kapynResearch

Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning

Internalized Visual Thinking is a post-training framework for proactive video reasoning. It jointly optimizes textual prediction and visual thinking during training, so models can reason directly at inference without generating intermediate images. That removes Visual CoT's overhead while preserving spatial and temporal reasoning.

Apple ML Research·Aug 24, 2026

Opening Kapyn…