A demo video with 510k views brought in zero stars, while a list post plus one official launch brought in 870 stars: How claude-motion took off

A 18-year-old developer in Serbia built a toolkit that lets coding agents cut motion-design videos on their own, and named it claude-motion, a name nobody paid attention to at the time. In Oct 2 his demo video got 519k views, and the repo gained not a single star. Six days later, a reposting account ranked it #1 in a "20 repos" list. Then, 6 hours later, Anthropic officially launched a same-named feature, Claude Motion, available only to enterprise customers. Below: where these 870 stars came from, why product quality was not the main factor, and how to get the same kind of stars yourself.

A late-night video, and six days with no stars

In the evening of Oct 2, 2026, whaley (on X as @whaleyxbt, bio reading "18 years old, commercial software developer, Serbia," about two thousand followers at the time) posted:

i couldn't sleep after watching this. i work in osint. i asked my agent to visualise what my scrapers do all night while i sleep. it drew them as crawlers. (Couldn't sleep after watching this. I do OSINT. I asked my agent to visualize what my scrapers do all night while I sleep. It drew them as a swarm of crawlers.)

In the video, 1 agents split into 256 within 9 seconds, each crawling a web page and "reading" the words it stepped on, one by one. In 20 seconds they covered 16 pages and over 2,900 words. By 11-06, the post had 519k views, 3,760 likes, and 2,412 bookmarks.

But its repo whaleyxbt/claude-motion, created in Sep 29, gained 0 stars per day through Oct 8 (per Trendshift's daily table). The reason is simple: the post body didn't include a repo link, and the simulation feature from the video wasn't in the repo until 10-04. People thought it was cool but had no idea where to find it. (The repo may also not have been public yet in those days; GitHub doesn't expose a public-since date, unverified.)

What it actually is

A motion-video toolkit for Claude Code, Cursor, and Codex. MIT-licensed and free. It renders on Remotion, and the README's first line draws a line against Remotion's official skills: "Remotion skills teach your agent the API. This teaches it taste, makes it check its own frames, and gives it a sound engine."

It breaks down into four parts:

• Motion rules written as numbers: entrances use bezier(0.16,1,0.3,1), exit durations are halved, and word-by-word reveals are spaced 0.09 seconds apart, instead of vague advice like "make it smoother."
• Frame-by-frame self-check: agents can't watch video, so the tool exports a timestamped contact sheet of screenshots at each beat, checks it against a checklist for text collisions, cropping, too-brief holds, and out-of-safe-area elements, and revises before handing off.
• Sound effects synthesized in code: Python standard library only, generating typing, clicks, whooshes, and swells, mixing them on the same timeline and auto-normalizing to −14 LUFS. No hunting for assets, no licensing concerns.
• Web simulations recorded to video: each HTML page is one piece. A headless browser steps through it frame by frame and hands the frames to ffmpeg, recording at 60 fps. The crawler video was made this way.

The last commit was on 10-09, and there have been no updates in nearly a month since.

Where the stars came from: #1 on a list post, plus a name collision

(Trendshift daily table; 11-06 GitHub shows 870 stars and 110 forks)

The first push came from a list post. On 10-08 afternoon, an account with 9,500 followers, @N01ennn, posted: 「Netflix engineers posted 20 GitHub repos that turn your Opus 5.5 into a f*cking motion empire」 (Netflix engineers posted 20 GitHub repos that turn your Opus 5.5 into a motion empire). Entry 1 on the list was claude-motion, described as 「taste, checks its own frames, sound written in code」 with a link. By 11-06 the post had 525k views, 6,000 likes and 14k bookmarks. X's community notes later pointed out that the "Netflix engineers" claim was made up. Posts like "engineers at a big company posted N repos" often come from repost account networks. No evidence was found linking that account to the author.

Note that the two posts had nearly identical view counts (519k vs. 525k). One brought in 0 stars, the other brought in 229 stars over two days. The only difference: whether there was a link, and whether it was ranked first.

The second push was the name collision. 6 hours after the list post went out, Anthropic launched the official Claude Motion (launch post with 405k views), labeled "In beta on Team and Enterprise plans." Individual Pro and Max users couldn't use it, so they searched "claude motion github," and this repo, whose name had been set since late September, ranked first. Afterward, AI-written intro pages, aggregator sites, and Facebook AI Q&A pages all treated it as "the open-source version of Claude Motion." A third-party subscription site also used the name for search optimization and wrote several tutorials about it.

