FreeInit: Bridging Initialization Gap in Video Diffusion Models

TL;DR

This paper identifies an implicit gap in noise initialization between training and inference in video diffusion models, and proposes FreeInit, a strategy that refines initial noise to improve temporal consistency without extra training.

Contribution

The paper introduces FreeInit, a novel inference sampling method that addresses the initialization gap in video diffusion models, enhancing temporal consistency and quality.

Findings

FreeInit improves temporal consistency in generated videos.

Refining low-frequency components during inference enhances quality.

The method works across various text-to-video diffusion models.

Abstract

Though diffusion-based video generation has witnessed rapid progress, the inference results of existing models still exhibit unsatisfactory temporal consistency and unnatural dynamics. In this paper, we delve deep into the noise initialization of video diffusion models, and discover an implicit training-inference gap that attributes to the unsatisfactory inference quality.Our key findings are: 1) the spatial-temporal frequency distribution of the initial noise at inference is intrinsically different from that for training, and 2) the denoising process is significantly influenced by the low-frequency components of the initial noise. Motivated by these observations, we propose a concise yet effective inference sampling strategy, FreeInit, which significantly improves temporal consistency of videos generated by diffusion models. Through iteratively refining the spatial-temporal low-frequency components of the initial latent during inference, FreeInit is able to compensate the initialization gap between training and inference, thus effectively improving the subject appearance and temporal consistency of generation results. Extensive experiments demonstrate that FreeInit consistently enhances the generation quality of various text-to-video diffusion models without additional training or fine-tuning.