titles can borrow the template repost accounts love, "the highest mountain in the XX world is here." DM a few Douyin and Xiaohongshu accounts that specialize in reposting new GitHub repos. They're short on material.

When the window closes: this playbook itself doesn't expire, but each topic's window lasts only a few weeks. Mesh Avatar was open-sourced in Oct 4, and RuiC-2D followed in Nov 6, already late. The Big Fat Fish meme's heat will probably last until late 2026 (inferred). To ride it, you need to move before the end of December.

What would show this read was wrong: (1) if RuiC-2D or a similar Chinese version passes 1,000 stars within a month, the ceiling on this domestic distribution channel is higher than estimated here, and you should push harder; (2) if the #豆包工作 and #百度搭子 activities that repost accounts tag wind down and video volume clearly drops, the free distribution was driven by platform incentives and the window is closing; (3) if someone builds single-image-to-avatar as a one-click web app or App and it takes off on Douyin, the real opportunity here is the product, not open-source stars.

Data and sources

GitHub: the HRuiCcc/RuiC-2D repo, README, commits and forks, daily stars (11-08 to 11-15); the author's profile and other repos; shinshin86/mesh-avatar-studio, livliv-dev/player, and similar repos (11-14). Douyin: videos found by searching "RuiC-2D" and the author's works list (likes, saves, shares; Douyin doesn't expose play counts). Xiaohongshu: notes from 企鹅加德, 阿甘探AI, and 青青王-AI, original clips reviewed. Bilibili: 牧野v 11-11 video. Big Fat Fish meme: 钛媒体 11-10, CBNData 11-12, 东云实验室 11-04. PoseAtlas: 人人都是产品经理 11-03. Anything marked "inferred", "estimated", or "unverified" is an inference or unverified.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
shinshin86/mesh-avatar-studioOpen source2026-10-042,366 stars / 252 forks, 7 days +460A prototype of RuiC-2D (its runtime and layering code reference it, and it can export compatible .mavatar files). Mesh Avatar requires Claude Code / Codex to build characters; RuiC-2D switches to local automatic layering with no agent needed, is in Chinese, and adds a timeline, lip-sync dubbing, WebM export, and custom parts.
LivLiv(livliv.ai + livliv-dev/player)WebOpen-source player 2026-11-04Player, 1,550 stars, single-day +11-07 712Cloud AI builds characters and can export PSD files for Live2D; RuiC-2D bundles its player and 21 characters directly (non-commercial use only)
shitagaki-lab/see-throughOpen source2026-03-315,670 starAcademic single-image anime layering: outputs layers only, no rigging, driving, or streaming
psd2live / still2rig-psd / rigby2d / live2d-halfbody-skill / Rev2DOpen source2026-07~1041–1,090 starStarts from layered PSDs, or is a skill for agents; not a complete single-image, end-to-end editor
Inochi2D / open-vt / PNGTuber-RemixOpen source2020~2024324–1,867 starTraditional open-source Live2D / PNGTuber alternatives; you have to draw the layered assets yourself
Live2D Cubism + VTube StudioOtherYearsIndustry standardRequires layered PSDs and professional rigging, paid, steep learning curve; RuiC-2D is free, single-image, rough but it moves
Avatars for Bilibili Live Assistant / Douyin live companionBig-company featuresAlready there—Only sells Live2D models; doesn't do single-image-to-character itself
DSH Big Fat Fish Live2D desktop pets and skins (DSH-L2D, jingyu-live2d-agent, free Big Fat Fish public skins on Bilibili, etc.)Open source2026-08~11Dozens to thousands of stars; public skin videos 440k viewsAll are desktop pets or plugins built on existing characters, not editors for making your own character
Where users come from
  • 查实 抖音开源搬运号:TikHub 搜到约 9 条:梦随笔至 11-07(101 赞,#豆包工作)、念昔-吃口大模型 11-09(234 赞、294 收藏)、未来注释员 11-09(76 赞)、两清/、云云的朵、织一束月色、三兆兆等;标题多套「2D虚拟形象界最高的山来了」;11-08~09 集中发布,对上 11-09 最大单日 +34
  • Verified Xiaohongshu repost accounts: 企鹅加德 11-07 (42 likes, 49 saves, posted less than 21 hours after the repo went up), 阿甘探AI 11-08 (19 likes, 70 saves, #大肥鱼). Both used Gemini on the original clip. The 941-like post about "turning standing art into animatable avatars" is about Mesh Avatar, so it doesn't count.
  • Verified Bilibili: 牧野v 11-11 ruic-2d tutorial, 189 views (#百度搭子)
  • Inferred Attention carried over from the author's previous work: the same repost accounts (梦随笔至, 念昔) promoted RuiC-PoseAtlas (1,586 stars, and it was also featured on 人人都是产品经理). 10-30 The repo already had 40 stars two days after launch, before the videos had gained traction.
  • Inferred platform creator campaign (Doubao Work / Baidu Dazi): reposted videos all tagged #豆包工作, #百度搭子 #办公就用dumate, same title template; campaign rules and rewards not found
  • Verified Trending meme (Big Fat Fish / DeepSeek Harness): the example character is called Big Fat Fish 1, and the repost videos are tagged #大肥鱼. 钛媒体 and CBNData reported on the meme's popularity in November.

