A Claude Skill built from a million-follower Douyin account's method: the "Silicon Nonsense" AI animated series repo, 9 days, 360 stars
What it is A Skill loaded into Claude Code: fix two or three original characters, use character reference sheets plus a nine-panel storyboard, and hand it to Jimeng Seedance 2.5 to produce a 30-second dialogue-free animated short in one go.
Why it took off The author used a real million-follower Douyin account as the hook, first posting a breakdown that collected 2039 bookmarks, then posting the open-source release to harvest stars; it also coincided with Seedance 2.5 being able to output 30 seconds at once, and Chinese X was hyping Claude Code video Skills.
BHow you could follow it Don't build another generic comic-drama Skill. Copy the approach: find the next AI viral account nobody has broken down yet, get the workflow working, package it as a Skill, and publish it with two posts: "first the case breakdown, then the open-source release."
- Data (updated daily) · Repo created 2026-10-28 · 362 stars · 7 days +128
GitHub · stars Data 2026-10-30 → 2026-11-08 (10 days, 10 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
2026 year Oct 29 morning, the Chinese AI scene's X blogger 硅基废话 (@DDJCXX, 21.9k followers) posted: "Someone already got rich making AI animation with millions of fans." He broke down the Douyin account 呱呱历险记 (Gua Gua Adventures): a chubby frog and a little red snake. The first post went up in Sep 9. In under two months, 21 videos, and it hit 1M followers. The post ended: "Now writing it up as a Skill, free and open source once it's done." Three hours later he posted again: "I wrote the method up as a Skill and open-sourced it free," link in the comments. That repo is ai-anime-series, which by 11-06 had 360 stars and 78 forks. This piece covers what earned those stars, and whether copying the path for your own open-source project is worth it.
Two posts, one to build anticipation, one to cash it in
The breakdown post by 11-06 had 239k views, 1906 likes, and 2039 bookmarks. More bookmarks than likes means readers were thinking "save it, I'll follow this later," not "just passing through." The open-source post had 93.6k views and 837 bookmarks. The comment with the GitHub link alone had 12.3k views and 378 bookmarks.
The hook is real. As of 11-06, 呱呱历险记 had 1.17M Douyin followers (confirmed by searching Douyin users, 11-06). The whole post was about "an ordinary person made AI animation and hit a million fans in two months," and not one line was about the tech.
The repo was created late at night (Beijing time) on Oct 28, and the final push came ten minutes later. Nothing has changed since. So the repo was ready first, the post was the bait for demand, and the stars all came from the posts.
What it actually is
A Skill to install in Claude Code: one SKILL.md, 5 docs, 5 small Python scripts, plus one full example episode. The first screen of the README shows the finished video from that episode:
汤圆 and 油条: a round white hamster and a very long dachshund fight pigeons in a park over a pancake. 油条 stretches his body into a bridge, then a slingshot. 汤圆 curls into a ball and bowls into the flock. They grab the pancake back and it's down to a ring. 30 seconds, no dialogue, generated in one pass.
The method is called "storyboard to video" and has three steps:
- Make a reference sheet for each character (three views, expressions, poses, silhouette);
- Draw one black-and-white sketch storyboard, 30 seconds mapping to a 3×3 nine-panel grid;
- Feed the reference sheets, the storyboard, and a prompt that scripts shots and sound by the second into Jimeng's Seedance 2.5 "omni reference", and get a 30-second finished video in one go.
Claude writes the script (opening, conflict, twist, counterattack, punchline) and assembles the prompts. The scripts tile the grids, extract frames for checking, and join the 1-minute cut and compilations. Images and video are both made on the Jimeng canvas, which needs a Jimeng membership. Overseas users use Dreamina.
