Research map / Image generation / Fast diffusion sampling
PixelRush: Ultra-Fast, Training-Free High-Resolution Image Generation via One-step Diffusion
Hong-Phuc Lai, Phong Nguyen, Anh Tran
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 partial inversion (15 steps) to 50-step baseline (ablation) | PixelRush baseline (A) · 2K generation | FID: 54.70 -> 52.90 | Improved |
| add partial inversion, time (ablation) | baseline (A) · 2K generation | time (sec): 67 -> 18 | Improved |
| add few-step model (SDXL-Turbo, 1 step) (ablation) | PixelRush with partial inversion · 2K generation | FID: 52.90 -> 57.23 | Hurt |
| add Gaussian filter blending (ablation) | PixelRush few-step · 2K generation | FID: 57.23 -> 56.16 | Improved |
| add noise injection (ablation) | PixelRush few-step with Gaussian blend · 2K generation | FID: 56.16 -> 50.13 | Improved |
| add noise injection, IS (ablation) | PixelRush few-step with Gaussian blend · 2K generation | IS: 13.77 -> 14.32 | Improved |
| increase inversion timestep K from 249 to 499 (ablation) | PixelRush · 2K generation | FID: 50.13 -> 66.24 | Hurt |
| increase inversion timestep K from 249 to 999 (ablation) | PixelRush · 2K generation | FID: 50.13 -> 79.45 | Hurt |
| use SD-turbo instead of SDXL-turbo for refinement (ablation) | PixelRush · 2K generation | FID: 50.13 -> 52.75 | Hurt |
| use SANA as base instead of SDXL (ablation) | PixelRush · 2K generation | FID: 50.13 -> 57.48 | Hurt |
| use Pixart-delta instead of SDXL-turbo for refinement (ablation) | PixelRush · 2K generation | FID: 50.13 -> 50.31 | No clear effect |
| reduce patch overlap from 0.5 to 0.25 (ablation) | PixelRush · 4K generation | FID: 54.67 -> 54.28 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- FreeScale — FreeScale: Unleashing the Resolution of Diffusion Models via Tuning-Free Scale Fusion (ICCV 2025) · better on 6 of 6 shared results
- DemoFusion — DemoFusion: Democratising High-Resolution Image Generation With No $$$ (CVPR 2024) · better on 6 of 6 shared results
- FouriScale — FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis (ECCV 2024) · better on 6 of 6 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
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.90)
- Any-Size-Diffusion: Toward Efficient Text-Driven Synthesis for Any-Size HD Images (AAAI 2024 · similarity 0.89)
- Patched Denoising Diffusion Models For High-Resolution Image Synthesis (ICLR 2024 · similarity 0.89)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 2025 · similarity 0.89)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.89)
- Is One GPU Enough? Pushing Image Generation at Higher-Resolutions with Foundation Models. (NeurIPS 2024 · similarity 0.89)
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.
