Eternal Haven

Home · LYGO Resonance

Room · image to sound

LYGO Resonance

Point it at a picture and it answers with a soundscape or a creative brief. Useful as a starting sketch; not a composer, and it says so.

Justin Helmer
Room guide · updated 6 November 2026 · about a seven-minute read

There is a long tradition of turning pictures into music, and most of it is worse than the idea deserves. The usual failure is literal mapping: brightness becomes pitch, edges become rhythm, and the result sounds like a spreadsheet with a reverb. Resonance is an attempt to do the mapping with taste, and to be clear about where taste ends and the machine stops.

The live instrument is LYGO Resonance. What follows is what it actually does, in the order it does it.

Reading the image

First the picture is measured, not understood. Edges, contours, keypoints, dominant colours, the distribution of light across the frame, and a rough sense of texture from local contrast. None of this is semantic. The tool does not know it is looking at a face or a coastline; it knows density and direction and warmth.

That limitation is also the feature. A system that recognised "sunset" would produce the sunset it has seen in training. A system that measures an orange gradient and a low horizon produces something specific to your file, including the parts that are badly exposed. Specificity is worth more than recognition in a sketch.

Building the sound

The mapping is layered rather than one-to-one. A slow harmonic bed follows the large colour regions. A mid layer of filtered noise follows texture. Transient events — hard edges, bright points — trigger short percussive gestures. Contour direction sets the movement between layers, which is why two images with similar palettes can produce different pieces.

Then there is the profile mode. Instead of audio, the tool emits a structured description: mood words, tempo suggestions, instrumentation, a brief that a musician or a text-to-music model can start from. I use this more than the audio. A brief is easier to argue with, and arguing is how a sketch becomes a piece.

What it is not

It is not a composer. It does not develop a theme, resolve a phrase, or know when to stop. Left alone it will happily produce four minutes of texture with no argument in it. The output is a beginning: a bed to write over, a palette to steal from, a way to get past the blank page.

It is also not a substitute for listening. The most common mistake with a generative tool is to accept the first result because it exists. Play it twice. If the second listen is boring, the piece is boring, and no amount of visual justification fixes that. This is the same rule as Empty beats fake: a thing that sounds finished is not the same as a thing that is good.

The audio question

Everything here is generated locally in the browser or with a local model. Nothing you drop on the page is uploaded, which matters for pictures of people and pictures you do not own. If you are scoring something commercial, treat the output as a sketch you own the arrangement of, not as a sample library cleared for release.

Why an image-to-sound tool belongs in a fiction house

Because the books are built out of images that do not exist yet, and a saga needs a way to hear a scene before it is written. I have used this to find the tone of a chapter more than once: feed it a painting, listen to the bed, write toward it. That is not a workflow I would recommend to everyone. It is a workflow I would defend to anyone who has stared at a blank page at midnight.

If you want the music it feeds, the free player is asiancoastline.com and the explanation is Music in this house. If you want the wider map of rooms, the tools room is the door.

Justin Helmer

Writes the Eternal Haven saga and the essays on this domain, publishes music as Excavationpro, and keeps the lantern on at /start.html.