Why this month
两件事凑到了一起。一是 Seedance 2.5 能吃多张参考图、一次出 30 秒,角色一致性大体靠模型解决了,「设定图 + 故事板 → 一条成片」第一次走得通。二是 10 月中文 X 上「让 Claude Code 做视频的 Skill」正热,作者自己 11-01 发的一条同类仓库合集帖有 10.7 万浏览、3067 次收藏,里面的仓库多在几百到两三千星。
How the stars grew
| Date | Cumulative stars | Daily |
|---|---|---|
| 10-28 | 0 | Repo created |
| 10-29 | — | Breakdown post and open-source post went out in sequence |
| 10-30 | 154 | |
| 10-31 | 225 | +71, the biggest day |
| 11-01 to 11-04 | 234 → 266 | Single digits to low teens per day |
| 11-05 | 289 | +23 |
| 11-06 | 355 | +66, source not traced |
(Daily GitHub snapshots; 11-06 afternoon live check showed 360 stars.)
The first wave came from the two posts. After that, Skill directories like claudecodeagents, githubsignals, and repolode picked it up automatically, adding a bit of long tail. On 11-05 and 11-06 there were 89 more stars over two days, though the author had not posted about the repo those days. Likely a reshare or a Chinese platform repost, not verified. No media coverage, no big accounts sharing it, and nothing on Hacker News or Reddit.
The fork-to-star ratio is 22%, which is high. The likely reason is the install path: the README says to clone into ~/.claude/skills, while the post says "paste the link into Claude Code and let it install." Many people forked it on the spot. That doesn't mean 78 people are editing the code, and there's no way to know how many actually made a video and posted it.
Plenty of similar repos. This one is mid-tier.
| Who | What it is | Scale (11-06) | Compared with alternatives |
|---|---|---|---|
| zenstory-ai/drama-skills | Full Skill kits for short dramas and comics with dialogue | 3825 stars | Large and comprehensive; this one does only 30-second silent episodes |
| lixiaoxiao9888-create/manju-laoli-skill | Comic kits with 13 sub-skills | 1927 stars | Heavy; this one is a single narrow format |
| UnderleveledBuilder/awesome-short-drama-skills | Short-drama Skill directories | 1845 stars | Curated lists, not workflows |
| zhanghao123vip/ai-manju-skills | Full comic production pipelines across 21 genres | 586 stars | Published 3 days later, more stars |
| dexhunter/seedance2-skill and similar | General Seedance prompt and storyboard Skills | 765–4412 stars | Not locked to a format |
| Toonflow-app, dramaclaw, moyin-creator, etc. | Complete comic production platforms | 1453–20763 stars | Those are apps; this is just one Skill |
What sets it apart is its narrow format: two or three fixed original characters, 30 seconds, no dialogue, and one nine-panel storyboard that renders to a finished video in one pass. The no-dialogue constraint means the output works on Douyin, TikTok, and YouTube Shorts with no translation. Technically, though, it's prompt assembly plus a few ffmpeg scripts, and the bar is low. What earned the stars was the million-fan case and the author's account, not the workflow itself.
The author's other repos make this clear. Since September he has released roughly one "break down a hit account → open-source as a Skill" a week: ai-character-swap for outfit changes, 244 stars; ancient-rap-battle, with ancient figures rapping, 232 stars; Cleopatra, 53 stars. ai-science-series (alien science explainers) 49 stars and ai-fable-series (animal fables) 8 stars used the same formula re-skinned. Same person, same account, same writing style, but a gap of dozens of times in stars. The difference is how strong the hook case is.
What to learn from it
- Two-part posting: the first post breaks down a case and promises open source, to collect bookmarks; a few hours later the second delivers and turns bookmarks into stars. Have the repo ready beforehand.
- Borrow someone else's numbers: "a real account went from N videos to X0k followers" is verifiable, tied to money, and beats any technical highlight.
- Zero-friction install: package it as a Skill and tell readers "paste the link into Claude Code," so there's almost no friction between the post and the star.
- One account, one recipe, keep shipping: repos link to each other, the X profile points to GitHub, and the persona of "open-source author" builds up. The author's bio includes a business email.
