Training-Free Motion Customization for Distilled Video Generators with Adaptive Test-Time Distillation

TL;DR

MotionEcho is a training-free, test-time distillation method that enhances motion customization in distilled video generators by leveraging slow teacher models, improving motion fidelity without sacrificing efficiency.

Contribution

We introduce MotionEcho, a novel framework that enables training-free motion customization in distilled video models through adaptive test-time distillation using teacher forcing.

Findings

Significantly improves motion fidelity in distilled video generation.

Maintains high efficiency comparable to existing fast models.

Demonstrates effectiveness across various models and datasets.

Abstract

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/