Training-Free Motion Customization for Distilled Video Generators with Adaptive Test-Time Distillation
This work addresses a specific bottleneck in video generation for users needing efficient, customizable motion control, representing an incremental advancement over existing training-free methods.
The paper tackles the problem of motion customization in distilled video generators without requiring additional training, achieving significant improvements in motion fidelity and generation quality while maintaining high efficiency.
Distilled video generation models offer fast and efficient synthesis but struggle with motion customization when guided by reference videos, especially under training-free settings. Existing training-free methods, originally designed for standard diffusion models, fail to generalize due to the accelerated generative process and large denoising steps in distilled models. To address this, we propose MotionEcho, a novel training-free test-time distillation framework that enables motion customization by leveraging diffusion teacher forcing. Our approach uses high-quality, slow teacher models to guide the inference of fast student models through endpoint prediction and interpolation. To maintain efficiency, we dynamically allocate computation across timesteps according to guidance needs. Extensive experiments across various distilled video generation models and benchmark datasets demonstrate that our method significantly improves motion fidelity and generation quality while preserving high efficiency. Project page: https://euminds.github.io/motionecho/