If it were you
Verdict: worth trying, but don't copy the format; copy the approach. The "comic / Seedance Skill" topic already has dozens of repos, and a generic one won't earn stars. What earns them is the next hit account nobody has broken down yet.
Who it suits: people with a few thousand to tens of thousands of followers on X, Xiaohongshu, or Jike who already break down AI content cases; creators running AI video accounts with real workflows.
Who it doesn't suit: people with no account who expect the repo to grow stars on its own; people who want to add to this repo's format (more genres, a platform). That path already has projects with thousands of stars.
How to do it: each week, watch Douyin and TikTok for AI content accounts that hit hundreds of thousands of followers within two months. Pick one with a clear format (fixed characters, fixed length, fixed structure). Run one episode yourself, turn the character sheets, storyboards, prompts, and scripts into a Skill, and put your own finished video on the README's first screen. Or change direction: apply the same breakdown to TikTok English accounts and ship an English Skill for overseas Claude Code users. The Chinese-language repos haven't covered that side (inference).
How to get users: follow the two-part posting. The first post covers the case and numbers, with no link. After it collects bookmarks, post the open-source announcement with the link in the comments. Then submit the repo to Skill directories and ask Chinese GitHub recommendation accounts to share it.
Timing: within the next month or two (inference). "Claude Code makes video" is still fresh, and hit accounts keep appearing. Once Jimeng or Dreamina ships "fixed characters + storyboard to video" as an official template, this kind of Skill is only worth tutorial value.
Signs you read it wrong: the case you break down gets only a few hundred bookmarks; or posts get decent traffic but the repo has only dozens of stars, meaning readers are tired of this kind of repo; or Jimeng releases the same one-click template officially.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one zenstory-ai/drama-skills Open source 2026-07-16 3825 stars, +7 in 302 days Full dialogue short-drama/comic Skill; this one only does 30-second silent fixed-character episodes UnderleveledBuilder/awesome-short-drama-skills Open source 2026-10-13 1845 stars, +7 in 590 days A short-drama Skill directory, not a workflow lixiaoxiao9888-create/manju-laoli-skill Open source 2026-08-15 1927 stars A heavy comic kit with 13 sub-skills; this one is a single narrow format zhanghao123vip/ai-manju-skills Open source 2026-10-31 586 stars Full comic pipelines across 21 genres, broader coverage dexhunter/seedance2-skill and other Seedance prompt Skills Open source 2026-02~08 765–4412 stars General Seedance prompt/storyboard; this one locks fixed characters + a 9-panel storyboard for a 30-second output in one pass Toonflow-app / dramaclaw / moyin-creator / alibaba lumenx Open source 2026-01~03 1453–20763 stars A full comic-drama production platform; it's just a Skill that runs on the Jimeng canvas Jimeng / Dreamina canvas and agent Big-company features Already there ByteDance official It's essentially a how-to tutorial for Jimeng; the output still comes from Jimeng Elser AI / KomikoAI / Toonkit / MagicLight Web 2025–2026 Paid overseas SaaS Paid, English, general anime generation; this one is free, open source, Chinese, and tied to a real 1M-follower case - Where users come from
- Verified The author's own two-post thread on X (breakdown + open source): @DDJCXX 10-29 01:53 UTC breakdown post 239k views / 1906 likes / 2039 bookmarks; 04:51 UTC open-source post 93.6k views / 569 likes / 837 bookmarks; link in the replies 12.3k views (TikHub 2026-11-06)
- Verified Real hit case used as the hook: Douyin "Gua Gua Adventure" now has 1,167,237 followers (TikHub Douyin search 2026-11-06)
- Verified Listed by Skill directories and aggregator sites: claudecodeagents.com, repolode, githubsignals.io, gitofanton and whereismyai.space all have pages
- Verified Author's repo cross-linking: the README footer links to the ancient rap battle and Cleopatra repos; the X profile website points to GitHub; sibling repo ai-character-swap has 244 stars, ancient-rap-battle has 232 stars
- Inferred Source of the 11-05~06 second-wave stars: +89 stars over two days; the author posted nothing related, and no source was found
Why 10291124jcjc-dotcom/ai-anime-series took off
Channel: GitHub; research day 2026-11-06. Star counts come from our github database (repo_day) and a live GitHub API check; X data comes from a live TikHub pull of @DDJCXX's timeline.
