Research map / Image generation / Text-to-image diffusion
UltraFlux: Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios
Tian Ye, Song Fei, Lei Zhu
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 SNR-Aware Huber Wavelet loss (ablation) | Flux + F16 VAE · 4K ablation | HPSv3: 9.22 -> 9.70 | Improved |
| add SNR-Aware Huber Wavelet loss (ablation) | Flux + F16 VAE · 4K ablation | FID: 151.40 -> 148.81 | Improved |
| add SACL on top of SNR-HW (ablation) | UltraFlux · 4K ablation | HPSv3: 9.70 -> 10.30 | Improved |
| add SACL on top of SNR-HW (ablation) | UltraFlux · 4K ablation | ArtiMuse: 67.23 -> 67.31 | No clear effect |
| add Resonance 2D RoPE with YaRN (ablation) | UltraFlux · 4K ablation | HPSv3: 10.30 -> 10.91 | Improved |
| add Resonance 2D RoPE with YaRN (ablation) | UltraFlux · 4K ablation | FID: 147.32 -> 146.93 | Improved |
| swap Diffusion-4K-v2 data for MultiAspect-4K-1M with plain Flux (ablation) | Flux, latent L2 · 4K data-model ablation | HPSv3: 8.57 -> 9.17 | Improved |
| swap Diffusion-4K-v2 data for MultiAspect-4K-1M with UltraFlux (ablation) | UltraFlux · 4K data-model ablation | HPSv3: 10.03 -> 10.78 | Improved |
| use UltraFlux model instead of Flux latent L2 on MultiAspect-4K-1M (ablation) | MultiAspect-4K-1M · 4K data-model ablation | FID: 151.41 -> 145.81 | Improved |
| post-train VAE (UltraFlux-F16-VAE) vs Flux-VAE-F16 | Flux F16 VAE · Aesthetic-4K@4096 | rFID: 2.201 -> 0.547 | Improved |
| add VAE post-training vs Flux-VAE-F16-SC | Flux-VAE-F16-SC · Aesthetic-4K@4096 | LPIPS: 0.097 -> 0.102 | Hurt |
| add prompt refiner to UltraFlux vs Seedream 4.0 | UltraFlux · 4K vs Seedream 4.0 | HPSv3: 12.03 vs 11.98 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ScaleCrafter — ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024) · better on 4 of 4 shared results
- FouriScale — FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis (ECCV 2024) · better on 2 of 2 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
- Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR 2025 · similarity 0.90)
- UltraHR-100K: Enhancing UHR Image Synthesis with A Large-Scale High-Quality Dataset (NeurIPS 2025 · similarity 0.89)
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.87)
- UltraDiff:Transferring High-Fidelity Priors to Compressed Latent Spaces for High-resolution Image Generation (NeurIPS 2026 · similarity 0.87)
- PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis (ICLR 2024 · similarity 0.86)
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.86)
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.
