Research map / Image generation / Fast diffusion sampling
Adding Additional Control to One-Step Diffusion with Joint Distribution Matching
Yihong Luo, Tianyang Hu, Yifan Song, Jiacheng Sun, Zhenguo Li, Jing Tang
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 |
|---|---|---|---|
| train one-step student with joint distribution matching instead of DI + standard ControlNet | one-step SD 1.5 with control · controllable generation average | FID: 22.21 -> 14.58 | Improved |
| JDM one-step vs 50-step ControlNet | JDM · controllable generation average | Consistency: 0.093 -> 0.071 | Improved |
| JDM one-step vs 50-step ControlNet on depth and super-resolution | JDM · Depth and 8x SR | FID: 15.25 -> 16.24 (Depth); 11.93 -> 12.48 (SR) | Hurt |
| train ControlNet via denoising loss on one-step generator (ablation) | DI one-step generator · Canny | FID: 13.07 -> 28.83 | Hurt |
| implicitly train ControlNet with original VSD loss (ablation) | one-step student · Canny | Consistency: 0.102 -> 0.151 | Hurt |
| use two-phase warm-up for shared one-step UNet (ablation) | JDM · Canny | Consistency: 0.122 -> 0.102 | Improved |
| model log p(c|x_t) with a consistency model (ablation) | JDM · Depth | FID: 18.56 -> 16.24 | Improved |
| use decoupled CFG from SD-v2.1 teacher | JDM text-to-image · text-to-image SD-v1.5 | HPS average: 29.46 -> 29.75 (vs HFL) | Improved |
| add human feedback learning jointly with one-step training | JDM · text-to-image SD-v1.5 | HPS average: 26.35 -> 29.46 (vs Diff-Instruct) | Improved |
| use log p(c|x_0) directly for HFL (ablation) | JDM · human feedback learning | visual artifacts | Hurt |
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.
- Diff Instruct — Diff-Instruct: A Universal Approach for Transferring Knowledge From Pre-trained Diffusion Models (NeurIPS 2023) · better on 1 of 1 shared result
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
- One-step Diffusion with Distribution Matching Distillation (CVPR 2024 · similarity 0.90)
- Inductive Moment Matching (ICML 2025 · similarity 0.89)
- Multistep Distillation of Diffusion Models via Moment Matching (NeurIPS 2024 · similarity 0.89)
- One-Step Diffusion Distillation through Score Implicit Matching (NeurIPS 2024 · similarity 0.88)
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 2024 · similarity 0.88)
- Continuous-Time Distribution Matching for Few-Step Diffusion Distillation (NeurIPS 2026 · similarity 0.88)
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.
