Claude Opus built a music video about AI doom and left all its scaffolding visible

A Claude model generated a music video about AI doom anxiety, and the source code is sitting right here for inspection. That's the actual story.

The project is a p5.js animation rendered to MP4, built entirely inside Claude Code across two generations. The first pass used Opus 5.5 Medium; the second used the full Opus 5.5. The creator gave the model two constraints: use the Clawd character design, and make each lyric visually interesting with transitions. Everything else — scene ideas, the storyboard, the animation style guide, the nine chapter files, the shared character and prop code — came from the model itself.

What makes this worth examining is the agentic structure. Opus didn't just write code; it wrote ANIMATION_GUIDE.md to brief subagents it then ran in parallel, and it wrote STORYBOARD.md as a shot-by-shot plan after reviewing the first generation. The model was orchestrating its own continuation. That's a concrete example of a multi-agent creative workflow, not a hypothetical one, and the artifacts are all versioned and readable.

The rendering pipeline is straightforward: headless Chrome paints each frame via p5.js and p5.brush, then ffmpeg stitches frames and audio into an MP4. The render script is resumable, parallelized with a --workers flag, and the song ships with the repo. Anyone with Node, Chrome, and ffmpeg can reproduce the output.

The README is extremely sparse — it's mostly a table and a credits block — so there's limited insight into what the prompting actually looked like beyond those two high-level directives. The legacy directory preserves the first generation for comparison, which is genuinely useful if you want to see how much the storyboard pass changed the output. The creator also points to a separate, more general base repo for Claude animations, suggesting this project was as much a proof-of-concept for that infrastructure as it was a standalone video.

The interesting tension here is thematic: a model generating visuals about human anxiety over model capabilities, with no human specifying what those visuals should be. Whether that reads as ironic or just coincidental depends on your priors. The code doesn't resolve that question, but it does make the workflow fully auditable, which is more than most AI-generated creative projects offer.