AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Trajectory Consistency for One-Step Generation on Euler Mean Flows

Zhiqi Li, Yuchen Sun, Duowen Chen, Jinjin He, Bo Zhu

ICML 2026 · Image generation › Fast diffusion sampling · approach: Guidance and alignment for flow matching · 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 JVP computation (JVP-free training)EMF vs MeanFlow · CelebA-HQ and ImageNet training costpeak memory and time per iteration: ImageNet memory 101.9GB -> 57.9GB; 400.9ms -> 198.8msImproved
add local-linearization surrogate to the JVP-free objective (ablation)EMF · CelebA-HQFID: 13.0 -> 10.9Improved
replace JVP-based objective with the JVP-free surrogate (ablation)EMF · CelebA-HQFID: 12.4 -> 10.9Improved
use fixed delta t between 6e-4 and 1e-2 (ablation)EMF · CelebA-HQFID: 10.9 at 6e-4 to 10.7 at 1e-2No clear effect
use very small or very large delta t (1e-4, 1e-1) (ablation)EMF · CelebA-HQFID: 10.9 -> 25.8 (1e-4); 10.9 -> 25.2 (1e-1)Hurt
add x1-prediction instead of u-prediction (ablation)EMF in pixel space · CelebA-HQ-256 pixel space (JiT-B/16), 1-stepFID-50k: 324.6 -> 36.3Improved
use x1-prediction with u-space loss (x1/u) rather than x1/x1 (ablation)EMF in pixel space · CelebA-HQ-256 pixel space, 1-stepFID-50k: 36.3 -> 30.6Improved
use EMF instead of MeanFlow objectiveDiT-B/2 on ImageNet-256 · ImageNet-256, 1-stepFID-50k: 11.1 -> 7.2Improved
use EMF instead of MeanFlow objectiveDiT-B/2 on CelebA-HQ · CelebA-HQ-256, 1-stepFID-50k: 12.4 -> 10.8Improved
use EMF instead of MeanFlow at 128 steps on CelebA-HQDiT-B/2 · CelebA-HQ-256, 128-stepFID-50k: 7.8 -> 7.6 (Flow Matching 7.3 is better)No clear effect
add auxiliary branch to predict u(t->t)EMF · CelebA-HQpeak memory / FID: memory 23.3GB -> 17.6GB; FID 10.9 -> 11.7Mixed
use EMF instead of MeanFlowSDF generation · ShapeNet SDF from 64 points, 1-stepChamfer distance: 0.060 -> 0.046Improved

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