Since Anthropic released Claude Opus 5.5 on September 22, 2026, short animated videos "made by Opus 5.5" have been everywhere: motion graphics showreels, history explainers, product launches. A community index, claudevideo.org, had collected 1,276 of them from 1,007 creators by mid-October, and the awesome-opus5-5-videos list passed 6,000 stars in under a month.
The surprising part: Opus 5.5 cannot output a single video frame. Its models read text and images and write text and code. Every one of those videos is a program.
The four steps
1. The model writes a program
You describe the video. Opus 5.5 decides the layout, the colors, the type, the motion curves and the order of shots, and writes all of it down as code: usually HTML with CSS and a canvas or SVG, sometimes Three.js for 3D, sometimes a React framework such as Remotion.
Because the result is code, nothing is left to chance. A circle that moves from left to right at second three will do exactly that every time the program runs.
2. A browser draws each frame
A headless browser opens the program and steps through time one frame at a time, 30 frames for every second of video. Each frame is captured as an image.
This is also where good pipelines check their work. A frame can be inspected for text that runs off screen, shapes that overlap, or seconds where nothing is on screen.
3. Sound is added
The viral clips are often silent, or set to a track afterwards. An explainer needs more: a script read by a voice, subtitles timed to that voice, music and sound effects that follow the scenes.
4. ffmpeg encodes the MP4
ffmpeg stitches the frames and the audio into one H.264 file that plays anywhere.
Why it looks like motion design, not film
Since every pixel comes from shapes, text and illustration drawn by code, the results look like motion graphics: diagrams, charts, kinetic type, flat or isometric illustration. That is exactly what explainers need, and exactly what this method cannot do: filmed scenes, real people, photo-real places. For those you need a video model, which we compare in code-drawn video vs AI video models.
Doing it yourself vs. hosted
To make one of these videos on your own computer, you install Node.js, Chrome and ffmpeg, run Claude Code with Opus 5.5, and prompt it through each step. One widely shared Hacker News thread reported about $3 to $4 of API usage for a single launch video, plus the time to set up and iterate.
That is what our Opus 5.5 video generator hosts: you write one sentence, and the director, which runs on Claude Opus 5.5, writes the script, records the narration, draws and checks every scene, scores it and hands back the MP4 with subtitles.
Where to go next
- Six Opus 5.5 video prompts for explainers, one per visual style
- Why every Opus 5.5 showreel looks the same
- How long should an explainer video be?
Claude and Opus are trademarks of Anthropic. This site is independent and not affiliated with Anthropic.
