KlingAI Research (Yiyan Ji et al.) · Agentic Code2Video Framework

Coda-DaVinci: Reimagining Video Generation with Code2Video

Coda-DaVinci is a task-adaptive agentic framework for Code2Video generation, which guides video generation with agent-authored 3D proxies. The authors study proxy granularity, physics simulation and prompting through controlled experiments and blinded human evaluation, finding that coarse proxies favour natural motion while medium-grained proxies favour consistency. The paper reports leading instruction following on VTR and scene consistency on R2M, plus better physical plausibility and visual realism than the compared Code2Video baselines. It also notes limitations in transition stability and adherence to scene settings over long videos.

Open the source ↗

Similar Video & Animation releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.