Train a real AI model and watch it collapse into a single pixel. The pixel's RGB values are the model's weights. Download it as a 1x1 PNG. Load it back. Run predictions. The pixel is the AI.
Training Data
Train
Load a Pixel
Drag a 1x1 PNG here
or click to browse
Examples
What is this?
A real trained model
This runs actual gradient descent with sigmoid activation and binary cross-entropy loss. The model learns weights from your data through real optimization, not hardcoded rules.
Encoded in RGB
A 2-input classifier has 3 parameters: two weights and a bias. Each is quantized to 8 bits and mapped to one color channel. R = weight 1, G = weight 2, B = bias. The pixel is the model.
What you see above
Each dot is one example answer to a yes-or-no question. Training draws the best dividing line between the yes-dots and the no-dots, then packs what it learned — just three numbers — into a single pixel's color. That pixel is the entire AI.
A note about limits
A single neuron draws a straight line. It can't learn XOR or any pattern that isn't linearly separable. Try the XOR example above to see it fail. That's why deeper networks exist.