FilmFlow plans 3D camera paths from text or reference video
Original title: FilmFlow: Agentic System for 3D Virtual Film Cinematography
CinemaBench covers 150 Blender scenes, 750 text-directed tasks, and 60 reference videos.
FilmFlow is an agentic framework that turns natural-language instructions or reference videos into explicit camera trajectories in dynamic 3D scenes. A Director agent builds executable plans from concurrent atomic operations for translation, rotation, and framing, with Interpreter, Scout, and Critic agents for intent, viewpoints, and visual feedback. The authors introduce CinemaBench: 150 Blender scenes, five instructions per scene (750 text tasks), and 60 reference videos on a subset. They report the strongest scores and highest user preference among the baselines they evaluated.
- Director, Interpreter, Scout, and Critic agents split planning and review.
- Atomic operations jointly control translation, rotation, and framing.
- CinemaBench has 150 Blender scenes from self-created and licensed sources.
- The benchmark covers text-directed and video-reference camera generation.
- Authors claim top benchmark scores and highest user preference.
From the original article
1 min read at huggingface.co
Complex camera motion is central to visual storytelling, yet translating cinematographic intent into controllable, executable camera trajectories remains challenging. We introduce FilmFlow, an agentic framework that converts natural-language instructions or reference videos into explicit camera trajectories for dynamic 3D environments.
By Yiming Zhang, Jiarui Guo, Haojia Wei · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.