AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Elastic Diffusion Transformer

Jiangshan Wang, Zeqiang Lai, Jiarui Chen, Jiayi Guo, Hang Guo, Xiu Li, Xiangyu Yue, Chunchao Guo

ICML 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration · 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
add adaptive block skipping (ablation)E-DiT (Qwen-Image) · Qwen-Imagelatency (ms): 2431 -> 1967 (skip only); GenEval 0.895Improved
add adaptive MLP width reduction on top of block skipping (ablation)E-DiT (Qwen-Image) · Qwen-Imagelatency (ms): 1967 -> 1643; GenEval 0.895 -> 0.857Improved
add block-wise caching (ablation)E-DiT (Qwen-Image) · Qwen-Imagelatency (ms): 1643 -> 1283; GenEval 0.857 -> 0.853Improved
use adaptive routing instead of non-adaptive random removal (ablation)E-DiT (Qwen-Image) · Qwen-ImageGenEval: 0.843 -> 0.853 at similar latency (1514 vs 1283 ms)Improved
directly skip borderline blocks instead of caching (ablation)E-DiT · Qwen-Imagegeneration qualityHurt
increase maximum reuse count K of cached features (ablation)E-DiT block-wise caching · Qwen-Imagegeneration qualityNo clear effect
increase cache threshold delta (ablation)E-DiT block-wise caching · Qwen-Imagegeneration quality: delta 0.03 gives T=1223ms vs 1319ms at 0.015Hurt
initialize routers with full capacity instead of random (ablation)E-DiT (Qwen-Image) · Qwen-ImageGenEval: 0.801 -> 0.853; DPG 78.6 -> 85.4Improved
apply E-DiT to Qwen-Image (turbo)Qwen-Image · Text-to-imagelatency (ms): 2431 -> 1283; DPG 88.9 -> 85.4Improved
apply E-DiT to Hunyuan3D-3.0Hunyuan3D-3.0 · Image-to-3Dlatency (ms): 5012 -> 2587; ULIP-I 0.1446 -> 0.1473Improved
apply E-DiT to FLUX.1-devFLUX.1-dev · Text-to-imagelatency (ms): 715 -> 374; GenEval 0.665 -> 0.671Improved

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