An old 2.6k-star repo paves the way for a new one: a hand-drawn narration Skill got 1066 stars in 25 days, while another with the same tech has only 107
What it is A Skill you install into Claude Code or Codex: give it a topic or an article, and it produces a narrated hand-drawn explainer video with lines drawn stroke by stroke and Chinese characters written in correct stroke order.
Why it took off The author's earlier 2.6k-star repo and his X followers gave it its first audience. Finished videos with voiceover stand out at a glance from silent competitors. It was also pitched as a "platform traffic bonus" tool for self-media creators, right as Douyin and Xiaohongshu's AI knowledge-video push was heating up in October.
CHow you could follow Don't build another hand-drawn video Skill. Copy its two-step approach: first use a lightweight Skill to build an audience, then ship an end-to-end finished-video Skill. Pick a video format that isn't crowded yet, show a finished video on the README's first screen, and use it to run your own content account.
- Data (updated daily) · Repo created 2026-10-12 · 1,084 stars · 7 days +110
GitHub · stars Data 2026-10-14 → 2026-11-08 (26 days, 25 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
threerocks/sweety-handdraw-video is a Skill installed into Claude Code or Codex: give it a topic or an article, and it produces a narrated hand-drawn explainer video with lines drawn stroke by stroke and Chinese characters written in stroke order. Created in Oct 12; by 11-06 it had 1066 stars and 177 forks. It took off partly because the finished videos really work, and partly because the author already had an audience. But its category already had seven or eight players before it launched. Below: what it is, where the stars came from, why it's this one, and whether you should still build another.
"All rise" and a 4-minute 51-second sample video
On Oct 13 2026, at 1:30 PM Beijing time, "Liu Laoban" (@leo_xiaolei) on X posted:
"All rise!! V1 of the 'hand-drawn narration Skill' has been committed to the repo!! Showcase: men have an extremely boring but irrepressible instinct: the urge to shoot."
Under the post was a 4-minute 51-second hand-drawn explainer made with this Skill. The post got 12k views and 180 bookmarks. That evening he posted a voiceover sample with a single line: "Listen to it, can you tell real from fake?" That one got 15.5k views. The same day he also posted a long Codex install tutorial on X, and on Juejin published "Major open source release! The strongest hand-drawn video pipeline on the market right now," with his own watermark on the article's images.
His X bio reads "AI R&D at a leading security company | instructor." He's not new: back in June he had open-sourced a hand-drawn-style prompt Skill called hand-drawn-styles, which already had 2,621 stars. When the new repo launched, that older repo was bundled as one of its "five built-in dependencies."
What it is
A pipeline from "one topic sentence" to "a video you can publish directly" splits into two halves:
- The Agent does the creative work: writing the narration script, designing storyboards, generating illustrations.
- Local programs do the craft work: Python, Pillow and FFmpeg first block out the forms, then draw lines part by part, erase the construction lines, color with markers from light to dark, write characters stroke by stroke in standard stroke order, and finally sync to the voiceover and export. No external image-to-video model is called.
- Audio first: The full narration is generated first, each segment is checked for hum, background noise and clipping, then the visuals and subtitles are aligned to timestamps, calculating how many seconds each frame can last and the maximum number of strokes. Voice can come from local VoxCPM2 or be routed to Volcano or Gemini TTS.
- Partial rework: Subtitles, audio and images are stored separately, so changing one subtitle line doesn't require regenerating images.
Install with one command: npx skills add threerocks/sweety-handdraw-video. The skills directory skills.sh shows it has been installed 484 times in total (11-06), a closer measure of real usage than stars. Actually producing a finished video requires local image generation and a TTS setup, so the bar is not low.
