Introduction
Text-guided video editing with diffusion models is often slow due to costly multi-step sampling and inversion. We present OSVE, the first framework to adapt one-step Text-to-Image (T2I) models for high-quality video editing, addressing core challenges like inversion, editability, and temporal consistency.
Key Technologies
To bypass slow iterative inversion, we train a learnable encoder that predicts the initial noise for each frame in a single forward pass. This encoder is trained with a novel Structure-Aware Editing (SAE) loss on a curated dataset of structurally-aligned image pairs, ensuring the preservation of the source video's geometry during edits.
For temporal coherence, we introduce Unified-Frame Editing (UFE), which concatenates frame latents for cross-frame attention in a single generation step. Additionally, a sliding-window strategy with an anchor frame maintains global consistency for long videos.
Extensive experiments show that OSVE achieves editing quality comparable or superior to state-of-the-art multi-step methods, operating approximately 155 to 171 times faster. This breakthrough paves the way for practical, real-time video editing applications.
Code is available at GitHub.
Blogger's Review: The OSVE framework significantly enhances video editing efficiency and quality through innovative encoding and editing techniques, showcasing the immense potential of diffusion models in practical applications. This advancement not only provides powerful tools for video creators but may also lead to a transformation in the video editing industry.