HRuiCcc/RuiC-2D: a local editor that turns a single standing image into an animatable 2D virtual avatar

Research date 2026-11-14 (data card refreshed to 11-15). Repo github.com/HRuiCcc/RuiC-2D, web demo hruiccc.github.io/RuiC-2D/demo/, MIT license.

Bottom line

RuiC-2D is a domestic prolific indie developer's (HRuiCcc) take on the Japanese open-source project Mesh Avatar Studio, reusing its approach and runtime to make a local version that is "in Chinese, needs no coding agent, and comes with Big Fat Fish examples and 21 ready-made characters." Its stars came mainly from domestic short-video "open-source repost accounts": starting the second day after the repo went up, at least 9 Douyin videos, 2 Xiaohongshu posts, and 1 Bilibili videos introduced it, most titled with the "2D virtual avatar world's highest mountain is here" template, many tagged with platform activity hashtags like #豆包工作 and #百度搭子. The author's previous repo, RuiC-PoseAtlas (1,586 stars), had just gone viral with the same repost accounts, which are watching his new repo. The idea isn't original: the original author shinshin86 has already grown it (2,366 stars). Its rise comes from the author's consistent shipping, repost-account distribution, and the Big Fat Fish meme's heat; the product's differentiation is a secondary factor. Scale is still small: 9 days, 162 stars.

1. Data

  • GitHub (data card / our github repo_day): repo created 2026-11-06 07:14 UTC (Beijing time 15:14); 11-08 at first observation it already had 40 stars. Daily stars since: 11-09 74 (+34, largest single day), 11-10 84, 11-11 97, 11-12 105, 11-13 117, 11-14 144 (+27), 11-15 162; forks 27 (11-14 live API check: 166 stars / 28 forks). Fork-to-star ratio about 17%. Most fork accounts are 1766211376, 15838073269, 1574549397, the kind of numeric handles common among domestic newcomers, so the audience is mostly in China.
  • Commits: 11-06 first release, same day added a WeChat tip QR code and a GitHub Pages demo; 11-09 integrated the LivLiv character library, putting 21 characters on the homepage and the web demo; no new commits since. 0 open issues.
  • Author HRuiCcc: GitHub account 2026-08-22 newly registered, 89 followers, opened 12 repos in three months, roughly 3,550 total stars. These include RuiC-PoseAtlas 1,586 (created 10-28, grew 200–290 per day from 10-30~11-01), RuiC-card-skill 668, and RuiC-oc-ferry 505. All focus on AI toolchains and the DSH (DeepSeek Harness) ecosystem, and every README includes a WeChat tip QR code.
  • Searches across our X corpus returned nothing for RuiC-2D, HRuiCcc, mesh avatar studio, or livliv. Exa and Jina found no media coverage either.

