Baoyu asked for a collection, and someone delivered a style library two and a half hours later: lemo-opuscar 41 days 12.6k stars
What it is A "make films with code" style library for Claude Code: 66 film styles, each with a style prompt and an Opus 5.5 sample film made purely in code. Install with two commands, make a film with one sentence.
Why it took off After Opus 5.5 launched, everyone wanted it to write code to make videos, but there were no ready-made styles. The author built the library two and a half hours after Baoyu publicly asked for a collection, then kept amplifying through frequent updates, the big film "OPUSCAR 98", and reposts from Chinese and English creators.
CHow to play it Don't build another Opus video style library. Wait for the next model launch, then in its first week make dozens of finished pieces by hand, package them as a one-click plugin, reply to KOLs' "any recommendations?" posts, and ship one standalone showpiece that can spread on its own.
- Data (updated daily) · Repo created 2026-09-26 · 13,340 stars · 7 days +3,246
GitHub · stars Data 2026-09-28 → 2026-11-08 (42 days, 32 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
lemo-opuscar is a set of styles for "making films with code" that installs into Claude Code: 66 film styles, each with its own style prompt, plus a sample film made entirely in code by Claude Opus 5.5, with no video model involved. Built in Sep 26, it hit 12,613 stars and 1,462 forks on GitHub by Nov 6. It took off for two reasons: it went live on day four after Opus 5.5 launched, and the sample film was good enough to share as-is. Below: how the stars came in, who's competing for the same attention, and whether building one yourself is worth it now.
A "recommend me" post
In 2026, Sep 26 at 3:42 Beijing time, Baoyu (@dotey) posted one line on X:
Asking for a recommendation of an open-source Awesome Opus 5.5 Video Prompt collection on GitHub
Four days earlier, on Sep 22, Anthropic had just released Claude Opus 5.5. "Have Opus write code to make videos" was everywhere in the Chinese AI community in those days, and Douyin had an official #Vibe Knowledge Awards event running at the same time, so people making knowledge videos were all hunting for ready-made styles and prompts.
Two and a half hours later, Melbourne indie developer Lemomo (X @lemomo_ai, about 7,000 followers) created the lemo-opuscar repo and replied under Baoyu's post with their own library. The next day the author published the Chinese launch post:
I've been hooked on making videos with Claude Opus 5.5 these past two days, made 100+ clips, picked 39 styles, and open-sourced them as Lemo-Opuscar... each style has a style prompt and a sample film made purely in code, plus method docs on directing, storyboarding, and scoring.
The same day, @GoSailGlobal reposted it (6,049 views). About 111 stars on day one, Sep 28 454 stars.
What it actually is
The README's opening line is blunt: "Not an awesome list. Every film here was made by me, with Claude Opus 5.5." This isn't a list of other people's work; the author made every film.
Every film is made the same way: the visuals are frame-by-frame canvas / WebGL rendering in a web page, the score is assembled from free sample libraries, narration uses TTS, and ffmpeg composites the final cut. The 66 styles range from ink wash, shadow puppetry, and ukiyo-e to documentary live-action and lidar survey maps. The repo also includes two method docs: DIRECTOR.md on how to direct, and TECHNIQUE.md on how to make it.
Getting started takes two commands:
`` claude plugin marketplace add lemomo-ai/lemo-opuscar claude plugin install lemo-opuscar@lemolab ``
Then just say something like "make a 45-second watercolor-style film about my family's coffee farm, with a warm female narrator," and it's done. Before starting, the plugin asks three things: what language, whether you have your own voiceover and assets, and whether you want to see the storyboard first.
A commenter put the difference from others well: "I've seen lots of Opus5.5 videos before, but they were all image-generation style. Opuscar actually did live-action style."
Where the stars came from
| Date | Stars | What happened |
|---|---|---|
| 09-26 | 0 | Baoyu asked for a collection; two and a half hours later someone built the library and replied |
| 09-27 | About 111 | The author's Chinese launch post; @GoSailGlobal reposted it, 6,049 views |
| 10-09 | 1,414 | Released the big film "OPUSCAR 98" around then |
| 10-20 | 3,658 | +647 in a single day; the same day English creator Duncan Rogoff posted "OPUSCAR 98" on X and TikTok |
| 10-22 | 4,800 | Major update: 8 documentary / editing styles and "make your own style"; reposted by Chinese creators like @indie_maker_fox and @LinearUncle |
| 10-28 | 8,860 | Single-day peak +903 |
| 11-06 | 12,613 | Last 7 days +3,233, the 7 days before that +3,960. It's cooling. |
(GitHub daily snapshots)
Break this curve apart and it's three forces stacking on top of each other.
