CVPR2024

Peekaboo: Interactive Video Generation via Masked-Diffusion

Yash Jain, Anshul Nasery, Vibhav Vineet, Harkirat S. Behl

摘要

Figure 1. Zero-training No-latency interactive video generation. PEEKABOO allows users to control the output (object size, location and trajectory) for any off-the-shelf video diffusion models, through specially designed masking modules. First row shows a panda playing PEEKABOO by following an expanding mask in left direction.