How might music become an input for image manipulation?
Some things are almost impossible to describe with words, yet music conveys them effortlessly. The moment you try to name a feeling, much of it disappears. Poets and novelists have wrestled with this problem for centuries. Text is inherently lossy. It is often too explicit and too logical to capture emotional texture. As Marcel Proust wrote, "We feel in one world, we think and name in another."
This project explores music as a new interface for image editing. Think of it as color correction driven by sound.
A song reshapes a photograph to match its emotional character: warmer or cooler, brighter or darker, grainy or clean, crisp or dreamy. Instead of relying on text prompts or AI interpretations, the system analyzes the physical properties of sound: its energy, brightness, texture, tonality, rhythm, and spatial breadth. Those continuous, multidimensional signals are mapped directly to visual parameters, preserving the nuance of the waveform all the way to the pixels.
AI is deliberately kept out of the creative loop. A vision model is used only at the edges as a guardrail. Its job is to keep the music's intent from breaking the photograph, never to author the mood.
Color correcting with music. It makes image editing more intuitive, more expressive, and, in the process, a lot more fun.