Research map / Image generation / Text-to-image diffusion
Margin-Aware Preference Optimization for Aligning Diffusion Models Without Reference
Jiwoo Hong, Sayak Paul, Noah Lee, Kashif Rasul, James Thorne, Jongheon Jeong
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 Diffusion-DPO with reference-free MaPO | SDXL · Pick-a-Pic v2 test | Aesthetics: 6.03 -> 6.34 | Improved |
| replace Diffusion-DPO with MaPO | SDXL · Pick-a-Pic v2 test | HPS v2.1 / PickScore: 31.1 -> 31.2 HPS; 22.9 -> 22.9 PickScore | No clear effect |
| use MaPO instead of DreamBooth for personalization | SDXL · DreamBooth low-shot datasets | SigLIP (T-I): 61.60 -> 73.60 | Improved |
| use MaPO instead of DCO (which uses a reference model) | SDXL · DreamBooth low-shot datasets | DINOv2 (I-I): 89.12 -> 89.51 | Improved |
| remove reference model (MaPO vs Diffusion-DPO) on training cost | SDXL fine-tuning · Pick-a-Pic v2, 4 A100s | training time: 63.5 -> 54.3 (-14.5%) | Improved |
| remove reference model on peak GPU memory | SDXL fine-tuning · Pick-a-Pic v2, 4 A100s | peak GPU memory: 55.9 -> 46.1 (-17.5%) | Improved |
| use MaPO instead of Diffusion-DPO as reference mismatch grows (safe generation, style learning) | SDXL · Pick-Safety, Pick-Cartoon (VLM judge) | VLM judge score | Improved |
| use MaPO instead of Diffusion-DPO in low-mismatch cultural representation | SDXL · Pick-Culture (VLM judge) | VLM judge score | No clear effect |
| use SFT alone instead of preference optimization | SDXL · safe generation / cultural representation | VLM judge score | Hurt |
| increase beta (weight of MSE matching term) in personalization (ablation) | MaPO · personalization | qualitative fidelity: beta 128 to 1024 | Improved |
| lower DPO beta to reduce reference dependence | Diffusion-DPO · general | generation quality | 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.
- SDXL (base) — SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis (ICLR 2024) · 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
- Calibrated Multi-Preference Optimization for Aligning Diffusion Models (CVPR 2025 · similarity 0.93)
- Aligning Diffusion Models by Optimizing Human Utility (NeurIPS 2024 · similarity 0.92)
- DyMO: Training-Free Diffusion Model Alignment with Dynamic Multi-Objective Scheduling (CVPR 2025 · similarity 0.92)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.91)
- A Dense Reward View on Aligning Text-to-Image Diffusion with Preference (ICML 2024 · similarity 0.91)
- Improving Long-Text Alignment for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.91)
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.