2. Competitors

NameWhereLaunchedScaleHow it differs from RuiC-2D
shinshin86/mesh-avatar-studioOpen source2026-10-042,366 stars / 252 forks, 7 days +460; trilingual README in Japanese, English, and ChinesePrototype. The RuiC-2D README states that "the base runtime and layering code reference Mesh Avatar Studio," and it can export compatible .mavatar files. Mesh Avatar requires Claude Code / Codex to first split the image and build the character following agent-guide, then fine-tune in the editor. RuiC-2D switches to a local Python pipeline (MediaPipe + OpenCV) for automatic layering, with three manual points to fix inaccurate splits, so no coding agent is needed. It also adds a timeline, lip-synced dubbing, transparent WebM export, and custom parts like tails and sleeves.
moyuxianren01/mesh-avatar-studio-zhOpen source2026-10-045 starA Chinese mirror of Mesh Avatar with no new features
livliv-dev/player + livliv.aiOpen-source player / web SaaSOpen-source player 2026-11-041,550 stars, single-day +11-07 712; lots of reposts on IG and YouTubeLivLiv uses AI to build characters and can export layered PSDs for Live2D; its player only plays them. RuiC-2D bundles that player and 21 curated characters directly (characters are non-commercial use only).
shitagaki-lab/see-throughOpen source (SIGGRAPH 2026)2026-03-315,670 starAcademic single-image anime character layering: outputs layers only, no rigging, driving, or streaming
tsunehimatoi/psd2live, shinshin86/still2rig-psd, jaisurya-dev-art/rigby2d, bellisji233/live2d-halfbody-skill, RevStudio/Rev2D, umamoorg/umamoOpen source2026-07~101,090 / 165 / 248 / 54 / 41 / 200 starEither starts from layered PSDs, or is a skill or JSON rig for agents. No complete single-image + local editor + streaming end-to-end option.
Inochi2D, erodozer/open-vt, PNGTuber-RemixOpen source2020~20241,867 / 324 / 409 starTraditional open-source Live2D / PNGTuber alternatives; you have to draw the layered assets yourself
Live2D Cubism + VTube StudioCommercial softwareYearsThe industry standard; Live2D tutorials on Bilibili routinely pull millions of viewsRequires layered PSDs and professional rigging, steep learning curve; a custom skin costs thousands to tens of thousands of yuan. RuiC-2D positions itself as "free, single-image, rough but it moves."
Virtual avatars for Bilibili Live Assistant and Douyin live companionBig-company featuresAlready there—Only accepts Live2D format, doesn't convert single images to characters itself
Big Fish / DSH ecosystem Live2D desktop pets and skins (DSH-L2D 168, jingyu-live2d-agent 69, dsh-pet-indesktop 3k+, free Big Fish Live2D skin on Bilibili 440k views)open source / Bilibili2026-08~11dozens to thousands of starsAll are ready-made character desktop pets or plugins, not editors for making your own character

Bottom line: lots of competitors, and the closest one, Mesh Avatar Studio, launched a month earlier and is 15x bigger, and it was gaining even faster in the same window (11-10/11 in a single day, +53 / +228). RuiC-2D is a follower, not the originator.

