icon Glance : Accelerating Diffusion Models with 1 Sample

Zhuobai Dong1 *, Rui Zhao2 *, Songjie Wu3, Junchao Yi4, Linjie Li5,
Zhengyuan Yang5, Lijuan Wang5, Alex Jinpeng Wang3 †,
1WuHan University,  2National University of Singapore
3Central South University,  4University of Electronic Science and Technology of China,  5Microsoft
*Indicates Equal Contribution, †Corresponding author
 
Paper   Code   Hugging Face Model
Qwen-Image or Glance — Can you spot which model generated each image? Click to find out!
🔍 Click to zoom in

Figure 1: Comparison of data usage and training time.. Glance achieves comparable generation quality with only 1 training samples and within 1 GPU-hour, demonstrating extreme data and compute efficiency. Note that the x-axis is in logarithmic scale, and values equal to zero are therefore not representable.
Abstract: Diffusion models have achieved remarkable success in image generation, yet their deployment remains constrained by the heavy computational cost and the need for numerous inference steps. Previous efforts on fewer-step distillation attempt to skip redundant steps by training compact student models, yet they often suffer from heavy retraining costs and degraded generalization. In this work, we take a different perspective: we accelerate smartly, not evenly, applying smaller speedups to early semantic stages and larger ones to later redundant phases. We instantiate this phase-aware strategy with two experts that specialize in slow and fast denoising phases. Surprisingly, instead of investing massive effort in retraining student models, we find that simply equipping the base model with lightweight LoRA adapters achieves both efficient acceleration and strong generalization. We refer to these two adapters as Slow-LoRA and Fast-LoRA. Through extensive experiments, our method achieves up to 5× acceleration over the base model while maintaining comparable visual quality across diverse benchmarks. Remarkably, the LoRA experts are trained with only 1 samples on a single V100 within one hour, yet the resulting models generalize strongly on unseen prompts.
Comparison Overview

Figure 2: Comparison of distill and accelerate strategies. Prior distillation pipelines rely on large training sets and costly retraining. Glance requires only one training sample to obtain Slow-LoRA and Fast-LoRA, providing plug-and-play acceleration of the base generation model.
Visual comparison of different Slow–Fast configurations.

BibTeX

@misc{dong2025glanceacceleratingdiffusionmodels,
                  title={Glance: Accelerating Diffusion Models with 1 Sample}, 
                  author={Zhuobai Dong and Rui Zhao and Songjie Wu and Junchao Yi and Linjie Li and Zhengyuan Yang and Lijuan Wang and Alex Jinpeng Wang},
                  year={2025},
                  eprint={2512.02899},
                  archivePrefix={arXiv},
                  primaryClass={cs.CV},
                  url={https://arxiv.org/abs/2512.02899}, 
            }
×