Research map / Image generation / Text-to-image diffusion
Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models
Zhenyu Zhou, Defang Chen, Siwei Lyu, Chun Chen, Can Wang
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 |
|---|---|---|---|
| use TORS schedule instead of uniform outer schedule at 10 steps | Flux.1-Dev · DrawBench (Flux.1-Dev) | ImageReward: 0.71 -> 0.97 | Improved |
| use TORS schedule instead of uniform outer schedule at 10 steps | Flux.1-Dev · DrawBench (Flux.1-Dev) | HPSv2: 27.70 -> 29.30 | Improved |
| use TORS schedule instead of uniform outer schedule at 10 steps | SD 3.5 medium · DrawBench (SD3.5 medium) | ImageReward: 0.55 -> 0.86 | Improved |
| use TORS instead of GITS schedule | Flux.1-Dev · DrawBench (Flux.1-Dev) | ImageReward: 0.90 -> 0.97 | Improved |
| add fast solvers DPM-Solver or UniPC (ablation) | Flux.1-Dev 10 steps · DrawBench (Flux.1-Dev) | ImageReward: 0.71 -> 0.78 | Improved |
| add feature caching with FORA (ablation) | Flux.1-Dev 10 steps · DrawBench (Flux.1-Dev) | ImageReward: 0.71 -> 0.71 | No clear effect |
| add TaylorSeers feature caching (ablation) | SD 3.5 medium · DrawBench (SD3.5 medium) | ImageReward: 0.55 -> 0.54 | No clear effect |
| use TORS 10 steps instead of uniform 10 steps | Flux.1-Dev · GenEval (Flux.1-Dev) | overall: 0.57 -> 0.67 | Improved |
| apply Flux-derived TORS schedule to unseen Qwen-Image | Qwen-Image · DrawBench (unseen model, N=10) | HPSv2: 27.10 -> 29.28 | Improved |
| apply Flux-derived TORS schedule to Flux + Skin LoRA | Flux + Skin LoRA · DrawBench (unseen model, N=10) | HPSv2: 26.59 -> 28.24 | Improved |
| profile TORS statistics with 100 steps instead of 25 steps (ablation) | TORS · DrawBench (Flux.1-Dev, schedule profiled with 25 steps) | HPSv2: 28.23 -> 29.30 | Improved |
| vary the number of prompts used for profiling TORS from 25 to 100 (ablation) | TORS · DrawBench (Flux.1-Dev) | ImageReward: 0.96 -> 0.97 | No clear effect |
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.
- UniPC — UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models (NeurIPS 2023) · better on 3 of 3 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
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- Accelerating Parallel Sampling of Diffusion Models (ICML 2024 · similarity 0.90)
- SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation (CVPR 2024 · similarity 0.90)
- Morse: Dual-Sampling for Lossless Acceleration of Diffusion Models (ICML 2025 · similarity 0.90)
- SCott: Accelerating Diffusion Models with Stochastic Consistency Distillation (AAAI 2025 · similarity 0.90)
- Multistep Distillation of Diffusion Models via Moment Matching (NeurIPS 2024 · similarity 0.90)
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.
