AI Paper Atlas

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

Sample Reward Soups: Query-efficient Multi-Reward Guidance for Text-to-Image Diffusion Models

Yinghua Yao, Yuangang Pan, Guoji Fu, Ivor Tsang

ICLR 2026 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement

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
use SRSoup K=20 instead of weighted-sum guidanceSRSoup · Aesthetic & Compressibilityreward queries per prompt: 15,000 -> 10,200Improved
use SRSoup K=20 instead of weighted-sum guidanceSRSoup · Aesthetic & Compressibilityinference time per prompt (min): 28.8 -> 21.1Improved
use SRSoup instead of weighted sum for a matched hypervolumeSRSoup · Aesthetic & Compressibility / Aesthetic & PickScorereward query efficiency: 1.8x fewer reward queriesImproved
use SRSoup instead of weighted sum for a matched hypervolumeSRSoup · Aesthetic & HPSv2reward query efficiency: about 2.7xImproved
add sample reward soup steps in the early denoising stage (K soup steps) (ablation)SRSoup · Aesthetic & Compressibilityreward queries: reduces reward queries substantially (by up to 40%)Improved
reduce K from 50 so weighted-sum search gradients enter earlier (ablation)SRSoup · Aesthetic & CompressibilityPareto frontImproved
increase queries per reward per step N (10, 20, 30) (ablation)SRSoup · Aesthetic & CompressibilityPareto frontImproved
use sample reward soup guidance instead of unconditional sampling for first K steps (ablation)SRSoup · Aesthetic & CompressibilityPareto frontImproved
replace fine-tuned model soups (DDPO, TDPO, AlignProp) with SRSoupSRSoup · animal and HPD prompts, SD1.5Pareto frontImproved
replace AlignProp (soup) with SRSoupSRSoup · Aesthetic & Compressibility, CLIP w1=0CLIP score: 0.1844 -> 0.2692Improved
replace TDPO (soup) with SRSoupSRSoup · Aesthetic & HPSv2, CLIP w1=1CLIP score: 0.1128 -> 0.2324Improved
replace FT-WS, RiC, D3PO (soup) and Flow-GRPO soup with SRSoupSRSoup · SD1.5 / SD3 multi-rewardPareto frontImproved

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