ViD-GPT: Introducing GPT-style Autoregressive Generation in Video Diffusion Models

TL;DR

ViD-GPT introduces a causal, autoregressive approach with GPT-style generation and kv-cache acceleration to improve long video generation quality and efficiency in diffusion models.

Contribution

It pioneers the integration of GPT-style causal attention and prompt-based conditioning into video diffusion models for long video synthesis.

Findings

Achieves state-of-the-art results in long video generation.

Significantly improves inference speed with kv-cache mechanism.

Demonstrates superior qualitative and quantitative performance.

Abstract

With the advance of diffusion models, today's video generation has achieved impressive quality. But generating temporal consistent long videos is still challenging. A majority of video diffusion models (VDMs) generate long videos in an autoregressive manner, i.e., generating subsequent clips conditioned on last frames of previous clip. However, existing approaches all involve bidirectional computations, which restricts the receptive context of each autoregression step, and results in the model lacking long-term dependencies. Inspired from the huge success of large language models (LLMs) and following GPT (generative pre-trained transformer), we bring causal (i.e., unidirectional) generation into VDMs, and use past frames as prompt to generate future frames. For Causal Generation, we introduce causal temporal attention into VDM, which forces each generated frame to depend on its previous frames. For Frame as Prompt, we inject the conditional frames by concatenating them with noisy frames (frames to be generated) along the temporal axis. Consequently, we present Video Diffusion GPT (ViD-GPT). Based on the two key designs, in each autoregression step, it is able to acquire long-term context from prompting frames concatenated by all previously generated frames. Additionally, we bring the kv-cache mechanism to VDMs, which eliminates the redundant computation from overlapped frames, significantly boosting the inference speed. Extensive experiments demonstrate that our ViD-GPT achieves state-of-the-art performance both quantitatively and qualitatively on long video generation. Code will be available at https://github.com/Dawn-LX/Causal-VideoGen.