Video generation gets layout and motion control without retraining
Latent Space Jam: Layout Guided Video Synthesis
Computer Vision and Pattern Recognition
Summary
Making videos where you control where objects are and how they move usually needs lots of extra training and computing power. The authors found a way to guide video creation just by telling the computer roughly where objects should be and how they should move, using a pre-trained video model. They do this in two steps: first, creating a rough guide for object placement and motion, then generating the final video based on that guide. This method works well for videos with different objects and movements without needing to train new models.
What this means in practice
- •For video production teams: Create videos with specific object layouts and natural motion without retraining models for each new scenario.
- •For game developers: Generate dynamic video scenes with controlled object trajectories to prototype or produce game animations faster.
Authors
Asaf Joseph, Moshe Mandel, Ohad Fried
Abstract
Adding control to video generation often requires training, which demands additional computational resources, data, or architectural changes. We propose a training-free approach that utilizes pretrained video diffusion models to control object layout and motion using coarse layouts. Our method operates in two passes: first, it steers spatiotemporal placement through prompt-guided cross-attention to produce a coarse visual guide; second, this guide conditions the same model to generate a high-quality, layoutconsistent video. The approach enables coordinated motion with strong temporal consistency, supporting diverse trajectories and object interactions without additional training. Extensive experiments show that our method consistently outperforms prior work in overlap and trajectory accuracy, as well as visual quality across static, dynamic, and multi-object scene settings.