FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion
This solves the problem of flexible test-time resolution generation for users of diffusion models, offering an incremental improvement over prior methods that suffered from artifacts and high latency.
The paper tackles the problem of generating high-resolution images with pre-trained diffusion models without retraining, which often causes repetitive patterns and distortions. The proposed FAM Diffusion method achieves state-of-the-art performance by addressing structural and local artifacts with negligible latency overhead.
Diffusion models are proficient at generating high-quality images. They are however effective only when operating at the resolution used during training. Inference at a scaled resolution leads to repetitive patterns and structural distortions. Retraining at higher resolutions quickly becomes prohibitive. Thus, methods enabling pre-existing diffusion models to operate at flexible test-time resolutions are highly desirable. Previous works suffer from frequent artifacts and often introduce large latency overheads. We propose two simple modules that combine to solve these issues. We introduce a Frequency Modulation (FM) module that leverages the Fourier domain to improve the global structure consistency, and an Attention Modulation (AM) module which improves the consistency of local texture patterns, a problem largely ignored in prior works. Our method, coined Fam diffusion, can seamlessly integrate into any latent diffusion model and requires no additional training. Extensive qualitative results highlight the effectiveness of our method in addressing structural and local artifacts, while quantitative results show state-of-the-art performance. Also, our method avoids redundant inference tricks for improved consistency such as patch-based or progressive generation, leading to negligible latency overheads.