用代码逐帧生成UI动效的完整流程

The trick is NOT the prompt. It is one rule: every frame is a pure function of time.

song -> numpy beat grid -> Opus 5.5 writes one HTML file -> Playwright renders -> ffmpeg blends

step 1 -> pick a royalty-free track around 120 BPM

step 2 -> give Opus 5.5 8 to 12 UI states: button, loader, toggle, chart, command palette

step 3 -> one HTML file, 1440x1440, every style computed inside seek(t). no CSS transitions, no timers

step 4 -> springs as closed-form math, so any frame renders in any order

step 5 -> numpy finds the beats, every UI sound lands on a measured peak

step 6 -> Playwright shoots 4 subframes per frame, ffmpeg tmix blends them into 60fps motion blur

step 7 -> render one frame per beat first, fix what is off the grid, then the full render

the math:

120 BPM x 7 bars = 28 beats = 14 seconds
14 s x 60 fps x 4 subframes = 3,360 browser renders