Research map / Image generation / Fast diffusion sampling
From Sketch to Fresco: Efficient Diffusion Transformer with Progressive Resolution
Shikang Zheng, Guantao Chen, Landis He, Jiacheng Liu, Yuqi Lin, Chang Zou, Linfeng Zhang
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 |
|---|---|---|---|
| replace unified noise field with independent re-noise (ablation) | Fresco · DrawBench 1024 | ImageReward: 0.9632 -> 1.0369 | Improved |
| use variance-based token selection instead of random selection (ablation) | Fresco · DrawBench 1024 | ImageReward: 0.9143 -> 1.0369 | Improved |
| use variance-based selection instead of attention score (ablation) | Fresco · DrawBench 1024 | ImageReward: 0.9849 -> 1.0369 | Improved |
| use variance-based selection instead of edge detection (ablation) | Fresco · DrawBench 1024 | ImageReward: 0.9482 -> 1.0369 | Improved |
| use Hadamard expansion instead of interpolated upsampling (ablation) | Fresco · DrawBench 1024 | ImageReward: 0.9876 -> 1.0369 | Improved |
| combine Fresco with TaylorSeer feature caching (ablation) | TaylorSeer · DrawBench 1024 | CLIP-IQA: 0.8872 -> 0.9116 | Improved |
| combine Fresco with step distillation (4 steps) (ablation) | FLUX.1-schnell · DrawBench 1024 | CLIP-IQA: 0.8771 -> 0.8693 | No clear effect |
| generate at higher resolution with Fresco | FLUX · FLUX 2048x2048 | ImageReward: 0.8499 -> 0.9348 | Improved |
| Fresco at 18 NFE on HunyuanVideo vs baseline | HunyuanVideo · VBench | Semantic Score: 71.04 -> 69.65 | Hurt |
Closest papers by meaning
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.91)
- Dynamic Diffusion Transformer (ICLR 2025 · similarity 0.90)
- Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decomposition (CVPR 2025 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration (ICML 2025 · similarity 0.89)
- Relay Diffusion: Unifying diffusion process across resolutions for image synthesis (ICLR 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.
