surface
A noisy particle field begins without a predetermined picture, target, or correct answer.
signal / live study
FEEDBACK FIELD
Find coherence in a field that does not contain a hidden image.
Drag, touch, or use the keyboard to tune the field. Structure appears because your behavior creates enough coherence for the system to begin interpreting itself. Enable local perception and a small browser model periodically changes the rules according to what it thinks the field resembles.
drag or touch to tune · arrow keys move the tuner · space introduces signal
noise field ready
coherence00%
entropy100%
machine belief—
fieldnoise
Local perception is optional. When enabled, the model downloads to this browser and examines only the generated Feedback Field canvas. No camera, microphone, prompt, or uploaded image is used.
under the surface
The interpretation changes what is being interpreted.
disturbance
Your movement adds local coherence. The same field can become radial, bilateral, flowing, vertical, or cellular.
recognition
When local perception is on, a tiny image classifier periodically reads the canvas and its belief selects the next attractor family.
feedback
The machine's interpretation changes the physics, which changes the next image it interprets. The loop is the medium.
system
Small intelligence. Large behavior.
- simulation
- Canvas 2D particle field
- perception
- MobileViT Small / quantized ONNX
- runtime
- Transformers.js / browser only
- model download
- about 6 MB, only after opt-in
- server inference
- none
- status
- experimental / mobile-first v0
The first build intentionally uses classification rather than a larger VLM. If the interaction earns the extra weight, a later device-tier can add SmolVLM-class semantic interpretation without making it the baseline experience.