AI Paper Atlas

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

Golden Noise for Diffusion Models: A Learning Framework

Zikai Zhou, Shitong Shao, Lichen Bai, Shufei Zhang, Zhiqiang Xu, Bo Han, Zeke Xie

ICCV 2025 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 68 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 singular value prediction (ablation)NPNet (SDXL) · Pick-a-PicImageReward: 49.03 -> 65.01Improved
add residual prediction (ablation)NPNet (SDXL) · Pick-a-PicImageReward: 63.05 -> 65.01Improved
add data selection with human preference model (ablation)NPNet (SDXL) · Pick-a-PicImageReward: 62.91 -> 65.01Improved
replace standard noise with NPNet golden noiseSDXL · Pick-a-PicImageReward: 58.01 -> 65.01Improved
use NPNet instead of plain re-denoise sampling (Inversion)SDXL · DrawBenchHPSv2: 26.91 -> 27.14Improved
apply NPNet to DreamShaper-xl-v2-turboDreamShaper-xl-v2-turbo · HPD v2HPSv2: 30.96 -> 34.08Improved
apply plain re-denoise sampling (Inversion) without learningDreamShaper-xl-v2-turbo · DrawBenchHPSv2: 30.39 -> 30.10Hurt
fine-tune NPNet on 600 Hunyuan-DiT noise pairsHunyuan-DiT · Pick-a-PicImageReward: 91.33 -> 100.82Improved
add NPNet on top of DPODPO on SDXL · Pick-a-Pic (10 steps)ImageReward: 68.82 -> 75.06Improved
add NPNet on top of AYSAYS on SDXL · Pick-a-Pic (10 steps)ImageReward: 50.74 -> 59.50Improved
use NPNet instead of InitNO noise optimizationStableDiffusion-v1-4 · Pick-a-PicImageReward: -205.66 -> -95.89Improved
raise HPSv2 filtering threshold m from 0 to 0.01 (ablation)NPNet data selection (SDXL) · Pick-a-PicPickScore: 21.86 -> 21.72Hurt

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