Replicating Picbreeder Using Large Vision Models

SakanaAILabs · x · 2026-07-12

Sakana AI shared their attempt to replicate Picbreeder (2008): using AI agents / Large Vision-Language Models to "breed" images, recreating the early collaborative, open-ended generative website.

They built a tribute page where images can be branched, scored, and inspected, with experimental results continuously evolving in this manner. The post also links to the relevant blog and GECCO 2026 paper, titled "In Search of the Ingredients of Open-Endedness: Replicating Picbreeder with Large Vision-Language Models", noting it as a joint project with MIT and NYU.

Related event: Sakana AI Replicates Picbreeder with VLM Agents for Open-Ended Creativity(14 posts)→

Original post →

More from Multimodal

Multimodal channel →