The product's strength shows up later. In the same list, the #2 entry howseen-ai/claude-motion-design has a nearly identical name but a different approach (no Remotion), and now has 774 stars, about the same as this one. That suggests list placement and name determined star counts, not product differences. The payoff from a solid product showed up elsewhere, though: 10-30, an account with 41k followers, @elvar_official, posted a long-form teardown of its own project (a 3D machine-teardown simulator). Its README says "Built on whaleyxbt/claude-motion (MIT)." The long post got 2.12M views, and the teardown alone gained 2,150 stars in a week, more than the original repo, which gained only twenty to thirty stars in the same stretch. Another educational motion project, holdframe, also reused its skill text and credited the original.

Who it's competing with

From late September into October, dozens of repos for "have a coding agent write code that produces motion-design videos" suddenly appeared (GitHub, 11-06):

There's also a string of small repos that borrow the name outright (claude-motion-skill, claude-motion-explainer, claude-motion-studio-skill), with a few dozen stars each, riding on the official name.

This niche is already crowded, and the biggest players didn't get there with this trick: they had company backing, big-name promotion, and mature frameworks. At 870 stars, claude-motion is only mid-tier here.

If it were you

Verdict: C. You can learn a practical playbook for earning stars, but building another one in the motion-skill niche isn't worth it. Its traction came from a one-time set of coincidences (the name was claimed early, the official launch landed the same day, and the list post ranked it first on that day). Followers can't reproduce the same timing, and the niche already has dozens of similar projects plus a Chinese big-name version.

Who it suits: people who watch big companies' release schedules, move fast, and can ship a working open-source version within a few days before or after an official feature launch; people already building in the agent-skill space who want to capture early stars.

Who it doesn't suit: people who want to build yet another "Claude motion skill" now. The name and the slot are both taken, and you'd be number 30. It also doesn't suit anyone who can code but can't post and has no one to post for them. This case shows exactly that with no link and no list placement, 510k views can still mean 0 stars.

How to do it. Copy the method, not the topic:

1. Camp on the next official feature that has a gate. When a big company's new feature launches enterprise-only, US-only, or on a single platform, individual users go looking for an open-source alternative. Claim a repo name in the form "product-feature" ahead of time, so a working version exists the same day the official launch drops.
2. Do what it didn't do well, and build on top. claude-motion's sound engine and frame-by-frame recording were taken by others to make derivatives, with credit given. You can do the same in reverse: build a vertical on top of it (product launch videos, data visualization shorts, tutorial videos), credit it, borrow the name, and collect the stars yourself. That's how teardown got 2,150 stars in a week.
3. Turn 'taste' into numbers. Whatever skill domain you're in, the first line of your README should draw a clear line against the biggest similar project, and your rules should be written as concrete parameters. That's why it got retained and reused in a crowded field.

How to get users: every demo video should include the repo link, and the first reply should too. Proactively contact accounts on X that post 'N repos' lists and aim to be listed first. After launch, submit to agent-skill directories and aggregator sites. Don't count on GitHub's daily trending; it only charted for two days.

When is the window: there's no fixed window, watch the big companies. The next time Anthropic, OpenAI, or Google ships a feature 'enterprise-first', the week before and after launch is the window. The motion-design niche itself is already late now.

What would show I read it wrong: ① Anthropic opens Claude Motion to individual Pro/Max users and the same-named open-source repos keep gaining stars. That would mean demand isn't just 'the official one is unavailable', the motion skill itself has room, and the rating should go up. ② The next big-company launch with a gate comes and same-named repos registered in advance get no noticeable stars. That would mean this was coincidence, and the playbook isn't repeatable.

Data and sources

GitHub: whaleyxbt/claude-motion repo page, README, commit history, stars (11-06), comparable repo stars (11-06). Trendshift: the repo's daily new stars and daily ranking. X: @whaleyxbt 10-02 demo post, @N01ennn 10-08 list post, @claudeai 10-08 launch post, @elvar_official 10-30 long post (views are public counts of 11-06). Claude Help Center Claude Motion page. Aggregators and intro pages: agentsuses, claudemontiondesignskills.org, thereelstack, xtxian, skills.sh. The match between LinkedIn reposts and the 10-19 small spike is inferred.