How the stars climbed, step by step
| Date | star | What happened |
|---|---|---|
| 10-12 | 0 | Repo created, v1.0.0 |
| 10-13 | — | The V1 post, the voiceover sample post, the Codex tutorial and the Juejin launch all landed on this day |
| 10-15 | Single day +230 | The biggest day so far; the same day an independent blog review appeared, and SegmentFault's "GitHub Trending Today" (10-17) and others reposted it afterward |
| 10-16 | 501 | |
| 10-19 | 683 | v1.4; the author published a long post, "The traffic harvester 'hand-drawn narration video Skill'" |
| 10-27 → 10-28 | 875 → 952 | v2.0, +77 in one day |
| 11-03 | 991 | v3.0.0 to 3.0.2 |
| 11-06 | 1066 | Last 7 days +102, first 7 days +162 |
(Daily GitHub star counts, 10-16 through 11-06.)
The shape is clear: the first week rocketed to five or six hundred, then each major release brought another step up. The 25 days produced 12 releases, with v1.4, v2.0 and v3.0 each paired with a long post and a few samples. Growth is slowing but hasn't stopped. Who drove the +230 on 10-15 can't be determined from public information.
Why this one got noticed
First, it had a ready-made foundation. The 2.6k-star predecessor, 471 followers on the author's GitHub, and roughly 3,900 stars across all his repos. People were looking on day one, so there was no need to find an audience from zero.
Second, the sample videos were the ad. The README's first screen shows a finished video, such as "The Snake That Died of Anger" and "The Way Happiness Imprisons You"; on X he kept posting short videos made with it, such as "A Takeout Order's Life," explicitly saying he was "testing short-video traffic." The content the tool produces goes out to chase traffic, and that traffic then promotes the tool.
Third, it sold a developer tool to self-media creators. The v1.4 long post was written as an SEO article, with a "Meta Description" and "core keywords" at the top. The body lists, one by one, Xiaohongshu's "AI tells it for me" creator fund, Douyin's #vibe knowledge contest, YouTube long-video revenue share and X creator revenue share, and concludes with "helping you grab the mid-to-long video traffic dividend." Readers weren't just programmers but also people who wanted to run faceless AI knowledge channels, a far larger group.
Fourth, the timing was right. Douyin's #vibe knowledge contest in October and Xiaohongshu's support for AI knowledge content were both hot, and plenty of people were looking for a tool that makes hand-drawn explainer videos in one click.
Fifth, it's instantly distinguishable from peers. Peers mostly output silent visuals or just prompts. This one outputs complete videos with sound, narrative, and the look of a human drawing. The line "can you tell real from fake" was aimed squarely at that.
The crowd of peers
| Project | Launched | star(11-06) | How it differs |
|---|---|---|---|
| lemomo-ai/lemo-opuscar | 09-26 | 12.6k, +3,233 in the last 7 days | General "Agent makes video" tools, 66 styles, whiteboard is just one of them |
| geeklee/srt-whiteboard-animation | 07-27 | 5,417, +246 in the last 7 days | Category leader; takes subtitles as input and outputs only whiteboard-marker animation, no script writing or voiceover |
| gnipbao/story-to-handdrawn-video | 07-21 | 2,380 | 3:4 vertical picture-book animation, many art styles, finished videos are silent; recommended by GitHubDaily |
| trustfuture/simon-skills | 09-20 | 520 | Faceless-account bundle, Excalidraw style, with fact-checking |
| hi-nikola/hand-drawn-explainer-video-nikola | 09-02 | 459 | Programmatic animation, can output prompts only |
| gnipbao/codex-whiteboard-video-skill | 06-26 | 346 | Turns existing images into whiteboard animations |
| kaomei/hand-drawn-video-prompts | 07-27 | 302 | Storyboards and prompts only |
| liangdabiao/whiteboard-explainer | 09-09 | 107 | Pixel-level strokes, frame-level audio-visual sync. The approach is closest to sweety's, but nobody is promoting it. |
(GitHub, 11-06.)
