AI Cappella transforms a collection of characters, each singing a single note, into a playable choir.

A challenge with AI video generation is making something intentional that you can reliably iterate on. It can feel like you’re playing a slot machine and losing real money. For AI Cappella, I needed to create a series of videos that followed a specific combination of pitch, timing, animation, and visual style, without triggering Seedance 2.5’s aggressive safety moderation. A convincing singing character was only useful if its performance blended believably with the group.

That became a practical test for DAMAGE, a tool I’ve been building to compose prompts from reusable, customisable elements. I went from struggling to produce one video that met all the requirements to generating a different character for each of the 12 notes in the chromatic scale in almost a single pass.

I keep imagining a synthesizer whose modules can produce believable voices, faces, movements, or entire environments, real or otherwise. What would its controls look like? What would it feel like to improvise with it? How could you preserve the surprise of experimentation while developing enough control to play with intention?

AI Cappella is a small exploration in that direction, combining generated clips with modes and interactions that turn them into a playable choir.