Bottom line
This wasn't a "product" that broke out; it was a content playbook from a Chinese AI-circle X blogger. The author "Silicon Nonsense" (@DDJCXX, 21.9k followers) first posted a breakdown of "Gua Gua Adventure" on Douyin, which reached 21M followers with 1 videos (239k views, 2039 bookmarks), promising "I'll write it up as a free open-source Skill." 3 hours later he delivered and published the repo. The stars came from the traffic of those two posts; the repo itself is a thin prompt-plus-script workflow, and dozens of similar Seedance / comic-drama Skill repos already exist, most of them bigger.
What it is
- A Claude Code Skill (SKILL.md + 5 docs + 5 Python scripts + a set of examples) that teaches you to make "AI animated episodic series": lock 2 original characters → one character sheet per character → a 3×3 black-and-white pencil storyboard → feed the character sheets + storyboard + per-second shot/sound prompts into Jimeng Seedance 2.5 "Omni Reference", generating a 30-second dialogue-free finished video in one go.
- Example "Tangyuan and Youtiao" (a white hamster and a sausage dog fighting over a pancake); demo.mp4 is in the repo.
- Dependencies: a Jimeng membership (canvas + Seedance 2.5; Dreamina overseas); the scripts need Python + ffmpeg + Pillow. Scripts are MIT, everything else CC BY-NC 4.0.
- The repo was created 2026-10-28 14:31 UTC, last pushed 14:42, and has not been updated since. Source: GitHub API.
Data
| Date | star | fork | Daily adds |
|---|---|---|---|
| 10-30 | 154 | 40 | — |
| 10-31 | 225 | 53 | +71 |
| 11-01 | 234 | 54 | +9 |
| 11-02 | 238 | 54 | +4 |
| 11-03 | 255 | 58 | +17 |
| 11-04 | 266 | 63 | +11 |
| 11-05 | 289 | 64 | +23 |
| 11-06 | 355 | 77 | +66 |
Source: our github database repo_day; live GitHub API check 11-06 shows 360 stars / 78 forks. There are no observations before 10-30 (the repo was 9 days old, already at 154 stars when we started tracking).
- A third-party X post (@wangcha58403845,10-30) recorded "about 189 stars as of Oct 30"; gitofanton showed 218 stars when indexed; claudecodeagents.com showed about 233 stars when indexed.
- Fork/star ≈ 22%, which is high. A likely explanation: the README install method is to clone into ~/.claude/skills, and the X post told people to "drop the link into Claude Code and let it install," so many people forked it casually. That doesn't mean 77 people are actually modifying the code.
- The source of the +66 on 11-06 is unknown (the author's X posts that day didn't mention this repo; see "Still uncertain").
Where users come from (growth levers)
- The two-post "breakdown first, open-source promise second" thread on X (confirmed, main channel)
- Post on 2026-10-29 01:53 UTC 2115623148033101845: "Someone is already making a million followers with AI animation. The Douyin account 'Gua Gua Adventure'... posted its first video in September 9, and in under two months, with 21 videos, it reached 1M followers... I'm writing it up as a Skill now, and it'll be free and open source once it's done." — 239k views, 1906 likes, 201 reposts, 2039 bookmarks.
- Same day, 04:51 UTC post 2115667766703489355: "That AI animation account that hit a million followers in under two months: I wrote up how it's done as a Skill and open-sourced it for free... GitHub link in the replies 👇" — 93.6k views, 569 likes, 65 reposts, 837 bookmarks; the reply with the GitHub link got 12.3k views, 201 likes, 378 bookmarks.
