AI Paper Atlas

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

AAAI 2026 · Image generation › Text-to-image diffusion · approach: Preference optimization for diffusion · 19 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace Diffusion-DPO with reference-free MaPOSDXL · Pick-a-Pic v2 testAesthetics: 6.03 -> 6.34Improved
replace Diffusion-DPO with MaPOSDXL · Pick-a-Pic v2 testHPS v2.1 / PickScore: 31.1 -> 31.2 HPS; 22.9 -> 22.9 PickScoreNo clear effect
use MaPO instead of DreamBooth for personalizationSDXL · DreamBooth low-shot datasetsSigLIP (T-I): 61.60 -> 73.60Improved
use MaPO instead of DCO (which uses a reference model)SDXL · DreamBooth low-shot datasetsDINOv2 (I-I): 89.12 -> 89.51Improved
remove reference model (MaPO vs Diffusion-DPO) on training costSDXL fine-tuning · Pick-a-Pic v2, 4 A100straining time: 63.5 -> 54.3 (-14.5%)Improved
remove reference model on peak GPU memorySDXL fine-tuning · Pick-a-Pic v2, 4 A100speak 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 scoreImproved
use MaPO instead of Diffusion-DPO in low-mismatch cultural representationSDXL · Pick-Culture (VLM judge)VLM judge scoreNo clear effect
use SFT alone instead of preference optimizationSDXL · safe generation / cultural representationVLM judge scoreHurt
increase beta (weight of MSE matching term) in personalization (ablation)MaPO · personalizationqualitative fidelity: beta 128 to 1024Improved
lower DPO beta to reduce reference dependenceDiffusion-DPO · generalgeneration qualityHurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy