DyMoS：无需训练即可提升图像到视频模型运动表现

View PDF HTML (experimental)

Abstract:Image-to-video models often generate videos that remain overly static, compared to text-to-video models. While prior approaches mitigate this issue by weakening or modifying the image-conditioning signal, they often require additional training or sacrifice fidelity to the reference image. In this work, we identify reference-frame dominance as a key mechanism behind motion suppression. We observe that non-reference frames in I2V models allocate excessive self-attention to reference-frame key tokens, causing reference information to be over-propagated across time and suppressing inter-frame dynamics. Based on this finding, we propose DyMoS (Dynamic Motion Slider), a training-free and model-agnostic method that rebalances the attention pathway from generated frames to the reference frame during initial denoising steps. DyMoS leaves both the input image and model weights unchanged and introduces a single scalar parameter for continuous control over motion strength. Experiments across multiple state-of-the-art I2V backbones demonstrate that DyMoS consistently improves motion dynamics while maintaining visual quality and fidelity to the reference image.

Comments:	Accepted to NeurIPS 2026. Project page: this https URL
Subjects:	Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2605.19398 [cs.CV]
	(or arXiv:2605.19398v5 [cs.CV] for this version)
	https://doi.org/10.48550/arXiv.2605.19398 arXiv-issued DOI via DataCite

Submission history

From: Wooseok Jeon [view email]
[v1] Tue, 19 May 2026 05:50:24 UTC (6,059 KB)
[v2] Wed, 20 May 2026 05:07:06 UTC (6,059 KB)
[v3] Wed, 3 Jun 2026 07:04:53 UTC (6,060 KB)
[v4] Fri, 25 Sep 2026 12:23:35 UTC (6,060 KB)
[v5] Mon, 26 Oct 2026 05:03:41 UTC (6,060 KB)