The last row is the one to watch. liangdabiao's technical route is closest to sweety's, and it launched a month earlier, yet today it has only 107 stars. Similar tech; the difference is distribution. By stars, sweety still ranks only mid-pack in this group. Before it launched, a Cnblogs post (10-12) and an X account (9-27) had already shared a "hand-drawn video Skill bundle." In late October, a batch of similar small repos with 0 stars each appeared. The author is also branching out: 11-03 newly opened a crayon-style state-diagram short-video Skill, which hit 58 stars in three days.
How many people know about it
A small circle. Juejin, CSDN, SegmentFault, the Tencent Cloud developer community, and Zhihu bundles all repost it, and several Skill directories such as skills.sh list it. The author's single posts get 1 to 15k views. I found no English-language community, mainstream outlet, or major account covering it. GitHubDaily promoted a competitor instead. I could not verify whether anyone used it to make a hit video on Douyin or Xiaohongshu.
If it were you
C: You can learn a playbook from it, but building another hand-drawn video Skill isn't worth it. Since 6, the field has grown to seven or eight players, with the top four or five at several thousand stars. A project with the same technique and no audience has only about a hundred stars. A large share of sweety's stars came from the author's 2.6k-star legacy repo and the followers already on X. You don't get those.
Who it suits: developers who already have a project with a few hundred to a few thousand stars, or a batch of readers on X or Juejin, and want to monetize that audience again with a new project; people who understand a specific content niche (knowledge accounts, product-selling accounts, children's picture books) and can tell the tool as a story about the platform traffic wave.
Who it doesn't suit: people starting from zero with no audience who hope to gain stars with "yet another hand-drawn / whiteboard video Skill" (see liangdabiao's 107 stars); people who only want to tinker with technical details and won't make content and post samples; people who want to copy the same category into the English market (I didn't see this category take off in English. Demand is unverified, not checked).
How to do it: three ways to follow, best to worst:
- Copy its two-stage structure, not its subject matter: first release a lightweight Skill (art style, prompts, templates) to build stars and followers, then after a month or two release the heavy end-to-end Skill, with the earlier work serving as its dependency and entry point.
- Pick a format that produces ready-to-post output: the same "Agent writes script + local program renders + voice first" approach, applied to a form that isn't crowded yet, such as the crayon state diagrams the author is testing, or data-chart explainers, map animations, and reading-card short videos. Before you start, search GitHub. If there are five or more similar projects, switch.
- Fix what it does poorly: its barrier is high, requiring local image generation and a TTS environment. A lightweight version that runs right after install with no environment setup, or a post-processing Skill that matches Xiaohongshu and Douyin dimensions and subtitle conventions, can both ride the category's momentum.
How to get users: put the finished video on the README's first screen, not a feature table; post interactive samples like "guess real or fake" on X, and run a content account with the tool to generate traffic; syndicate each article to Juejin, CSDN, Zhihu, and SegmentFault; submit to directories like skills.sh so it installs with one command; ship a major version every one to two weeks, each with a long post and new samples.
Window closes: this heat rides on Douyin and Xiaohongshu's AI knowledge-video support programs. These usually run quarterly, so ship before Q1 2027 (inferred).