The author kept shipping new things. Style count went from 39 to 43, 51, 59, 66, and each update came with a new sample film, which itself became repost material. The most useful one was "OPUSCAR 98": 98 Best Picture Oscar winners, each in a different art style, 6 min 31 sec. It isn't a tool pitch; it's a standalone piece people could share, and when they shared it, the repo came along.
Chinese and English creators passed it along. The Chinese side took off first: @GoSailGlobal, @Voxyz_ai, @indie_maker_fox, @LinearUncle, and xiaohu.ai (小互) wrote two long-form pieces; articles on Zhihu, OSChina, and Tencent Cloud Developer all ranked it near the top of "Opus video" projects. The English side came a bit later: the README was already bilingual, and on the day Duncan Rogoff's post appeared in Oct 20, it gained 647 stars. But his TikTok had only 385 views, so who actually drove that jump is still unclear.
Aggregator sites picked it up. Skill directories, Trendshift, and the official Star History account recommended it in Nov 5. Videos about it on YouTube have low view counts; the most-viewed one is about 6,300.
No sign of paid promotion found. The fork-to-star ratio is about 11.6%, similar to new repos of the same size at the time. Release assets (the sample film pack, web pack, and tools) have been downloaded about 430k times in total (GitHub), suggesting many people actually used it, not just starred it.
Who's competing alongside
Over a month after Opus 5.5 launched, dozens of "Opus video" repos appeared. The largest few:
| Project | Repo created | Stars | How it differs from lemo-opuscar |
|---|---|---|---|
| yihui-dev/awesome-opus5-5-videos | About 09-27 | 13.7k, cooling | Scraped viral videos and prompts from the web; not one person's tuned styles |
| alchaincyf/huashu-art-motion (Huashu) | 10-06 | 6.4k | 35 art styles, from a big V, 10 days later, fewer sample films |
| Vincentwei1021/mg-styles-15 | October | 3.1k | 15 motion-graphics styles, also written by Opus in code |
| HeyChengC/vibe-knowledge-film | 10-06 | 1k | Reverse-engineered from viral Douyin knowledge videos |
| XiaoyeCodes/code-animation-prompts | End of October | 622 (8 days) | Count: 341 code-animation style prompts |
| Remotion / HeyGen HyperFrames | Early / March | 71k / 79.9k | A rendering engine: it handles how to render, not what style to shoot in |
The rendering engines underneath already existed. The "collection" Baoyu wanted was built by someone the next day, with slightly more stars than his. lemo-opuscar sits between the two: a style pack one person hand-built and tuned, with finished footage for every style as proof, plus a plugin installed with one command, covering the whole path from picking a style to getting a video out.
But this position has no moat. A style prompt is just a Markdown file, anyone can write one. Those who followed include ones racing on count (341 styles), ones riding a Douyin campaign, and ones focused on motion design. It stays at the top because it came first and updates more often, not because others can't build it.
If it were you
C: Learn its rhythm, but don't build another "Opus video style library." This category produced dozens of libraries in a month. The top two both passed 12k stars and are now cooling. A style library is just a Markdown file anyone can copy. What it earned came from the two and a half hours in Sep 26, and that window has closed.
Who it suits: developers who want to get stars early when the next new model ships, who are willing to hand-build dozens or hundreds of finished pieces and have decent taste; or people with an existing body of work in a niche (infographics, PPT, game art, music visualization) who can carry the "style + sample footage + one-click plugin" package over to it.
Who it doesn't suit: anyone planning yet another "Opus video library with more styles." Someone has already reached 341 styles on count and only collected a few hundred stars. Also those who won't build finished pieces themselves and only want to curate others' work; the awesome-list route is already taken.
How to do it:
- Watch for the next new model or platform capability launch. In the first week, find one thing "it can do, but people don't yet know how to do."
- Hand-build dozens of finished pieces, each with footage you can watch right away, and put the gallery in the first screen of the README.
