HARIVO: Harnessing Text-to-Image Models for Video Generation

TL;DR

HARIVO introduces a novel architecture that leverages pretrained text-to-image models for generating temporally consistent videos, enhancing realism and diversity while requiring limited video data.

Contribution

It advances existing methods by integrating a mapping network and frame-wise tokens, along with new loss functions and gradient sampling techniques for improved video quality.

Findings

Achieved realistic, temporally smooth video generation

Successfully integrated video-specific inductive biases

Built on frozen StableDiffusion for simplified training

Abstract

We present a method to create diffusion-based video models from pretrained Text-to-Image (T2I) models. Recently, AnimateDiff proposed freezing the T2I model while only training temporal layers. We advance this method by proposing a unique architecture, incorporating a mapping network and frame-wise tokens, tailored for video generation while maintaining the diversity and creativity of the original T2I model. Key innovations include novel loss functions for temporal smoothness and a mitigating gradient sampling technique, ensuring realistic and temporally consistent video generation despite limited public video data. We have successfully integrated video-specific inductive biases into the architecture and loss functions. Our method, built on the frozen StableDiffusion model, simplifies training processes and allows for seamless integration with off-the-shelf models like ControlNet and DreamBooth. project page: https://kwonminki.github.io/HARIVO