- Source: live TikHub pull of @DDJCXX's timeline, 2026-11-06.
- These two posts correspond to about 225 stars across 10-29~10-31. Bookmarks far outnumber likes, which suggests the audience is side-hustlers and account-starters who save things to follow later.
- The hook is real: Douyin's "Gua Gua Adventure" currently has 1,167,237 followers (TikHub Douyin user search, 2026-11-06), so the author's "million followers" claim holds up. The hook is built on a "money / follower count + ordinary people can do it" narrative, not on technology.
- Auto-indexed by Skill directories and aggregator sites (confirmed, secondary): claudecodeagents.com/skills/ai-anime-series, repolode, githubsignals.io, gitofanton and whereismyai.space all have pages (found via Exa/Jina search). These sites index passively, which brings long-tail traffic.
- Small/repost accounts relaying it (confirmed, tiny volume): @wangcha58403845 (15 followers)10-30 posted a structured recap; accounts like this look like auto-generated summary feeds.
- The author's own account-network effect (inferred): the author only joined X in January 2026 and since September has been posting almost weekly "breakdown-of-a-viral-account → open-source Skill" repos: ai-character-swap (244 stars), ancient-rap-battle (232 stars), cleopatra-modern-life (53), ai-science-series (49), ai-fable-series (8). Each post links the older repos in the README and the profile's website field, funneling traffic between them.
- Not found: media coverage, big-account reposts (the relay posts found were all small), Reddit/HN, or ads. GitHub trending doesn't list repos younger than 60 days, so it wasn't a factor.
Why now
- Seedance 2.5's "omni-reference" mode (multiple reference images + generating 30 seconds in one pass) is what made "design sheets + grid storyboard → one finished video" possible. Character consistency is largely solved by the model. This workflow wasn't feasible before the model could do it.
- 2026 October on X: Skills that make Claude Code produce video are having a moment. A post by 11-01 rounding up "a batch of Claude Code video and animation Skills that appeared on GitHub this month" (107k views, 3067 bookmarks) lists repos mostly sitting between 370 and 2700 stars.
- The Douyin AI-content follower surge was hot (Zhihu 2026 article "'Ordinary people' gain 1.63M followers in a week... AI content is going wild on Douyin"), and the demand "I can build a million-follower animation account too" lived there.
Competitors
| Name | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| zenstory-ai/drama-skills | Open source | 2026-07-16 | 3825 stars, +7 in 302 days | A full short-drama / comic-drama Skill suite (scripts, character assets, storyboards, review), for long dialogue-driven series; it only does 30-second dialogue-free unit dramas |
| UnderleveledBuilder/awesome-short-drama-skills | Open source | 2026-10-13 | 1845 stars, +7 in 590 days | A curated directory of short-drama Skills that collects rather than builds a workflow |
| lixiaoxiao9888-create/manju-laoli-skill | Open source | 2026-08-15 | 1927 stars | A heavy suite of 13 sub-skills from the comic-drama creator Lao Li; it covers one narrow format |
| zhanghao123vip/ai-manju-skills | Open source | 2026-10-31 | 586 stars | A full comic-drama pipeline Skill across 21 genres, released 3 days later and with more stars |
| dexhunter/seedance2-skill, songguoxs/seedance-prompt-skill, liangdabiao/Seedance2-Storyboard-Generator, liyue-aigc/seedance-2-5-video-director | Open source | 2026-02~08 | 4412 / 2987 / 2773 / 765 stars | General-purpose Seedance prompt/storyboard Skills; this one locks the format to "fixed characters + 9-grid + 30 seconds in one pass" |
| HBAI-Ltd/Toonflow-app, dramaclaw, moyin-creator, alibaba/lumenx, and others | Open-source apps | 2026-01~03 | 20k / 7250 / 4824 / 1453 stars | A complete comic-drama production platform (canvas + agent); this one is just a Skill that relies on Jimeng's official canvas for output |
| Jimeng (Dreamina)'s own canvas and agent | Big-company features | Already there | ByteDance official | Its workflow runs inside Jimeng; it's essentially a "how-to" for Jimeng |
| Elser AI, KomikoAI, Toonkit, MagicLight, AnimeShorts | Web | 2025–2026 | Overseas SaaS | All focus on "character-consistent AI anime" as paid SaaS; this one is free, open source, Chinese, and tied to a real viral case |
Source: our github database repo_profile (2026-11-06); Jina search results from each site.