- Package it as a plugin installable with one command, and have it ask questions before it starts working.
- Prepare one standalone showpiece that can spread on its own (like "OPUSCAR 98"), so the tool library breaks out through it.
How to get users: reply directly under KOLs' "looking for recommendations" posts. Zero cost, and this is its first batch of stars. Make the launch post clear: "how many were made, how many were picked, each one has finished footage." Start the README bilingual, build traction in one language community first, then wait for bloggers in the other to repost it. After that, post an update every one or two weeks with new footage, giving people a fresh reason to share.
When to get in early: within 7 days of a new model launch. Going by this round's pace, the first decent library appears on day 4 after launch, the lists on day 5, and within two weeks the same category has more than ten entries.
Signs you've misread it: ① A Opus video style library that doesn't exist today appears before year-end and passes 3,000 stars without big-account backing. That means this category still has organic traffic, and it's easier than C. ② lemo-opuscar stops updating for a month and still gains more than 1,000 stars a week. That means it runs on word of mouth and real users, not hype. ③ Other models (GPT, Gemini) see the same "write code, make video" craze, but this batch of Opus library styles won't run there. Then rebuilding the style library on another model is a new window.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one Remotion + remotion-dev/skills Open source 2020 / 2026-01 71k / 6.5k stars A React video engine plus usage docs. It provides no styles; taste is up to the user. HeyGen HyperFrames Open source 2026-03 79.9k stars An HTML-to-video rendering framework. It handles how to render, not what style to shoot in. yihui-dev/awesome-opus5-5-videos Open source About 2026-09-27 13.7k stars, cooling An awesome list collecting viral Opus videos and prompts from the web. lemo-opuscar is a single-author release: tuned styles + method docs + a one-click plugin. Vincentwei1021/video-shotcraft Open source 2026-07 17.2k stars Uses Remotion to make product promo videos, leaning commercial promo rather than a multi-style library. Huashu (花叔) huashu-art-motion / huashu-motion Open source 2026-10-06 / 10-18 6.4k / 2.4k stars A similar art-style / promo video skill from a big-name creator, released 10 days later, with fewer styles and samples. hypit-ai/hypit Open source About 2026-07 28k stars A workflow for recreating viral videos, not a style library. Manim / Motion Canvas Open source 2015 / 2022 96k / 19.7k stars An explainer animation engine for humans writing code. Video generation models (Sora, Kling, Seedance, etc.). Big-company features — Huge They generate pixels, so changing one thing means re-rendering the whole clip. This is editable, reproducible code. - Where users come from
- Verified Platform event / demand window: Opus 5.5 9-22 released; Douyin's official "Vibe Knowledge Showcase" campaign and Xiaohongshu's "AI Says It For Me" popularized code-made knowledge videos; Baoyu @dotey 9-26 07:42 asked for an Opus video prompt collection, and the author built the library and replied under his post 10:10.
- Verified Author posting work on X: 9-27 Chinese launch post (39 styles); at about 1.5k stars released "OPUSCAR 98"; 10-22 "epic update"; 10-29 added 7 new styles; every time with new footage.
- Verified Chinese AI creators and news sites: @GoSailGlobal 9-27 (6,049 views, x library), @Voxyz_ai 10-15, @indie_maker_fox / @LinearUncle 10-22; two feature articles on xiaohu.ai; articles on Zhihu, OSChina, and Tencent Cloud Developer.
- Verified English AI creators: Duncan Rogoff 10-20 posted "OPUSCAR 98" on X / TikTok, gaining +647 stars that day (timing matches, but can't confirm he drove them); also Instagram reels driving traffic and Facebook repost pages.
- Verified Skill aggregators / GitHub derivative sites: skillsllm, typingmind, tutti, opuskill, SourceForge mirror, Trendshift badge, Star History 11-05 recommendation.
- Verified YouTube: repo walkthrough videos have low view counts, with the top one around 6.3k (Alex Hitt 10-18).
- Inferred Paid promotion / HN / GitHub trending: none found. A repo under 60 days old can't reach trending.