Signs you've misread it: (1) if sweety ships v4.0 with a long post and still can't add 30 stars in a day, even people with an audience can't push it, and the category is cooling; (2) if an all-in-one project like lemo-opuscar makes "voiced hand-drawn" the default option, the room for single-feature Skills gets eaten; (3) conversely, if a new hand-drawn Skill with no predecessor and an author with no following passes 1000 stars in a month, demand is bigger than estimated here, and you should move faster.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one geeklee/srt-whiteboard-animation Open source 2026-07-27 5417★, 7-day +246 Takes SRT subtitles as input, makes only whiteboard-pen animations, no scripting or voice-over gnipbao/story-to-handdrawn-video Open source 2026-07-21 2380★, promoted by GitHubDaily Story picture-book 3:4 animation, many art styles, silent output trustfuture/simon-skills (whiteboard-video) Open source 2026-09-20 520★ Faceless-account bundle, Excalidraw style, with fact-checking and per-platform copy hi-nikola/hand-drawn-explainer-video-nikola Open source 2026-09-02 459★ Programmatic animation "Grotesque Little Black," can output prompts only gnipbao/codex-whiteboard-video-skill Open source 2026-06-26 346★ Turns existing images/SVGs into whiteboard animations, no content production kaomei/hand-drawn-video-prompts Open source 2026-07-27 302★ Outputs storyboards and image/video prompts only, leaving generation to Flow/Nano Banana liangdabiao/whiteboard-explainer Open source 2026-09-09 107★ Technical approach is close (pixel-level strokes, frame-level audio-visual sync), but no promotion lemomo-ai/lemo-opuscar Open source 2026-09-26 12.6k★, 7-day +3233 An all-in-one Agent video project with 66 visual styles; whiteboard is just one of them VideoScribe / Doodly / Vyond Other Years ago Commercial software Timelines must be built by hand. You can't go from a topic to a finished video in one step - Where users come from
- Verified Author's X account + audience from earlier work: the predecessor hand-drawn-styles (2621★) is a built-in dependency; a V1 post got 12.2K views/180 bookmarks, and the voiced "guess real or fake" post got 15.5K views
- Verified Using the tool's output to run a content account: the author keeps posting short videos made with it on X (e.g., "A Takeout Meal's Life") and openly says they're testing traffic
- Verified Author posts articles and SEO pieces to Chinese tech sites: Juejin and CSDN "Major open source release! The strongest hand-drawn video pipeline on the market right now" (10-13, with the author's watermark); V1.4 X long post with meta description and keywords
- Verified Third-party reposts and Skill directories: mushroom review 10-15, SegmentFault daily recommendation 10-17, Tencent Cloud, Zhihu bundle, skills.sh/360 skill hub/awesome-claude-video-skills
- Verified Riding platform trend language: the V1.4 long post cites Xiaohongshu's "AI replaces me speaking," Douyin's #vibe knowledge awards, and YouTube/X revenue sharing, positioning it as a "traffic harvester"
- Inferred Big-account endorsement: found no big account such as GitHubDaily specifically recommending it (GitHubDaily promoted a competitor)
Why threerocks/sweety-handdraw-video (Hand-drawn Talking-Head.Skill) took off
Research date 2026-11-06. Sources: our GitHub library (repo_profile / data cards, daily records from 2026-10-16 to 11-06), GitHub API, the author's X posts (read via r.jina.ai), and Jina/Brave/Exa web search. The local x-radar corpus returns nothing for this project (0 hits); all X evidence comes from reading the post pages directly.
Bottom line
This is an Agent Skill that plugs into Claude Code / Codex. Give it a topic or an article and it produces a hand-drawn explainer video with voiceover, strokes drawn line by line, and characters written one by one. It takes off because three things stack up: the author already has a hand-drawn-styles predecessor with 2.6k stars, which brings ready-made followers and trust; the author posts sample cuts on X and ships versions often (12 releases in 25 days), while also syncing articles to Chinese tech sites like Juejin, CSDN, Tencent Cloud and Zhihu; and it rides the AI science-explainer video wave pushed by October's Douyin "#vibe Knowledge Awards" and Xiaohongshu's "AI speaks for me". Its edge is "finished video with voice, drawn like a real person is drawing" (voiceover first, then drawing on a timeline, tracing by part, characters written in stroke order, with a QA gate), while most rivals ship silent visuals or only prompts. The difference is clear but not exclusive: the niche already has seven or eight similar projects.
1. What it is
- The repo was created on 2026-10-12, MIT license, by authors threerocks / LeiLiu, X account @leo_xiaolei ("Boss Liu"). As of 2026-11-06 it has 1066 stars and 177 forks; latest version v3.0.2 (11-03).
