Generative Pre-trained Autoregressive Diffusion Transformer

TL;DR

GPDiT is a novel model that combines diffusion and autoregressive techniques in a continuous latent space for high-quality, long-range video synthesis, with improved efficiency and representation capabilities.

Contribution

It introduces a unified diffusion-autoregressive framework for video generation, along with lightweight attention and rotation-based time-conditioning mechanisms.

Findings

Achieves high-quality long-range video synthesis.

Demonstrates strong video representation and few-shot learning performance.

Improves training and inference efficiency through novel mechanisms.

Abstract

In this work, we present GPDiT, a Generative Pre-trained Autoregressive Diffusion Transformer that unifies the strengths of diffusion and autoregressive modeling for long-range video synthesis, within a continuous latent space. Instead of predicting discrete tokens, GPDiT autoregressively predicts future latent frames using a diffusion loss, enabling natural modeling of motion dynamics and semantic consistency across frames. This continuous autoregressive framework not only enhances generation quality but also endows the model with representation capabilities. Additionally, we introduce a lightweight causal attention variant and a parameter-free rotation-based time-conditioning mechanism, improving both the training and inference efficiency. Extensive experiments demonstrate that GPDiT achieves strong performance in video generation quality, video representation ability, and few-shot learning tasks, highlighting its potential as an effective framework for video modeling in continuous space.