Skip to content

Dataset · picbreeder-vlm

VLM-agent image evolution archive

picbreeder-vlm/picbreeder-vlm-archive

Generated images, genomes, and VLM agent reasoning logs from the Picbreeder study.

What's inside

Contains images generated by a swarm of multimodal models (VLMs), the CPPN genomes they were derived from, English reasoning transcripts of the agents, lineage graphs, and analysis artifacts. The dataset size ranges from 100,000 to 1 million records. The data was collected as part of the GECCO 2026 research work replicating the Picbreeder crowdsourcing project using neural networks.

What you can do with it

  • Studying reasoning chains and visual selection in VLM agents
  • Training or analyzing evolutionary image generation systems
  • Researching open-endedness in multimodal models
Tasks
Image → text
Languages
English
Size
100K–1M rows
Downloads this month
125.1K
Likes
7

How to load it

from datasets import load_dataset

ds = load_dataset("picbreeder-vlm/picbreeder-vlm-archive")