AI Paper Atlas

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

ECCV 2026 · Image generation › Text-to-image diffusion · approach: T2I evaluation and controllable adapters · 6 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
add quality reward (ablation)POET general rewriter (FLUX.1-schnell) · Pick-a-Pic v2image quality win rate: 0.364 -> 0.494Improved
add general alignment reward (ablation)POET general rewriter · Pick-a-Pic v2image-text alignment win rate: 0.521 -> 0.561Improved
add physical alignment reward (ablation)POET general rewriter · Pick-a-Pic v2image-text alignment win rate: 0.504 -> 0.561Improved
add aesthetics reward (ablation)POET general rewriter · Pick-a-Pic v2image aesthetics win rate: 0.476 -> 0.818Improved
add aesthetics reward (ablation)POET general rewriter · Pick-a-Pic v2image-text alignment win rate: 0.561 -> 0.424Hurt
add POET rewriting to the original promptFLUX.1-dev · GenEvaloverall score: 0.70 -> 0.79Improved
add POET rewriting to the original promptJanusPro · GenEvaloverall score: 0.79 -> 0.68Hurt
add POET rewriting to the original promptJanusPro · T2I-CompBench++ Coloraccuracy: 0.5294 -> 0.7861Improved
scale rewriter LLM from Llama 3 3B to 70BPOET on FLUX.1-schnell · GenEvaloverall score: 0.73 -> 0.75Improved
replace Llama 3 70B rewriter with DeepSeek 70BPOET on FLUX.1-schnell · Pick-a-Pic v2average win rate: 0.510 -> 0.463Hurt
use full-model finetuning instead of LoRAPOET rewriter (Llama3-70B) · Pick-a-Pic v2win rate vs DALL-E 3No clear effect
train DPO beyond six roundsPOET rewriter · Pick-a-Pic v2image quality, aesthetics, average win rateMixed

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