Claude made this. Even the voice. All it needed was a Folder.

Everything in this Short was run by Claude Code: the script, my cloned voice, the word timings, the storyboard, the animation in four styles, the render and the captions. I read what each step made and said go. How it's built: It's one folder, and Claude Code walks it one stage at a time. There's a CLAUDE.md at the top that works as the map (what lives where, where to go for each job), and a folder per stage, each with a short CONTEXT.md saying what it reads, what it makes and who checks it. • Voice. Claude writes the script in how I actually talk, then a small Python script sends it to ElevenLabs in chunks using my voice clone. I listen before anything else happens. • Words. Whisper runs locally on my GPU and writes down every word with its start and end time. That timing file drives everything after it. • Storyboard. A plain text file, one beat per visual change, each beat pinned to a few spoken words. A command matches them to the timings. • Scene. Claude writes the animation as JavaScript on a small kit we built up over time: pixel characters, hand-drawn line art, a clean modern look, a blueprint look, and real screens captured straight off my computer. Every frame is a function of time, so the pictures land on the words. It grabs stills of every beat and fixes what overlaps or gets cut off. • Render. Playwright plays the scene one frame at a time in Chrome and FFmpeg mixes in the voice, music and effects. The same folder writes the cover and the description. The files on screen in the video are the real files from this Short. A person reading each stage's output before the next one runs is the part that keeps it from turning into slop. Happy to answer questions on any stage. I have a little bit longer tutorial for free on youtube if anyone wants to dive deeper: How to Make AI Videos That Don't Feel Like AI Slop (https://www.youtube.com/watch?v=EhWlGingCl0)