CogOmniControl: Reasoning-Driven Controllable Video Generation via Creative Intent Cognition

Researchers introduced CogOmniControl, a reasoning-first framework for controllable video generation.

It splits intent understanding from video synthesis by training a specialized reasoning model on anime production data and aligning a diffusion generator to those outputs.

The team also released CogReasonBench and CogControlBench, benchmarks built from real production workflows.

Results show stronger performance than existing open-source models, especially for abstract and sparse creative inputs.

The work narrows a key gap between user intent and usable AI video tools for production settings.