Free video from code: the canvas + Playwright + FFmpeg pipeline
No Remotion and no generator credits: Claude writes a function of time, a headless browser draws each frame, FFmpeg stitches them, and the agent checks its own frames.
Chase AI's walkthrough starts from one idea: Claude never makes a video. It writes a function. You give the function a time t and it returns one picture. Ask for 60 pictures a second and play them in order, and you have video. Because the same t always gives the same picture, any frame can be redrawn or checked on its own. Playwright runs a headless browser that draws the frames, FFmpeg stitches them into an MP4, and the sound can be another function of t or a narration track. His one-shot demo explainer was made in a 3Blue1Brown style, with ElevenLabs narration.
Effort level matters less than the prompt. He gave Opus 5.5 the same 'make a dynamic 15-second motion graphics showreel, no skills' prompt at low, high and the top effort setting. Low took about 20 minutes, high about 30, and the top setting about eight hours. The top setting was best, but all three were usable. His conclusion is that if your results are weak, the prompt is the more likely problem.
Ask for a storyboard first. Before any code, have the agent lay out the main scenes as individual frames so you can redirect the piece cheaply. Renders can take 20 minutes or more, and skipping this step is how people end up in hour-long regenerate loops.
His prompt checklist, in our words: (1) the length and type, e.g. '30-second animated explainer about X'; (2) the aspect ratio, 16:9 or 9:16 for shorts; (3) build it in JavaScript; (4) the soundtrack, either generated in code or a track you supply; (5) if you have narration, give the MP3 first and have the visuals timed to it; (6) render to MP4 with Playwright and FFmpeg; (7) before calling it done, pull a contact sheet with FFmpeg, inspect it, fix, re-render and repeat.
References come next. You can attach images to copy a look, or a reference video to borrow its composition while you swap in your own subject. In his example, the agent remade a Remotion-built currency explainer as a JavaScript video about the history of electricity, and the result looked like the same family of design. He recommends Skillry's AI-video gallery and Kevin Ngo's posts on X for references.
Finally, he packaged all of this as the free animate skill. It walks you through intake questions (subject, format, audio, voice, whether to have a recurring hero character), then story, then storyboard, then build. It ships seven styles (cut paper, crosshatch, riso and sketchbook are among them, plus isometric), installs Playwright, FFmpeg and faster-whisper, and offers to set up the ElevenLabs connector. He calls it a work in progress and accepts pull requests.