How it works
Different models hear different things
This prototype does not send the image to a cloud model. The browser reads pixels and tries three simple model ideas: color mood, shape contour, and light rhythm. The same image can become three different pieces of music.
Try these
Compare Color Mood and Shape Contour. Which one matches the picture better?
Turn on Mix and listen for chords where models agree.
Send the result to Remix or Music Coding and change it again.
Why this lab
Real image-to-music systems use vision encoders, emotion mapping, audio tokens, and generative models. This kid-safe version exposes the idea underneath: a model is a set of choices about what to notice.