Why lemomo-ai/lemo-opuscar took off
What it is: a "code-made video" style library for Claude Code (plugin / skill). It has 66 video styles, each with a STYLE.md style prompt, plus a sample film the author made with Claude Opus 5.5 entirely in code: visuals rendered frame by frame with canvas / WebGL, music stitched from free sample libraries, narration via TTS, and final composition with ffmpeg, no video model used. The repo also includes DIRECTOR.md (how to direct) and TECHNIQUE.md (how to build), and installs as a Claude Code plugin with one command. MIT license. The author, Lemomo, is a Chinese indie developer in Melbourne (X @lemomo_ai, about 7,090 followers).
Conclusion: 9-22 Opus 5.5 launched, and "have Opus write code to make videos" immediately blew up in the Chinese AI community and on Douyin (the "Vibe Knowledge Showcase" campaign). People lacked ready-made styles and prompts. 9-26 Baoyu (@dotey) publicly asked on X for an "Awesome Opus 5.5 Video Prompt" collection, and the author built the library and replied to him 2.5 hours later. It was the earliest in its category to offer a "style library + one hand-made sample per style." It kept scaling through three things: the author frequently posting new styles and the showpiece "OPUSCAR 98", Chinese and English AI creators reposting it, and skill aggregator sites listing it. It worked mainly because of timing plus the work itself being good to look at. Distribution was all organic reposting; no paid acquisition was found.
---
1. Growth data (GitHub repo_day; GitHub API)
| Date | star | Notes |
|---|---|---|
| 2026-09-22 | — | Anthropic releases Claude Opus 5.5 (Reuters, evolink, etc.) |
| 09-26 07:42 UTC | — | Baoyu @dotey posts "looking for recommendations for open-source Awesome Opus 5.5 Video Prompt collections on GitHub" |
| 09-26 10:10 UTC | 0 | Repo created; the author replies under Baoyu's post with their own work library |
| 09-27 | About 111 on day one | Author's Chinese launch post: "made 100+ pieces, picked 39 styles"; reposted by @GoSailGlobal, 6,049 views (x library) |
| 09-28 | 454 | Wayback data points filled in |
| 10-09 | 1,414 | |
| 10-10~10-19 | Daily gain 100–280 | At about 1.5k stars, released "OPUSCAR 98" (98 Best Picture Oscar winners, each in a different art style, 6 min 31 sec) |
| 10-20~10-28 | Daily gain 500–900 | 10-20 Duncan Rogoff English post; 10-22 major update (documentary / editing 8 styles, "make your own style"); 10-22 Fox and LinearUncle Chinese posts; Xiaohu and Zhihu articles; 10-28 single-day peak +903 |
| 10-29~11-06 | Daily gain 250–610 | 10-29 added 7 more styles; 11-05 recommended by the official Star History account |
| 11-06 | 12,613 (live on GitHub 12,895), forks 1,462 | Last 7 days +3,233, the first 7 days +3,960; cooling, but volume is still high |
- Fork/star ratio about 11.6%, roughly the median of similarly aged, similarly sized new repos (about 11%). No sign of star inflation.
- Direct fork creation times track the star curve: 9-27 had 33, 10-19 had 91, and 10-22–10-26 hovered around 80 per day (GitHub forks API).
- Cumulative release attachment downloads: films about 200k, web about 137k, assets about 90k (GitHub API). These are samples, web packages, and tools, showing people are actually downloading the samples and tools, not just starring.
2. Competitors and peers (a very crowded field)
The category's base layer is engines that render video from code, with agent skills on top. In the month-plus after Opus 5.5 launched, dozens of "Opus makes video" libraries appeared (GitHub repo search).