- The division of labor: the Agent writes the voice-over, designs storyboards, and generates illustrations; local Python/Pillow/SciPy/FFmpeg handles line-by-line strokes, marker coloring, stroke-order handwriting for Chinese characters, voice alignment, and export. No external image-to-video model is used.
- There are several voice options: local VoxCPM2 (MLX), a local VoiceStudio service, ready-made full narration, or hookups to Volcano / Gemini TTS. Every voice track must pass checks for hum, noise floor, and clipping.
- Install with
npx skills add threerocks/sweety-handdraw-video. skills.sh shows 484 cumulative installs (2026-11-06 read), a closer measure of actual usage than stars.
2. Growth data
| Date | star | Notes |
|---|---|---|
| 10-12 | 0 | Repo created, v1.0.0 |
| 10-13 | — | The author posted V1 on X (12.2K views, 180 bookmarks), and the same day published a Codex install tutorial long post and a Juejin "major open source release" article. |
| 10-15 | Single-day +230 (the largest single-day gain in the library) | The same day, blog.mushroom.cv published a review. SegmentFault (10-17) and Baffox (10-19) reposted it afterward. |
| 10-16 | 501 | First day recorded in the library |
| 10-19 | 683 | v1.4 released; the author posted the X long post "Traffic harvester 'Hand-drawn Talking-Head Video.Skill'" |
| 10-27→10-28 | 875→952 | v2.0 released, +77 in a single day |
| 11-03 | 991 | v3.0.0~3.0.2 |
| 11-06 | 1066 | Last 7 days +102, first 7 days +162 |
Source: data card / GitHub library repo_profile (2026-11-06). The curve shape is "burst in the first week, then a jump with each major release." Growth is slowing but hasn't stopped.
3. Competitors (doing the same thing)
All are Chinese "hand-drawn / whiteboard explainer video Skills" that appeared between 2026 6 and October. Star counts are from the GitHub API (2026-11-06); the last 7 days of data come from the GitHub library.
| Project | Launched | Scale | How it differs from sweety |
|---|---|---|---|
| geeklee/srt-whiteboard-animation | 2026-07-27 | 5417★, 7-day +246 | Category leader. Input is SRT subtitles, output is cream-paper whiteboard pen animation; it doesn't handle scripting or voice-over |
| gnipbao/story-to-handdrawn-video(@GeekCatX) | 2026-07-21 | 2380★ | Storytelling, 3:4 vertical picture-book animation, nearly 300 art styles, rendered with Remotion, silent output; promoted by GitHubDaily 10-30 |
| gnipbao/codex-whiteboard-video-skill | 2026-06-26 | 346★ | Turns images, SVGs, and line art into whiteboard animations, 30 styles |
| trustfuture/simon-skills(whiteboard-video) | 2026-09-20 | 520★ | Skill bundle for faceless accounts, Excalidraw style, with fact-checking and per-platform copy |
| hi-nikola/hand-drawn-explainer-video-nikola | 2026-09-02 | 459★ | Codex Skill, "Grotesque Little Black" programmatic animation, can output prompts only |
| kaomei/hand-drawn-video-prompts | 2026-07-27 | 302★ | Outputs storyboards and image/video prompts only, leaving generation to Flow / Nano Banana |
| liangdabiao/whiteboard-explainer | 2026-09-09 | 107★ | Pixel-level strokes, frame-level audio-visual sync; the approach is closest to sweety's, but it has almost no distribution |
| lemomo-ai/lemo-opuscar | 2026-09-26 | 12.6k★, 7-day +3233 | A bigger "Agent-made video" all-in-one project with 66 visual styles; whiteboard is just one of them (Zhihu's "9 god-tier projects riding the vibe knowledge awards" puts it first) |
| VideoScribe / Doodly / Vyond, etc. | Years ago | Commercial software | Manually building a timeline; sweety goes from a one-line topic to a finished video, fully automated. |
cnblogs(10-12) and @zhongying14 (Mazhiang AI Lab, 9-27) both posted "hand-drawn video Skill roundups" — the niche was already hot before sweety launched.
