AI Paper Atlas

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

Attention Sinks in Diffusion Transformers: A Causal Analysis

FANGZHENG WU, Brian Summa

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
remove dynamically identified top-1 attention sink (single layer 12) (ablation)SD3 inference · GenEval 553 promptsCLIP-T: -0.5e-3 [-1.7, +0.7]No clear effect
remove top-1 sinks in layers 6, 12 and 18 simultaneously (ablation)SD3 inference · GenEval 553 promptsCLIP-T: +0.9e-3 [-0.1, +2.0]No clear effect
increase masked sinks from top-1 to top-5 (ablation)SD3 inference, layer 12 · GenEval 553 promptsHPS-v2: -0.0020 (p = 0.001)Hurt
increase masked sinks from top-1 to top-5 (ablation)SD3 inference, layer 12 · GenEval 553 promptsCLIP-T: +0.0011No clear effect
sink masking versus equal-budget random masking at k=10 and k=50 (ablation)SD3 inference · HPS-v2 preferenceHPS-v2 difference-of-differences: -0.005 (k=10), -0.020 (k=50)Hurt
sink masking budget k in {1,5,10,20,50} (ablation)SD3 inference · CLIP-T alignmentCLIP-TNo clear effect
sink masking versus equal-budget random masking (ablation)SD3 inference, layer 12 · LPIPS perceptual shiftLPIPS: 0.053 -> 0.347 (k=1); 0.104 -> 0.436 (k=5)Hurt
suppress sinks in SDXL mid-block self- or cross-attention (ablation)SDXL inference · SDXL N=100CLIP-T: +0.0004 (attn1); -0.0003 (attn2)No clear effect
full removal sweep (eta from 1 to 0) on score path and value path (ablation)SD3 inference · dose-response sweepCLIP-TNo clear effect
intervene only in the early denoising phase (ablation)SD3 inference · phase-specific interventionsCLIP-T: +0.0006No clear effect
suppress sinks, measured with non-CLIP BLIP2-VQA (ablation)SD3 inference · GenEval N=553BLIP2-VQA score: +0.0001 [-0.0039, +0.0040]No clear effect
increase sink-suppression intensity from A1 to A3 (ablation)SD3 inference · GenEvalLPIPS vs baseline: 0.189 -> 0.314Hurt

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