Research map / Image generation / Text-to-image diffusion
FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis
Linjiang Huang, Rongyao Fang, Aiping Zhang, Guanglu Song, Si Liu, Yu Liu, Hongsheng Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add FouriScale guidance (milder-filter conditional estimate with attention maps from FouriScale estimate) (ablation) | FouriScale · LAION-5B, 2048x2048 (SD 2.1) | FID_r: -4.26 FID_r | Improved |
| add low-pass filtering to dilated convolution (ablation) | FouriScale without guidance · LAION-5B, 2048x2048 (SD 2.1) | FID_r: 46.74 without filtering | Improved |
| replace ScaleCrafter kernel enlargement with dilation plus low-pass filtering | SDXL 1.0 high-resolution generation · LAION-5B, 16x 1:1 | FID_r: 84.58 -> 56.66 | Improved |
| apply FouriScale instead of vanilla model at 4x resolution | SD 1.5 · LAION-5B, 4x 1:1 | FID_r: 26.96 -> 23.62 | Improved |
| use attention-entropy scaling (Attn-Entro) instead of vanilla | SD 1.5 · LAION-5B, 16x 1:1 | FID_r: 67.90 -> 67.45 | No clear effect |
| use a gentler low-pass filter (sigma 0.6) instead of the ideal filter for SDXL | FouriScale on SDXL · SDXL | visual quality | Improved |
| use FouriScale one-stage generation instead of SD followed by diffusion super-resolution cascade | SD 2.1 at 2048x2048 · LAION-5B, 2048x2048 | FID_r / KID_r | Hurt |
| use FouriScale instead of ScaleCrafter | SDXL 16x · SDXL 16x 1:1 | seconds per image: 577 -> 540 | Improved |
Later papers that compared with this method
4 later accepted papers report FouriScale as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- PixelRush: Ultra-Fast, Training-Free High-Resolution Image Generation via One-step Diffusion (CVPR 2026 · better on 6 of 6 shared results)
- Exploring Position Encoding Mechanism in Diffusion U-Net for Training-free High-resolution Image Generation (AAAI 2026 · better on 6 of 6 shared results)
- UltraFlux: Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios (CVPR 2026 · better on 2 of 2 shared results)
- FreeScale: Unleashing the Resolution of Diffusion Models via Tuning-Free Scale Fusion (ICCV 2025 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ScaleCrafter — ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024) · better on 11 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.89)
- Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR 2025 · similarity 0.89)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.88)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.88)
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.88)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
