VideoCanvas

Définition

xxxxx

Français

VideoCanvas

Anglais

VideoCanvas

Unified Video Completion from Arbitrary Spatiotemporal Patches via In-Context Conditioning

A unified framework for arbitrary spatio-temporal video completion that allows users to place content patches at any location and timestamp in a video, with the model filling in the remaining regions. This approach unifies many existing video generation tasks like image-to-video, video inpainting, and extension under a single paradigm. The method addresses fundamental challenges in modern video diffusion models, particularly the temporal ambiguity introduced by causal VAEs that compress multiple frames into single latent representations.
VideoCanvas addresses temporal ambiguity in latent video diffusion models to enable flexible spatio-temporal video completion using a hybrid conditioning strategy.

Sources

Sources : huggingface