Rebalancing Reference Frame Dominance to Improve Motion in Image-to-Video Models

arXiv:2605.19398 · cs.CV, cs.AI · Submitted 2026-05-19 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-05-19

Updated: 2026-09-25

Comments: Accepted to NeurIPS 2026. Project page: https://sh0xed98b8.github.io/DyMoS/

Code: https://github.com/black-forest-labs/flux

License: http://creativecommons.org/licenses/by/4.0/

The gist: Image-to-video models often generate videos that remain overly static, compared to text-to-video models.

Terminology

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.

Sources

Related papers