English

Diffusion Transformer-to-Mamba Distillation for High-Resolution Image Generation

Computer Vision and Pattern Recognition 2025-06-25 v1

Abstract

The quadratic computational complexity of self-attention in diffusion transformers (DiT) introduces substantial computational costs in high-resolution image generation. While the linear-complexity Mamba model emerges as a potential alternative, direct Mamba training remains empirically challenging. To address this issue, this paper introduces diffusion transformer-to-mamba distillation (T2MD), forming an efficient training pipeline that facilitates the transition from the self-attention-based transformer to the linear complexity state-space model Mamba. We establish a diffusion self-attention and Mamba hybrid model that simultaneously achieves efficiency and global dependencies. With the proposed layer-level teacher forcing and feature-based knowledge distillation, T2MD alleviates the training difficulty and high cost of a state space model from scratch. Starting from the distilled 512×\times512 resolution base model, we push the generation towards 2048×\times2048 images via lightweight adaptation and high-resolution fine-tuning. Experiments demonstrate that our training path leads to low overhead but high-quality text-to-image generation. Importantly, our results also justify the feasibility of using sequential and causal Mamba models for generating non-causal visual output, suggesting the potential for future exploration.

Keywords

Cite

@article{arxiv.2506.18999,
  title  = {Diffusion Transformer-to-Mamba Distillation for High-Resolution Image Generation},
  author = {Yuan Yao and Yicong Hong and Difan Liu and Long Mai and Feng Liu and Jiebo Luo},
  journal= {arXiv preprint arXiv:2506.18999},
  year   = {2025}
}
R2 v1 2026-07-01T03:30:07.844Z