4. Where users came from (growth tactics)
- 作者自己的 X 账号 + 前作攒下的受众(已确认)。同一作者的 hand-drawn-styles(手绘画风提示词 Skill,6 月建库)已经有 2621★,sweety 把它打包成"五项内置依赖"之一,README 链回作者 X。V1 帖 12.2K 浏览、180 收藏;10-13 配音样片帖 15.5K 浏览,文案是「你们听听声音,能分出来真假吗?」。
- Using sample videos to spread (confirmed). The README's first screen is finished videos like "The Snake That Died of Anger" and "The Way Happiness Imprisons You." On X the author keeps posting short videos made with it, like "A Takeout Order's Life" and a Mbappé topic, openly saying they're "testing short-video traffic." They use the tool's own output to run a content account, so the work itself is the ad.
- The author publishes to Chinese tech sites directly (confirmed). Juejin: "Major open source release! The strongest hand-drawn video pipeline on the market right now" (10-13, with the image watermark XiaoLei_Liu, i.e. the author). CSDN / AtomGit carry the same piece. The X long-form v1.4 version is written as SEO copy, with a "Meta Description / core keywords" block.
- Third-party reposts and indexing (confirmed, modest reach): a blog.mushroom.cv review (10-15), SegmentFault's "GitHub Trending Today" (10-17), Tencent Cloud developer community, iencoo, a Zhihu roundup, skills.sh / 360 skill hub / tool.lu / hotskills and other Skill directories, and zhuyansen/awesome-claude-video-skills. I found no evidence of big accounts like GitHubDaily or Baoyu specifically promoting it (GitHubDaily promoted the competitor story-to-handdrawn-video).
- Riding platform trends (confirmed as the author's own pitch). The v1.4 long post uses Xiaohongshu's "AI tells it for you" creator fund, Douyin's #vibe Knowledge Awards, YouTube long-video revenue share, and X creator payouts as selling points, positioning it as a "traffic harvester / monetizing mid-to-long video traffic." The target user is faceless AI-knowledge creators, not just developers.
- High-frequency iteration that keeps generating buzz (confirmed): 25 releases in 12 days; v1.4 / v2.0 / v3.0 each came with an X long post and sample videos, and each gave the star curve a small step up (10-28 in a single day, +77).
- Following hot host platforms (confirmed): first support for Codex's built-in image generation, then Claude Code, Tencent WorkBuddy, and DeepSeek, with a post for each one added.
5. Why it took off
- An existing foundation: the author's earlier 2.6k★ art-style Skill already earned trust, so the new project had an audience from day one (471 GitHub followers; about 3.9k★ across all the author's repos).
- The sample videos sell it: narrated, complete story videos from 49 seconds to 2 minutes, with sound. The "can you tell it's real?" voiceover post got 15.5K views. Competitors mostly ship silent footage or just prompts; sweety looks at a glance like a video you can post as-is.
- Good timing: in October, Douyin's #vibe Knowledge Awards and Xiaohongshu's support for AI knowledge content were both running hot, and a crowd of creators was looking for a one-click hand-drawn explainer tool.
- Good at promotional writing: the author packages a technical project as a "traffic-monetization tool," distributes the same piece across platforms, and Skill directory listings bring long-tail traffic.
- Frequent updates: each major version gives people a new reason to come back and share again.
6. Differentiation
- Narrated finished videos: audio comes first. It generates the full narration, runs quality checks per segment, then schedules visuals and subtitles against timestamps. The story-to-handdrawn-video output is silent; srt-whiteboard needs subtitles first; kaomei only outputs prompts.
- Looks like a human drawing: it sketches the shape first, traces the lines part by part, erases the construction lines, colors with markers from light to dark, and writes Chinese characters in standard stroke order. It also works out from the narration length how many seconds each drawing gets and how many strokes it can take.
