AI Paper Atlas

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

Exposing Hidden Biases in Text-to-Image Models via Automated Prompt Search

Manos Plitsis, Giorgos Bouritsas, Vassilis Katsouros, Yannis Panagakis

ICML 2026 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement · 0 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
raise classifier weight lambda from 10 to 100 (ablation)BGPS · Male-biased gender prompts, SD 1.5mean male frequency: 0.76 -> 0.91Improved
raise classifier weight lambda from 10 to 100 (ablation)BGPS · Female-biased gender prompts, SD 1.5mean female frequency: 0.43 -> 0.66Improved
raise classifier weight lambda from 10 to 100 (ablation)BGPS · Male-biased prompts, SD 1.5prompt perplexity: 53 -> 129Hurt
add bias-classifier guidance (BGPS lambda=100) to LLM-only decodingLLM prompts · Female-biased gender prompts, SD 1.5mean female frequency: 0.27 -> 0.66Improved
replace gradient-based PEZ with BGPSprompt search · Male-biased prompts, SD 1.5prompt perplexity: 1387 -> 53Improved
add BGPS guidance to LLM on underrepresented groupLLM (biased) · Black-biased prompts, SD 1.5mean black frequency: 0.05 -> 0.27Improved
use BGPS prompts for further debias fine-tuningLoRA-debiased SD 1.5 · Male-biased prompts, lambda=100male proportion: 0.71 -> 0.47Improved
raise lambda from 10 to 100 on age attribute (ablation)BGPS · Old-biased prompts, SD 1.5old frequency: 0.18 -> 0.35Improved
add BGPS guidance (lambda=10) to LLM-onlyLLM prompts · Young/old age promptsyoung frequency: 0.81 -> 0.80No clear effect
increase lambda from 5 to 100 (ablation)BGPS · Flux male-biased promptsmale frequency: 0.68 -> 0.61Hurt
apply BGPS to debiased DiffLens modelBGPS (lambda=100) · Male-biased gender prompts, DiffLensmean male frequency: 0.31 -> 0.67Improved
halve K from 10 to 5 (ablation)BGPS · Runtimeminutes per prompt: ~2.4 -> ~1.2Improved

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