AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

MeanFlow Transformers with Representation Autoencoders

Zheyuan Hu, Chieh-Hsin Lai, Ge Wu, Yuki Mitsufuji, Stefano Ermon

CVPR 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 7 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
run MeanFlow in RAE latent space with DiTDH instead of SD-VAE with DiT/SiT (ablation)MeanFlow, MFD, unguided · ImageNet 256 1-stepFID: 5.94 -> 2.03Improved
train with MeanFlow distillation from a flow matching teacher instead of one-point velocity MFT (ablation)MF-RAE on RAE latents · ImageNet 256 1-stepFID: 2.81 -> 2.03Improved
use MFD instead of MFT on SD-VAE latents with guidance (ablation)vanilla MF with SD-VAE · ImageNet 256 1-stepFID: 3.38 -> 3.15Improved
initialize MF from Consistency Mid-Training weights instead of random or flow matching init (ablation)MF on RAE latents · ImageNet 256/512 training stabilitytraining stabilityImproved
add MFT bootstrapping stage after MFD (ablation)MF-RAE · ImageNet 512 1-stepFID: 3.95 -> 3.23Improved
add MFT bootstrapping stage after MFD at 256 resolution (ablation)MF-RAE · ImageNet 256FIDNo clear effect
replace exact JVP with finite difference, dt between 1e-3 and 1e-2 (ablation)MF-RAE · ImageNet 256 1-stepFID: JVP 1.96 vs 2.03 at dt 5e-3No clear effect
use too small finite-difference step dt=1e-4 (ablation)MF-RAE · ImageNet 256 1-stepFID: 2.03 -> 5.63Hurt
remove classifier-free guidance by distilling an unguided class-conditional teacherMF-RAE · ImageNet 256hyperparameter complexityImproved
replace SD-VAE decoder with RAE decoderMeanFlow sampling · ImageNet 256 1-stepsampling GFLOPS: 424 -> 263Improved
use MF-RAE pipeline instead of training vanilla MF from scratchImageNet 256 training · ImageNet 256training cost (H100 GPU-days): over 600 -> about 100Improved
naive MF training on RAE latents without CMT (ablation)MF on RAE latents · S and B scales1-step FID before divergence: best FID still above 20Hurt

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