AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Dual-Seed Evolutionary Algorithm for Noise Optimization in Diffusion Models

Yuzheng Tan, Yuan He, Yao Zhu, Tianlin Huo, Huanqian Yan, Hang Su, Shuxin Zhang, Guangneng Hu

AAAI 2026 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement · 1 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 dual-seed interpolation and guided inversion on top of best-of-N seed selectionRewardSelect · Drawbench (SDv2.1)Average Gain: 3.8969 vs 0.8010 higher than RewardSelectImproved
add dual-seed interpolation and guided inversion on top of best-of-N seed selectionRewardSelect · Drawbench (SDXL)Average Gain: 1.3876, exceeding RewardSelect by 0.5173Improved
apply SOE instead of Golden noiseGolden noise · Drawbench (SDXL)Average Gain: exceeding Golden noise by 0.5116Improved
apply SOE instead of Z-samplingZ-sampling · PickaPic (SDv2.1)Average Gain: surpassing RewardSelect by 0.4583 and Z-sampling by 2.5879Improved
apply SOE to vanilla SDv2.1SDv2.1 · HPDv2ImageReward: 0.3969 -> 1.3405Improved
apply SOE to vanilla SDXLSDXL · HPDv2ImageReward: 0.9760 -> 1.5637Improved
apply SOE to vanilla SDXLSDXL · DrawbenchClipScore: 0.3291 -> 0.3539Improved
increase number of seeds per iteration to 3 (ablation)SOE on SDv2.1 · ImageReward vs vanilla SDXLImageReward: requires only 3 seeds to exceed the average performance of vanilla SDXLImproved
increase number of seeds per iteration to 12 (ablation)SOE on SDv2.1 · ClipScore vs vanilla SDXLClipScore: at least 12 seeds are necessary to outperform SDXL on ClipScoreImproved
increase number of optimization iterations (ablation)SOE · ImageRewardImageReward: increasing the number of rounds improves ImageReward scoresImproved
increase number of optimization iterations (ablation)SOE · ClipScoreClipScore: yields limited gains on ClipScoreNo clear effect
apply Initno attention-based noise optimizationSDv2.1 · HPDv2Average Gain: -0.1218No clear effect

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