Pix2Video: Video Editing using Image Diffusion (duyguceylan.github.io)

Summary:

• Large scale text to image generation models are being used for editing videos in concurrent work.
• Some of these methods include Video-P2P, Fate-Zero, Tune-A-Video, and Gen-1.

Key terms:

• Large scale text to image generation models: Models that generate images from text and are used for image editing tasks
• Video-P2P: A method that extends null inversion to video clips and adapts a cross attention control mechanism
• Fate-Zero: A training-free strategy for editing videos that utilizes cross attention maps to compute blending masks
• Tune-A-Video: A method that finetunes the image generation model given an input video and uses cross-frame attention for consistent edits
• Gen-1: A large scale video generation model that uses depth as a structural cue and is trained on a mixed dataset of images and videos

Tags:

Research arXiv Video Editing Image Diffusion Concurrent Work Video-P2P Fate-Zero Tune-A-Video Gen-1 Large Scale Text To Image Generation