NEvo: AI Generates Videos Optimized to Maximally Drive Specific Brain Regions in Visual Cortex

What happened

EPFL Schrimpf Lab researchers published the NEvo preprint: Neural-Guided Evolutionary Video Synthesis for Dynamic Visual Selectivity (arXiv 2607.02317). The work received 98 points on Hacker News and ranked among the top AI research stories of the day.

Context and impact

NEvo combines AI video generation with neuroscientific models of cortical activation. The system performs evolutionary search over a structured video-prompt space, predicting voxel-level activation via a dynamic encoding model. Synthesized videos consistently outperform handcrafted localizer videos. The technique could accelerate brain-computer interface research and deepen understanding of visual perception and temporal dynamics.

Details

• Framework: evolutionary search over a structured video-prompt space
• Goal: maximize activation of specific regions across visual cortex
• Covers: ventral, dorsal, and lateral pathways
• Result: NEvo videos outperform hand-crafted localizer videos
• Reveals: systematic differences in visual cortex sensitivity to temporal dynamics
• HN score: 98 points (July 10, 2026)