3. Growth: where users came from

  1. Douyin "open-source repo re-upload" accounts (confirmed, main channel): searching "RuiC-2D" on TikHub finds at least 9 videos (timestamps in Beijing time):
    • 梦随笔至 (79 followers) 11-07 13:48 "2D virtual avatar's highest mountain has arrived ruic-2d tutorial #豆包 #豆包工作", 101 likes, 18 shares. This account also posted RuiC-PoseAtlas (221 likes) 10-30; its other posts are all GitHub new-repo recommendations.
    • Zhi Yi Shu Yue Se 11-08, San Zhao Zhao 11-08, Xiao Li Zao A 11-08, Liang Qing/ 11-09 (38 likes): the titles copy "The highest mountain in the 2D virtual avatar (world) has arrived" word for word, with the same tags #Doubao #DoubaoWork.
    • Future Annotator (5,744 followers) 11-09,76 likes.
    • Nianxi (2,951 followers) 11-09 "Live2d is about to have its tech revolution? Who's hot and who's trash?" 234 likes, 294 saves, 80 shares. This is a head-to-head review, and this account 10-30 also made PoseAtlas.
    • Yunyun's Duo 11-10 (51 likes, #大肥鱼), Cyberpunk Herdsman Field Fun 11-13, Little Way 11-14. The publish times of these videos line up with the star curve: 11-08~09 saw 5 videos go out in a cluster, and 11-09 is exactly when the biggest single-day jump of +34 happened. Each video's reach is small (the top one has 234 likes), so it's a long tail of small accounts stacking up.
  2. Xiaohongshu re-upload accounts (confirmed): Penguin Jiadel ( 122 followers, Guangdong) 11-07 12:13 "Turn your character art into a virtual avatar", 42 likes, 49 saves; he also runs an "XX highest mountain has arrived" series, and the posting time was less than 21 hours after the repo was created. Ganjian Tan AI ( 10k followers, Hunan) 11-08 "Turn one image into a 2D virtual avatar #BigFish", 19 likes, 70 saves. Both videos were checked with Gemini against the original footage; the visuals really are RuiC-2D and Big Fish 1. Separately, the Xiaohongshu search also surfaced a "turn character art into a movable avatar" post with 941 likes and 1,715 saves (Qing Qing Wang-AI, 11-13), but the original footage is about Mesh Avatar Studio, not RuiC-2D, so it shouldn't be credited to it.
  3. Bilibili (confirmed, very small reach): Mu Ye v 11-11 "2D virtual avatar world's highest mountain has arrived ruic-2d tutorial #OfficeWithDumate #BaiduBuddy", 189 views, 8 saves. The data card says "Bilibili is the main traffic source", which doesn't hold up.
  4. Platform campaign hashtags (inferred): the Douyin batch all carries #DoubaoWork, and the Bilibili one carries #BaiduBuddy #OfficeWithDumate, with the same title template. It looks like big-company AI products running creator incentives, with re-upload accounts batch-picking new GitHub repos to fulfill the tasks. The actual campaign rules and rewards were not found; unverified.
  5. The author's own back-to-back launches (inferred, strong evidence): the same batch of repost accounts (梦随笔至, 念昔) 10-30 had just pushed his RuiC-PoseAtlas. PoseAtlas also got picked up by 人人都是产品经理 (11-03), earlysignal, aoxox and other sites, which brought followers and attention to the HRuiCcc profile. RuiC-2D 11-06 launched right on the tail of PoseAtlas's buzz (11-06 was still gaining 76 a day). Two days after the repo went up, before the repost videos had gained traction, it already had about 40 stars, most likely from followers and profile visitors.
  6. The big fat fish meme (background, confirmed): 2026-10~11 DeepSeek Harness's "blue big fat fish / whale girl" is the hottest meme in China's AI circles. TMTPost 11-10, CBNData 11-12, and Dongyun Lab 11-04 have all covered it; a single Bilibili video titled "Deepseek big fat fish" hit 1.28M views; on GitHub, repos tagged dsh-plugin claim over 30k. RuiC-2D used "Big Fat Fish 1" as its example, and 云云的朵 and 阿甘探AI both tagged #BigFatFish.
  7. Not found: X, HN, Reddit, Product Hunt, ads, media coverage. The data card says nicheloom has 74 upvotes, but that site sits behind Cloudflare verification and couldn't be read; unverified. nicheloom itself is an aggregator that auto-indexes new GitHub repos, not an active promoter.

4. Why it took off

  • Steady supply from Chinese re-upload accounts: on Douyin, Xiaohongshu and Bilibili, a batch of accounts picks new GitHub repos every day and makes 30–60 second demo videos. "One image becomes a moving virtual avatar" is visual, easy to demo, and comes with Big Fish, so it suits them well. Someone posted within a day of creation, and five or six videos appeared within three days.
  • The author is a familiar name to re-upload accounts: PoseAtlas had just been pushed by the same accounts, and the high-star repos on the author's profile lent credibility to the new repo.
  • Chinese-first, low barrier: Mesh Avatar needs Node 22 plus Claude Code / Codex to build characters, a high bar for ordinary Chinese users. RuiC-2D has a Chinese interface, runs with pip install and npm start, needs no agent or API key, and has a GitHub Pages live demo page where you can click and watch Big Fish talk.
  • Example content rides a hot trend: the "Big Fish 1" example plus 21 ready-made LivLiv characters give the re-upload videos and the demo page something appealing right away.
  • Real demand: Live2D skins are expensive and hard to learn; "talking on camera without showing your face" and "low-cost VTuber" have had steady demand in China. Beginner Live2D tutorials on Bilibili reach millions of views.