- Local rework is possible: subtitles, audio, and images are stored separately, so fixing a subtitle doesn't require regenerating images.
- Can it hold its position? Partly. "Narration + quality checks + directing method" makes it look more finished than silent competitors, which is one reason it's pulling more stars in a crowded field. But its star count clearly also rides on the author's audience and promotion. On stars it still ranks behind srt-whiteboard (5.4k) and story-to-handdrawn (2.4k). Technically, liangdabiao/whiteboard-explainer also does frame-level audio-video alignment, but nobody promotes it and it has only 107★. Same tech without distribution doesn't get traction.
7. How many people know about it
It's spreading in niche circles. It has a batch of reposts and directory listings in the Chinese AI Skill / creator-tool scene, and the author's posts have 1~15k views each. I found no English big accounts, mainstream media, or X influencers writing about it, and it's not in our x-radar corpus. It's an open source project, so no revenue is public.
8. Copycats
This niche is already crowded: since 2026-10-21, a bunch of similar small repos have shown up, like why3316-g/Why-HandDrawn-Video, xiao-steel/ai-srt, and wbin888/whiteboard-sb (11-04), each at 0 stars. liangdabiao also built story-handdrawn-video. There's no direct fork-and-rename evidence of copying, but "hand-drawn talking-head Skill" is already a commoditized idea. The author is branching out too: 11-03 opened sweety-statechart-video (crayon-style state-diagram shorts, 58★).
9. What to borrow
- Use a lightweight Skill (art-style prompts) to build an audience first, then ship the heavyweight end-to-end Skill; the earlier work becomes a dependency and a funnel for the new one.
- Put a finished video on the first screen of the open source README, push sample videos on X with "guess if it's real"-style engagement posts, and use the tool to run a content account so testing traffic doubles as promotion.
- Pitch the developer tool as "platform traffic support + monetization" and sell it to creators, not programmers.
- Major-version cadence (1.x → 2.0 → 3.0) paired with long posts, creating a new buzz moment every 1~2 weeks.
- Spread across Skill directories like skills.sh, with one-line install via
npx skills add.
10. Still unclear
- Which repost drove the 10-15 single-day +230: our token can't read the GitHub stargazer timestamp API. The mushroom review landed the same day, but its impact can't be verified.
- The author's X follower count wasn't readable (not shown on the page).
- I didn't find whether anyone on Xiaohongshu / Douyin used it to make a viral video, or whether it drove reverse traffic.
- Of 484 installs, how many people actually produced a finished video is unknown: it requires local image generation and a TTS setup, and it passed testing on Mac. The bar is not low.
Sources
- Repo and README: github.com/threerocks/sweety-handdraw-video; releases list (GitHub API)
- Author's X: V1 post x.com/leo_xiaolei/status/2109880459664752867; voiceover sample post x.com/leo_xiaolei/status/2109988867168444486; Codex tutorial long post x.com/leo_xiaolei/article/2109982650199060818; v1.4 long post x.com/leo_xiaolei/status/2112133426983932186; v2.0 post x.com/leo_xiaolei/status/2114979590305382616
- Juejin juejin.cn/post/7696047416342429747; CSDN tianqi.csdn.net/6ace3610536f457592afa0ba.html; mushroom review blog.mushroom.cv/blog/sweety-handdraw-video-pipeline/; SegmentFault segmentfault.com/a/1190000048357338; Zhihu roundup zhuanlan.zhihu.com/p/2101308452006000134
- Competitor roundups: www.cnblogs.com/xgss/articles/23264612; x.com/zhongying14/status/2104184165318918352; GitHubDaily x.com/GitHub_Daily/status/2116130539522683388
- Installs: www.skills.sh/threerocks/sweety-handdraw-video (484 installs)
- Stars and growth: our GitHub repo_profile (2026-11-06)