Does Semantic Noise Initialization Transfer from Images to Videos? A Paired Diagnostic Study

Yixiao Jing, Chaoyu Zhang, Zixuan Zhong, Peizhou Huang · 2026

Read Does Semantic Noise Initialization Transfer from Images to Videos? A Paired Diagnostic Study by Yixiao Jing, Chaoyu Zhang, Zixuan Zhong, Peizhou Huang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-06672.txt.

Description

Jing and colleagues test whether semantic noise initialization transfers from image diffusion to text-to-video generation, despite coupling instability. They benchmark it against noise on a VideoCrafter-style backbone with VBench on 100 prompts, using paired tests and a sign-flip permutation test. They observe a positive trend on temporal dimensions, but the 95 percent confidence interval includes zero, with p around 0.17, and score stays on par with baseline. Noise-space analysis finds weak or unstable signal, and the authors recommend evaluation and noise-space diagnostics as practice.

Catalog record

CREATORYixiao Jing, Chaoyu Zhang, Zixuan Zhong, Peizhou Huang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21480
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.06672

Browse the full library · About the Tower of Babel