AI-Generated Videos That Maximally Stimulate Specific Brain Regions

A technology called NEvo has been unveiled that builds a digital twin of the brain (Encoding Model) to automatically find the video that most strongly activates a specific visual region.

How NEvo Works:

• Using a Digital Twin: Based on a model that predicts brain responses, it sets a reward function to search for the optimal video.
• Evolutionary Algorithm: It evolves prompts that increase brain activity by crossing over and mutating 'genes' such as subject, lighting, movement, and mood.
• Two-Stage Optimization: It first finds the most powerful still image, then animates it into a moving 2-second video.

Key Research Findings:

• Region-Specific Stimulation: It succeeded in generating videos optimized for the characteristics of each brain region, including FFA (faces), PPA (places), EBA (bodies), and MT (motion).
• Effect of Dynamic Stimuli: The generated videos induced higher brain activity than existing hand-crafted clips or natural footage, and videos with motion showed a higher response than still images.
• Mapping Social Characteristics: Analysis along the visual pathway confirmed a gradient of Social Dynamics, starting from simple patterns and gradually progressing to faces and social interactions.