ElasticTTT: Prior-Preserving Test-Time Tuning for Video Editing
Updated
Updated · arxiv.org · Jul 23
ElasticTTT: Prior-Preserving Test-Time Tuning for Video Editing
1 articles · Updated · arxiv.org · Jul 23
Summary
Researchers have introduced ElasticTTT, a new framework to improve test-time tuning for video editing using diffusion models.
ElasticTTT addresses prior collapse by introducing Target Distribution Regularization, Contrastive Classifier-Free Guidance, and Asynchronous Noise Scheduling to preserve generative flexibility.
The method achieves state-of-the-art results on one-shot video editing, offering improved instruction adherence and video quality without sacrificing source preservation.