5. Differentiation

Compared with Mesh Avatar Studio, RuiC-2D differs in that it needs no coding agent, a local pipeline auto-splits layers and you only add three points when the split is off; it's in Chinese; it adds a timeline and dubbing lip-sync, transparent WebM export and custom movable parts; it has a built-in Big Fish example and a LivLiv character library, plus a live demo page. Compared with Live2D, it doesn't need layered PSD files and it's free. But this differentiation is not the main reason it took off. The core capabilities (single-image mesh deformation, lip-sync, camera-driven motion, OBS) are all in Mesh Avatar, which grew faster in the same period; the most popular Xiaohongshu post, "turn character art into a movable avatar", was also about Mesh Avatar. The 2 stars RuiC-162D got mainly came from distribution through domestic re-upload accounts, attention from the author's previous project, and the Big Fish tag. "Chinese, no agent needed" just made the re-upload accounts more willing to use it for videos.

6. Visibility and copycats

  • Visibility: circulating in small circles. Only small Chinese short-video re-upload accounts have posted about it, with a top single video of 234 likes; no media, X or English-community coverage; it's free and open source, with only a WeChat tip QR code and no disclosed revenue.
  • Copycats: none so far. It is itself a follower or localizer of Mesh Avatar Studio.

7. What others can learn

  • An open-source tool that had just gone viral overseas (here, a Japanese author's Mesh Avatar), turned into a Chinese, agent-free version with domestic meme examples, can be picked up by Chinese re-upload accounts quickly. The window from overseas launch to Chinese hit was only a few weeks.
  • On Douyin and Xiaohongshu there is a steady set of "GitHub new repo" re-upload accounts, and big-company AI products' creator campaign hashtags (#DoubaoWork, #BaiduBuddy) are paying them incentives. An author who makes demos instantly understandable, plus a live demo page, can catch this free distribution.
  • Authors who launch back to back get watched by re-upload accounts: when one repo takes off, someone posts about the next one the next day.
  • Opportunity: "single image to movable avatar" is currently all open-source tools; there is no paid product for ordinary Chinese users as a one-click web app or app (LivLiv is an overseas web SaaS). The demand for low-cost virtual avatars on Bilibili and Douyin live streaming is still unmet.

8. Still unclear

  • Mapping stars to specific videos: the GitHub stargazers endpoint returned 404 this time, so there are no per-star timestamps; only rough daily comparisons were possible.
  • The rules and rewards for the #DoubaoWork and #BaiduBuddy campaigns were not found; whether the re-upload accounts received platform incentives is unverified.
  • The Douyin API doesn't return view counts (play_count is 0), so only likes and saves are visible.
  • The claim of 74 upvotes on nicheloom is unverified.
  • Not checked: private spread via QQ groups, WeChat groups, Zhihu, Juejin, and similar channels.

Sources

  • GitHub API: repos/HRuiCcc/RuiC-2D, users/HRuiCcc, the author's repo list, commits, forks; README and metadata for shinshin86/mesh-avatar-studio and livliv-dev/player.
  • Our GitHub repos (agent-im, repo_day / repo_profile, 2026-10-28~11-15).
  • TikHub: Douyin search/fetch_video_search_v2, web/fetch_user_post_videos; Xiaohongshu app_v2/search_notes, get_video_note_detail, get_user_info, get_user_posted_notes, get_note_comments. The three Xiaohongshu videos were checked against the original footage with ask-gemini.
  • Bilibili search API (via a Shanghai exit); Dongyun Lab dylab.org/archives/1138; TMTPost www.tmtpost.com/8198794.html; CBNData www.cbndata.com/information/296007; Rensheng Shi Chan Pin Jing Li www.woshipm.com/ai/6488179.html (PoseAtlas).
  • x-radar local X corpus: no matches.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.