AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Self-Supervised Flow Matching for Scalable Multi-Modal Synthesis

Hila Chefer, Patrick Esser, Dominik Lorenz, Dustin Podell, Vikash Raja, Vinh Tong, Antonio Torralba, Robin Rombach

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 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
replace DINOv2-B with stronger DINO encoders (DINOv2-L, DINOv3-B, DINOv3-H+) in REPAREPA · text-to-image FID vs training stepsFIDHurt
use naive full masking or per-token diffusion forcing noise instead of uniform noise (ablation)flow matching · noise scheduler comparisonFIDHurt
add Dual-Timestep Scheduling alone without self-supervised loss (ablation)vanilla flow matching · noise scheduler comparisonFIDImproved
add Self-Flow self-supervised objective with EMA teacherSiT-XL/2 on ImageNet · ImageNet 256x256 (4M steps)FID: 5.89 -> 5.70 (vs REPA)Improved
add Self-Flow on top of RAERAE with SiT-XL · ImageNet 256x256 (1M steps)FID: 3.24 -> 2.95Improved
add Self-Flow vs vanilla flow matchingtext-to-image model · Text-to-image (1M steps)FID: 4.08 -> 3.61Improved
add external alignment with video encoders (V-JEPA2, Depth Anything 3)video flow model · Text-to-video (600K steps)FVD: 50.95 -> 53.55 (V-JEPA2)Hurt
add external alignment with MERTaudio flow model · Text-to-audio (350K steps)FAD (CLAP): 148.874 -> 148.883No clear effect
scale model from 290M to 1B parametersSelf-Flow vs REPA · text-to-image scalingCLIP scoreImproved
remove representation loss Lrep (ablation)Self-Flow · ImageNet class-to-imageFID: over 4 points worse withoutImproved
add masking (dual-timestep) to representation alignment (ablation)Self-Flow · ImageNet class-to-imageFID: over 1 point worse withoutImproved
sample both timesteps from the full noise distribution instead of s in [t, t-0.2] (ablation)Self-Flow · ImageNet class-to-imageFIDImproved

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