Generative Pre-trained Autoregressive Diffusion Transformer

Bo Wang (Sensetime) · Yuan Zhang (Ohio State University, Columbus) · Jiacheng Jiang (Tsinghua University) · Guoqing Ma (StepFun) · Zhiying Lu (University of Science and Technology of China) · Haoyang Huang (Microsoft Research Asia) · Jianlong Yuan (ByteDance Inc.) · Nan Duan (StepFun)

autoregressive modelingcausal attention variantcontinuous latent spacediffusion lossfew-shot learning tasksgenerative pre-trained autoregressive diffusion transformerinference efficiencylong-range video synthesismotion dynamicsparameter-free rotation-based time-conditioningrepresentation capabilitiessemantic consistency training efficiencyvideo generation qualityvideo representation ability

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.