‹ back to 2026-07-25

ElasticTTT: Prior-Preserving Test-Time Tuning for Video Editing

arXiv cs.CV Multimodal & Generative Yueyi Liu, Chi Zhang, Sen Cui, Miao Liu 2026-07-23

TL;DR - ElasticTTT is a test-time tuning framework for one-shot video editing that prevents pretrained diffusion models from collapsing toward the source video. It preserves the model’s generative prior while improving edit fidelity and protecting unchanged regions.

Identifies “Prior Collapse,” where standard single-point optimization causes lost text conditioning, source-video memorization, or entangled regional features.
Uses Target Distribution Regularization to discourage sharp memorization minima.
Introduces Contrastive CFG to steer generation away from source bias.
Applies an Asynchronous Noise Schedule to preserve unedited regions and reports state-of-the-art performance in extensive evaluations.

view merged work →