| Product | Type | Launched | Scale | How it differs from lemo-opuscar |
|---|---|---|---|---|
| Remotion + remotion-dev/skills | Open-source engine + official skill | 2020 / 2026-01 | 71k / 6.5k stars | A React video engine with usage docs. It provides no styles; taste is up to the user. |
| HeyGen HyperFrames | Open source (big vendor) | 2026-03 | 79.9k stars | An HTML-to-video rendering framework. It handles "how to render," not "what style to shoot in." |
| Manim / Motion Canvas | Open-source engines | 2015 / 2022 | 96k / 19.7k stars | A math / explainer animation engine for humans writing code. |
| browser-use/video-use | Open source | 2026-04 | 30.6k stars | Lets an agent edit existing video rather than drawing from scratch. |
| yihui-dev/awesome-opus5-5-videos | Open-source collection | About 2026-09-27 | 13.7k stars, cooling | Collects viral videos from the web and their prompts (the author claims to have spent $3,000 recreating 301 of them). It's an awesome list. lemo-opuscar is "single-author output: tuned styles + directing method + one-click plugin." |
| Vincentwei1021/video-shotcraft | Open-source skills | 2026-07 | 17.2k stars | Uses Remotion to make product promo videos, leaning commercial promo rather than a multi-style library. |
| alchaincyf/huashu-art-motion, huashu-motion (Huashu) | Open-source skills | 10-06 / 10-18 | 6.4k / 2.4k stars | Big V release, 35 art styles + 9 explainer / product-promo styles; 10 days after lemo-opuscar, fewer styles and fewer sample pieces |
| hypit-ai/hypit | Open source | About 2026-07 | 28k stars | An agent workflow that replicates viral videos: swap the material and regenerate |
| Video generation models (Sora, Kling, Seedance, etc.). | Big-company features | — | Huge | Generates pixels, so changing one thing means re-rolling the whole clip; lemo-opuscar is editable, reproducible code |
Difference: most rivals are either "engines" or "lists of other people's prompts." lemo-opuscar is a hand-built, tuned-one-by-one style pack. Every style comes with finished footage as proof, plus two method docs (director and technical) and a plugin installed with one command, covering the whole path from "pick a style" to "output a film." The showcase film, OPUSCAR 98, is itself shareable: 98 Best Picture Oscar winners, each in a different art style.
3. Growth tactics (evidence)
- Riding a platform event and a demand window (confirmed): Opus 5.5 9-22 launched, Douyin ran an official "Vibe Knowledge Awards" campaign at the same time (Baidu Wenku explainers, several #vibe知识大赏 Douyin videos, one posted 9-23 with 428k likes), and Xiaohongshu had "AI speaks for me" (@369Serena: "lots of people grew their accounts with videos"). People making knowledge videos urgently needed styles. 9-26 Baoyu publicly asked for a collection, and the author built the repo that same day and replied under his post.
- The author kept posting work on X (confirmed): launch post in Chinese 9-27; at about 1.5k stars posted OPUSCAR 98 and publicly shared 98 styles and their prompts; 10-22 published a long "epic update" post; 10-29 added 7 new styles, one from community author @EianLu. Each update came with a new sample film, and the samples themselves were the promo material. The posts tagged @claudeai and @ClaudeDevs.
- Relayed by Chinese AI bloggers and news sites (confirmed): @GoSailGlobal 9-27 (6k views); @Voxyz_ai 10-15; @indie_maker_fox 10-22 (2.2k views); @LinearUncle 10-22; @GeekCatX, Sac and others; xiaohu.ai wrote two feature articles and listed it among "ten open-source project breakdowns"; Zhihu columns, OSChina and Tencent Cloud developer articles all ranked it near the top of Opus video projects.
- English AI creator repost (confirmed): Duncan Rogoff 10-20 posted "OPUSCAR 98" on X and TikTok. The TikTok has only 385 views. Stars rose +647 the same day, and the timing lines up, but we can't be sure he drove it. There are also Instagram "comment VIDEO for the link" lead-gen reels, Facebook repost pages, and LinkedIn posts.
- Indexed by skill aggregators and GitHub derivative sites (confirmed): skillsllm, typingmind, tutti, opuskill, skillsdirectory, bestproductivityresources, a SourceForge mirror, Trendshift (badge in the README), and 11-05 Star History recommendations. Repo walkthrough videos on YouTube got low views; the top one was about 6.3k (Alex Hitt).
- Not found: no hits on the HN library; no paid promotion found. The repo was under 60 days old, too young for GitHub trending (a pattern in the data), so growth did not depend on trending.
4. Why it took off
- Timing was precise: it went live 5.5 day(s) after Opus 4 launched, just when demand peaked and supply was thin. Baoyu publicly asked for a collection, and it filled that gap the same day.
- Samples are persuasive: "not one frame came from a video model," and the styles vary widely, including ink wash, shadow puppetry, ukiyo-e and real-world documentary. A user's words: "I've seen plenty of Opus5.5 videos that were all image-generation style, but Opuscar actually did a live-action look." Using finished footage as the ad makes it instantly clear.
