Artbreeder

Artbreeder didn't start as an image generator in the Midjourney sense. It began as Ganbreeder, an experiment by Joel and Studio Morphogen in using breeding and collaboration to explore what the site calls "high complexity spaces." The name is a nod to Picbreeder, a research project that looked at the role of exploration in optimization, and the about page says it was also inspired by Joel's earlier project, Facebook Graffiti, which showed the creative capacity of crowds. That history explains the product: you're meant to wander, mix and build on other people's images, not just type a prompt and hope.

The community idea isn't marketing copy either. The about page credits specific contributors. Gwern made the anime image model. Peter Baylies made the painting model. Alex Reben built the dataset for the albums model, Max Hawkins made the Artbreeder Bot, and Arfa and Xander contributed the furry and science bio art models. The original engine runs on Generative Adversarial Networks, specifically BigGAN and StyleGAN, and there's a minimal open source version built on BigGAN if you want to look under the hood. Early backers included Stochastic Labs, DeepAI and TensorFlow.

The current product has moved past its GAN roots. The pricing page sells credits measured in SDXL images and highlights tools like Composer and Splicer. Paid plans start at $8.99/mo, or $7.49/mo if you pay yearly for 1,200 credits. Our take: Artbreeder is a good fit if you like finding an image through iteration and remixing rather than prompt engineering. The catch is that the public documentation is thin and the about page reads like it was written in the Ganbreeder era, so you'll learn most of what the current tools do by signing up and poking around.