2026.08 · 7 min read

From affective intent to visual decisions: Notes on music-to-image generation

Project Log Music-to-Image Affective Computing

In the project on affective music-to-image generation with human-AI hybrid intelligence, I am less interested in producing an image that merely looks related to music than in making generation respond to a person's subjective intent. Beat, spectral, and semantic cues that a model can capture do not sit at the same level as a person's judgment about mood, tension, and visual breathing room.

The project therefore feels like an ongoing negotiation: the model proposes cross-modal candidates, a person redirects the affective trajectory through feedback, and the system turns that feedback into controllable visual decisions. The important task is not to add more control variables, but to find which kinds of feedback genuinely change the experience.