Differentiation
- Ultra-narrow format: skip "short drama / manhua" and do episodic shorts with a fixed cast of two or three original characters, 30 seconds each, no dialogue, and a beat structure of setup → conflict → twist → comeback → punchline. No dialogue means it travels across languages, so it can run on Douyin, TikTok and YouTube Shorts at the same time.
- Storyboard straight to video: a black-and-white 3×3 sketch storyboard plus character design sheets serve as Seedance references, producing 30 seconds in one pass instead of generating shot by shot and editing. This is the key technique the author extracted from "Gua Gua Adventure"; it saves rerolls and editing.
- Benchmarked against a real million-follower account, not an abstract "AI short-drama tool."
- Are these differences the reason it took off? Partly: the narrow format plus a real case makes the X story credible and persuasive. But the technical barrier is very low (prompt assembly + ffmpeg scripts), there are many similar and bigger repos, and at 355 stars it's only mid-pack among its peers. What actually drew the stars was the author's distribution (post hooks + a 20k-follower account); the repo was the payoff the posts promised.
Copycats already out
- The author is batch-copying the same recipe: ai-science-series (10-30, alien science short dramas, 49 stars) and ai-fable-series (11-05, animal fable short dramas, 8 stars). Both are re-skins of "fixed characters + Jimeng storyboard prompts," but their star counts are far below this repo's, which shows that the strength of the hook case determines the outcome.
- Third-party check: I found no repository copying "Gagaga Adventure / episodic drama" directly. The zhanghao123vip/ai-manju-skills repo (586 stars) that appeared in 10-31 is in a similar direction but isn't a copy.
How many people have seen it
Word is spreading in small circles: in the Chinese AI-tools crowd on X, the two posts total about 330k views; several Skill directory sites have indexed it; there's no media coverage and no big-account endorsement. The author is himself a "repost/breakdown" blogger in this circle, and his posts are one of its information sources.
What to borrow
- A two-step pattern: a breakdown post with an open-source promise first to accumulate bookmarks (2039), then a second post to harvest stars a few hours later.
- The hook is a verifiable, money-linked number, like "a real account's N videos grew to X0K followers", not a tech highlight.
- Package the repo as a "Skill you can install into Claude Code", plus a zero-friction "just paste the link into Claude Code and it installs itself" instruction, to cut the friction from post to star/fork.
- One account keeps churning out repos from the same formula, the repos link to each other, the profile website points to GitHub, and it builds an "open-source author" persona. That then lets the X account land sponsored deals (the bio has a business email, and he complains that X ad partnerships only pay referral commissions 11-05).
- Takeaway for product teams: now that Seedance 2.5-style multi-reference-image generation exists, there is room for "narrow format + fixed character" template tools; but the moat is zero, and whoever has distribution wins.
Still unclear
- Where the +89 stars between 11-05 and 11-06 came from: the author posted nothing about this repo in those two days; it could be reposts by others, directory sites, or Douyin/Xiaohongshu re-uploads, not verified. This time the GitHub stargazers API returned 404/needs auth, so stars could not be split by time.
- How many people actually made and published videos with it is unverified.
- Whether the author of "Gua Gua's Adventure" really used the same workflow (storyboard straight to video) is the author's inference and unverified; only his Douyin following of 1.17M was verified.
- The author says they are in New York (X location); the account was registered 2026-01; the real identity is unverified.