pdoom-video: Render a 4K music video from one song with a single line of code

Analyze the song's lyrics and audio data to generate every frame as a deterministic function. The real-time preview in the browser and the exported 1080p60 or 4K60 video are pixel-perfect identical.

Using Demucs and Whisper to precisely extract lyric timing and beats, a three.js-based renderer visualizes them. For fast movements, it averages up to 324 subframes to achieve natural motion blur, rendering directly at physical resolution without upscaling even at 4K.

Unlike traditional video editing software, all visual effects are controlled by code, making edits and re-renders instantaneous. On an M5 Pro, rendering the entire track in 4K takes approximately 2.5 hours, and automated pipelines can be built using headless Chrome and ffmpeg.