Research map / Image generation / Text-to-image diffusion
POET: Preference Optimization for Enhanced Text-to-Image Generation
Ruibo Chen, Jiacheng Pan, Heng Huang, Zhenheng Yang
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 |
|---|---|---|---|
| add quality reward (ablation) | POET general rewriter (FLUX.1-schnell) · Pick-a-Pic v2 | image quality win rate: 0.364 -> 0.494 | Improved |
| add general alignment reward (ablation) | POET general rewriter · Pick-a-Pic v2 | image-text alignment win rate: 0.521 -> 0.561 | Improved |
| add physical alignment reward (ablation) | POET general rewriter · Pick-a-Pic v2 | image-text alignment win rate: 0.504 -> 0.561 | Improved |
| add aesthetics reward (ablation) | POET general rewriter · Pick-a-Pic v2 | image aesthetics win rate: 0.476 -> 0.818 | Improved |
| add aesthetics reward (ablation) | POET general rewriter · Pick-a-Pic v2 | image-text alignment win rate: 0.561 -> 0.424 | Hurt |
| add POET rewriting to the original prompt | FLUX.1-dev · GenEval | overall score: 0.70 -> 0.79 | Improved |
| add POET rewriting to the original prompt | JanusPro · GenEval | overall score: 0.79 -> 0.68 | Hurt |
| add POET rewriting to the original prompt | JanusPro · T2I-CompBench++ Color | accuracy: 0.5294 -> 0.7861 | Improved |
| scale rewriter LLM from Llama 3 3B to 70B | POET on FLUX.1-schnell · GenEval | overall score: 0.73 -> 0.75 | Improved |
| replace Llama 3 70B rewriter with DeepSeek 70B | POET on FLUX.1-schnell · Pick-a-Pic v2 | average win rate: 0.510 -> 0.463 | Hurt |
| use full-model finetuning instead of LoRA | POET rewriter (Llama3-70B) · Pick-a-Pic v2 | win rate vs DALL-E 3 | No clear effect |
| train DPO beyond six rounds | POET rewriter · Pick-a-Pic v2 | image quality, aesthetics, average win rate | Mixed |
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.
- RePrompt — RePrompt: Reasoning-Augmented Reprompting for Text-to-Image Generation via Reinforcement Learning (ICLR 2026) · 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
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.92)
- Preference Adaptive and Sequential Text-to-Image Generation (ICML 2025 · similarity 0.91)
- Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist Rewards (NeurIPS 2026 · similarity 0.90)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.90)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.90)
- Subject-driven Text-to-Image Generation via Preference-based Reinforcement Learning (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.
