A 630k-view video that sent a Claude skill to two thousand stars: how claude-animated-shorts came to be

In November 1, a YouTube channel with only 11.7k subscribers posted a tutorial: have Claude watch the latest 20 animated shorts from another channel frame by frame, then make one from scratch. The first link in the video description was a GitHub repo created 25 minutes earlier. Five days later the video had 630k views, and the repo had nearly two thousand stars and over four hundred forks. Anyone trying to earn stars on an open-source project can learn a clear playbook from it, but first note that nearly all of those two thousand stars came from that one video.

A video and a repo created 25 minutes earlier

On Nov 1 05:50 2026 (UTC), GitHub user mromidev created claude-animated-shorts, pushed its only code 7 minutes later, and never touched it again. 06:15, his YouTube channel "Mr omi" published "How to Make Viral Animated Shorts Using Claude (Full Workflow)." The video opens with this claim:

I gave Claude a channel's latest 20 shorts, it studied them frame by frame at 30 fps, and then it built a full animated short with our own story… No AI video generator.

The description lists chapters first ("Claude watches 30 animated shorts at 20 frames per second and writes a style guide," "why the first version looked cheap," "seven agents each handle one theme, comparing our frames against reference frames"), then two links: the first, "Free skill (copy it into your Claude and it learns the whole process)," points to this repo; the second, "More skills + realistic 3D (paid community)," points to his paid Skool community.

As of Nov 6, the video had 632k views, 16.7k likes, and 844 comments (YouTube, 11-06). The channel has 9 videos in total and about 920k total views; earlier videos mostly got a few thousand to a hundred thousand views. This one video alone accounts for two-thirds of the channel's total views, a small channel breaking out with a single video.

What it actually does

In one line: once installed in Claude Code, you say "use the cartoon short skill to make a short about..." and it renders a 55–70 second vertical animated story short on your own computer with three.js + Remotion. 2D characters walk through a colorful 3D town, with outlines, depth of field, camera pans, march music and sound effects. Everything is rendered from code, with no AI video generator involved.

What's in the repo isn't just a prompt, it's a complete finished set:

• SKILL.md: the full pipeline from topic selection, fact checking, scripting, voiceover, storyboarding and still frames through rendering, plus the pitfalls the author ran into along the way;
• The full source of a demo short, "The Bridge That Fell for a Song" (about the Broughton Bridge in 1831 England, which collapsed under marching feet, 67 seconds): the world, characters, outlines, camera, color grading, and composited sound effects;
• What was "measured" from 20 reference shorts: a style guide, 7 measurement reports (motion, depth of field, materials, characters, lighting, outlines, color grading), 38 reference frames, and 20 contact sheets of 60 frames each.

The bar is not low: it needs Remotion, Node, Python 3 and ffmpeg, and the public/ folder is about 1.3 GB; the README specifically warns to put temp directories on a large disk. Comments already include people saying "I gave Claude this skill and what it made doesn't look like the original" and "we need a longer video explaining how to install it."

The reference channel being ripped off isn't named in the video, but the comments mostly say "Rick's videos" and "Ricks Clips". It's a channel making short videos with a 2D character plus 3D world trivia (my inference, based on the comments and the titles of similar reupload videos).

How the stars climbed

(The day-by-day figures from 11-02 to 11-06 are snapshots taken at a fixed time each day; the 11-06 row is live numbers from that evening.)

A few things stand out:

The video is the main engine. The two days of star surge line up exactly with the two days of view surge; by a rough calculation, 630k views and about 2,000 stars means roughly 1 out of every 300 viewers went and starred it (estimate). The X account reposts were the second force; AI news accounts on Facebook reposting the original, the TypingMind skill marketplace, and aggregator sites like trendshift were passive pickups. I see no Hacker News, Reddit post or tech media coverage.

The people showing up want something they can use. Forks run about 21% of stars, roughly twice what new repos usually hit. The comments show who the audience is: "I'm a complete animation beginner, got hooked on Rick's videos two weeks ago, subscribed to Claude and tried it myself, and was disappointed" and "I've been making videos with Remotion for a while, two of my projects are earning, and I still can't figure out how that channel makes 2D look so good in a 3D world." They want "a faceless channel that can get traction," not a tech project.

"Free" is the selling point itself. The top-liked comment (1,188 likes): "People sell this kind of thing for $97 a month, and then this guy just gives it away for free." Similar remarks show up a dozen-plus times in the first 50 comments.

A free skill, a paid community

The author's Skool community, "AI Creator Lab by Mr Omi", charges $1.50 a day (about $45 a month), with the pitch "prices rise as new skills launch, join now to lock in your price", selling "faceless YouTube channels + AI workflows". Two days later (11-03) he posted another video, "Ultra-Realistic 3D Animated Shorts", with 74k views. This version of the skill is only available in the paid community.

So for him the repo is a hook: the video shows the process, GitHub gives a free finished product to earn trust and secondary sharing, and the more advanced version sits behind the paywall. The stars are a by-product. How many people are in the community and how much it earns has not been disclosed.

How crowded this niche is

"Have Claude Code write code to make videos" has become a hot category this year, and claude-animated-shorts is not the technically strongest entry:

It has only one real difference, but it's a strong one: others sell "capability" (can do motion graphics, can do explainers), while this one sells a format already proven to go viral on short-video platforms, plus the appealing process of "have the AI dissect 20 viral videos first, then replicate them." That difference mainly makes the video itself take off; what actually brought in the stars was the video.

Copycats came fast. On GitHub, artemnovitckii/claude-animation-skills was created the same day with 402 stars; shorts-kit, sunshower, leviathan and others appeared within 11-03 to 11-04 days, each with a few dozen stars. On YouTube, from 11-04 onward there were remixes like "Make animated shorts with claude code (Full Workflow)", "Claude Code can make Viral Animated 3d/2D shorts", and "How To Make Viral Ricks Clips Style Animations With Claude", plus people re-uploading under the original title and re-cutting it into segments on TikTok. One comment nails it: "competition is going to be so high now," another: "There's no edge left anymore."

If it were you

One-line verdict: copying this and building yet another "Claude makes a 2D character 3D world short" skill is already too late. The original is out, the copycats are out, and the repost videos are out too. But the playbook behind it is reusable: pick a content format that's currently hot → have AI break down the viral examples and turn them into a style guide → package that into a skill that can produce the video directly → post a process video alongside it. Swap in a different format and it still works. Whether stars come in, though, depends almost entirely on your video.

Who it suits: developers who have a YouTube, TikTok or X account and can make a tutorial video whose process is watchable; people who can write three.js / Remotion / canvas and who actually scroll short-video feeds and know which formats are picking up.

Who it doesn't suit: people with no distribution channel who plan to just drop a repo on GitHub and wait for people to find it. Without the video, this project would not have those two thousand stars. Also not for people who want to build "yet another general-purpose AI video skill": the generic slot is already taken by Remotion's official work and several projects with tens of thousands of stars.

How to do it:

• Pick a format, not a tech stack. Go to Shorts / Reels / TikTok and look for faceless formats that took off in the last month or two and are clearly being mass-produced from one template (fun-fact animations, one fixed art-style explainer, pseudo-documentaries, pixel-art stories, etc.). Then check whether anyone has turned it into a skill yet (inference: going by the repos you can find, 11-06 apart from the 2D-plus-3D kind, very few other formats have a "finished-format skill").
• Package the dissection output along with it: style guide, measurement reports, reference frames. This is what makes the repo look substantial and trustworthy, and it's also the most watchable part of the video.
• Ship a demo finished-video source that runs directly, not just a prompt.
• Fix what it got wrong: it's too heavy to install (one-command install, assets downloaded on demand instead of copying 1.3 GB in full), newcomers can't reproduce the original results (provide complete "zero to finished video" conversation examples), and there's only one style (make it a series with one sub-skill per format).
• Be clear about the boundaries of the material: its README says "stories, characters and voiceover should be your own; reference frames are only for learning the art style." Comments already include people saying "feels like copying the person who built it first," and stating this up front heads off a lot of controversy.

How to get users: create the repo and publish the video in the same hour, put the repo link as the first link in the video description, and title the video with a pattern like "How to Make Viral … Using Claude (Full Workflow)". Write the repo description as a single, directly quotable selling point (theirs is "Free Claude skill: viral animated story shorts…"), so AI news accounts like @RoundtableSpace on X can repost it verbatim. Afterward, submit the repo to skill marketplaces and directory sites to pick up long-tail traffic.

When to move: this wave of "Claude makes video with code" content has been hot since mid-October, and tutorials and copycat repos keep appearing daily. Once a format gets packaged into a skill, it gets reposted within days. If you want to do it, pick a format nobody has dissected yet before the end of 2026, and finish and publish within a week.

What would show you misread the situation: ① new "format skill + tutorial video" combos generally only get a few thousand views and a few dozen stars, which would mean audiences are already tired of "Claude makes video"; ② Anthropic, Remotion or AI video generators make "consistent-style, editable animated shorts" an official capability, which weakens the case for the code-rendering route; ③ platforms start cracking down on mass-produced AI faceless shorts (comments already say "The slop thickens"), the people using it to grow channels leave, and nobody looks for the skill anymore.

Data and sources

GitHub repo page, README and author's profile (11-06); GitHub daily star records (from 11-02); the YouTube video OtbLL6OBNE4 description, views and comments, Mr Omi channel data (11-06); X @RoundtableSpace 11-02 post; skool.com/mromi landing page; TypingMind and trendshift listing pages; star counts for competitors and copycat repos are as of 11-06, figures marked "about" are rounded.