- Low barrier to entry: two commands install the plugin, and one sentence produces a film; the plugin first asks about language, materials and whether to review a storyboard. Compared with awesome lists, users get a tool they can run directly, not a pile of links.
- Steady updates: 39 → 43 → 51 → 59 → 66 styles, plus OPUSCAR 98, documentary and editing styles, and "make your own style." Each update gave bloggers a new reason to repost, which is why another wave came after 10-20.
- Bilingual: the README has Chinese and English versions, so it built traction in the Chinese community first and was then reposted by English bloggers.
5. Visibility
Already widely discussed, just with no revenue (purely open source): 12.9k stars; xiaohu, Zhihu, OSChina and Tencent Cloud all wrote about it; Star History recommended it; Chinese and English X bloggers reposted it repeatedly; many similar projects and copycats exist. In the "Opus makes videos" niche it is a top-tier known project, and latecomers building style libraries will find it hard to stand out.
6. Copycats
- XiaoyeCodes/code-animation-prompts (8 days old, 622 stars, "341 code-animation style prompts," directly riding the style count)
- andyhuo520/vibe-knowledge ("44 style sets, code rendered frame by frame," aimed at the Douyin Vibe Knowledge Awards)
- HeyChengC/vibe-knowledge-film (1k stars)
- Vincentwei1021/mg-styles-15 (3.1k stars, "15 motion graphics styles, each coded by Opus 5.5")
- cth9191/animate (1.6k stars)
- huashu's huashu-art-motion (10-06,35 art styles). Closest in form, but we can't say it was copied from.
7. What others can learn
- In the first week after a new model launches, build a style or template library around what it "can do but people don't know how to do," with hand-built, tuned, each item backed by a finished piece. That is worth more than an awesome list.
- Use a standalone, shareable showcase work (OPUSCAR 98) to drive traffic to the tool library.
- Reply directly under KOLs' "looking for recommendations" posts. That's a zero-cost cold-start entry point.
- Package it as a plugin installed with one command, and have the plugin ask questions before acting, lowering the barrier to entry.
- Apply the same model to other verticals: code-generated infographics, slides, game art, music visualization. But people are already doing these (e.g., huashu-design).
8. Still unclear
- Who actually drove the 10-20 jump (+647): we didn't get Duncan Rogoff's X post engagement, it may have been a big Chinese account or Reddit; the stargazer breakdown couldn't be pulled via API this time.
- Exact views and engagement for the author's launch post and the OPUSCAR 98 post (X pages didn't show the full numbers).
- Actual usage: one film takes 30–60 minutes and a lot of tokens, and users must first download a tool over 1GB. How many stargazers actually made a film is unverified. Forks and downloads suggest some real use.
- The styles were tuned on Opus 5.5; after a new model ships, whether the library loses value depends on whether the author keeps up.
Sources
- GitHub: github.com/lemomo-ai/lemo-opuscar, README, releases, forks API (2026-11-06 captured)
- Our GitHub database: repo_profile / repo_day (2026-09-28~11-06), competitor search
- Our X database: @GoSailGlobal 2104028369688076457, @indie_maker_fox 2113130490161135678
- X:x.com/dotey/status/2103751976400101651 ,x.com/lemomo_ai/status/2104070024575242622 ,x.com/Voxyz_ai/status/2110838129326473353 ,x.com/DuncanRogoff/status/2112635070540390772 ,x.com/LinearUncle/status/2113222028178141540 ,x.com/369Serena/status/2111310833489793054 ,x.com/StarHistoryHQ/status/2118274845897056412
- Articles: xiaohu.ai/article/lemo-opuscar-code-filmmaking/, xiaohu.ai/article/opus-5-5-code-creative-projects/, zhuanlan.zhihu.com/p/2098566569270026623, my.oschina.net/jeecg/blog/19773689
- Opus 5.5 launch: evolink.ai/blog/claude-opus-5-5-release-date (2026-09-22)
- Douyin Vibe Knowledge Awards: kuku.baidu.com/landing/tscp_doc/a82a79448a45dc1443c8dcda1dcd…, m.douyin.com/share/video/7688645727061272805
- YouTube Data API (video views), TikHub (Duncan Rogoff TikTok 385 views)