NEvo, a new AI framework, evolves video content to maximally activate predefined visual‑brain regions. It trains a detailed encoding model—a digital twin—that predicts regional responses to any visual stimulus. Researchers then ask the twin: which video will light up a target area the most?

NEvo treats video attributes—subject, lighting, motion, mood—as genetic loci. A population of them is generated, scored by the twin, retained, crossed, and mutated across generations. Each iteration nudges predicted activation upward, converging on a high‑response stimulus.

The tool adopts a two‑step procedure to reduce compute costs. First, it hunts for a single still image that elicits peak activation functions. Next, it animates that frame into a 2‑second clip by optimizing motion parameters, producing a fully dynamic stimulus.

Across the lateral stream, NEvo’s videos surpass handcrafted localizers and natural footage, showing stronger activation for every region. The method maps a gradient from low‑level pattern detectors in V1/V3A to high‑level social‑dynamic hubs in pSTS/aSTS, revealing how visual selectivity evolves into a social domain.