MotiMotion: Motion-Controlled Video Generation with Visual Reasoning

Researchers introduced MotiMotion, a reasoning-then-generation framework for motion-controlled video generation.

It uses a vision-language reasoner to refine motion inputs, add plausible secondary motions, and tune control strength by confidence.

The team also released MotiBench, a benchmark for interaction-centered image-to-video generation.

Evaluations show MotiMotion produces more plausible interactions than existing methods.