Picbreeder‑style (Kenneth Stanley) CPPN Playground (RGB)

Mutate or breed CPPN images. In Breed mode, selecting 2 images automatically spawns the next generation.

✨ Mutate 🧬 Breed

Population 9 Generation 0 Selected 0

Quick controls: Mutate mode: click one image to mutate it immediately. Breed mode: select two images to automatically breed + mutate offspring and spawn the next generation. Use Next generation anytime to spawn from the current selection.

Settings

These parameters control how new images are generated. Hover tooltips on buttons/cells also contain useful details.

Evolution

Mutation sigma (σ) controls how strongly weights/biases are perturbed. Higher values explore faster but can destroy structure; lower values refine gradually.

Mutation sigma (σ)

Typical range: 0.05–0.30

Keep elites

Copy the first N individuals unchanged into the next generation. This is a simple “pin favorites” approximation.

Display

Upscale affects how large each 100×100 canvas appears on screen. It does not change the underlying CPPN resolution.

Upscale factor

Grid size

Export / Import

Save your current population (including weights, biases, and activation choices) as JSON. Paste JSON back in to restore.

JSON

What you are seeing

Each image is generated by a CPPN with inputs (x, y, d) where x,y ∈ [-1,1] and d = √(x²+y²). The CPPN outputs (R,G,B), each squashed with a sigmoid into 0–255. Hidden neurons use a mix of transfer functions: sin, cos, gauss, id, sigmoid.

Note: Picbreeder uses CPPN‑NEAT with innovation-number crossover. This demo uses a fixed feedforward topology and performs uniform crossover over parameters.