We introduce Hybrid Autoregressive Transformer (HART), an autoregressive (AR) visual generation model capable of directly generating 1024x1024 images, rivaling diffusion models in image generation quality. Existing AR models face limitations due to the poor image reconstruction quality of their discrete tokenizers and the prohibitive training costs associated with generating 1024px images. To address these challenges, we present the hybrid tokenizer, which decomposes the continuous latents from the autoencoder into two components: discrete tokens representing the big picture and continuous tokens representing the residual components that cannot be represented by the discrete tokens. The discrete component is modeled by a scalable-resolution discrete AR model, while the continuous component is learned with a lightweight residual diffusion module with only 37M parameters. Compared with the discrete-only VAR tokenizer, our hybrid approach improves reconstruction FID from 2.11 to 0.30 on MJHQ-30K, leading to a 31% generation FID improvement from 7.85 to 5.38. HART also outperforms state-of-the-art diffusion models in both FID and CLIP score, with 4.5-7.7x higher throughput and 6.9-13.4x lower MACs. Our code is open sourced at https://github.com/mit-han-lab/hart.
@article{arxiv.2410.10812,
title = {HART: Efficient Visual Generation with Hybrid Autoregressive Transformer},
author = {Haotian Tang and Yecheng Wu and Shang Yang and Enze Xie and Junsong Chen and Junyu Chen and Zhuoyang Zhang and Han Cai and Yao Lu and Song Han},
journal= {arXiv preprint arXiv:2410.10812},
year = {2024}
}
Comments
Demo: https://hart.mit.edu. The first two authors contributed equally to this work