Representation Distribution Matching for One-Step Visual Generation

Hugging Face Daily Papers Papers

Summary

This paper introduces Representation Distribution Matching (RDM), a method for one-step image generation by matching feature distributions under pretrained encoders, achieving state-of-the-art results on ImageNet and enabling post-training of FLUX.2 into a one-step generator with improved performance.

We elucidate the design space of Representation Distribution Matching (RDM), our name for the paradigm that trains a one-step image generator by matching generated and reference feature distributions under frozen pretrained encoders. We identify two design axes, how the distributions are compared and the representations they are compared in, and controlled studies along them yield three findings. First, the classical MMD, which could not train convincing generators a decade ago, becomes a strong and scalable objective once estimated right. Second, the generated batch is then the operative variable, with an optimum above 2048, far beyond customary batch sizes. Third, any single representation can be gamed, driven below the real score while images stay visibly fake, so we match against a balanced battery of encoders and evaluate with SW_r14, a Sliced-Wasserstein distance over 14 encoders that is independent of the training loss and resists gaming. Combining the preferred choices yields improved RDM (iRDM): it sets the one-step state of the art on ImageNet at SW_r14 1.30, corroborated by PickScore, a human-preference proxy our objective never optimizes, which prefers it over the prior best one-step generator on 71.2% of matched samples. The same recipe post-trains the four-step FLUX.2 [klein] into a one-step generator, surpassing the four-step version on GenEval, 0.826 to 0.794, and on PickScore, 22.76 to 22.58, in 90 H200 GPU-hours. Project page: https://alan-lanfeng.github.io/rdm/.
Original Article
View Cached Full Text

Cached at: 07/03/26, 03:52 AM

Paper page - Representation Distribution Matching for One-Step Visual Generation

Source: https://huggingface.co/papers/2607.02375

Abstract

Representation Distribution Matching enables high-quality image generation by matching feature distributions under pretrained encoders, with improved performance through optimized batch sizes and multi-encoder evaluation metrics.

We elucidate the design space ofRepresentation Distribution Matching(RDM), our name for the paradigm that trains a one-step image generator by matching generated and referencefeature distributionsunder frozenpretrained encoders. We identify two design axes, how the distributions are compared and the representations they are compared in, and controlled studies along them yield three findings. First, the classicalMMD, which could not train convincing generators a decade ago, becomes a strong and scalable objective once estimated right. Second, the generated batch is then the operative variable, with an optimum above 2048, far beyond customary batch sizes. Third, any single representation can be gamed, driven below the real score while images stay visibly fake, so we match against a balanced battery of encoders and evaluate withSW_r14, aSliced-Wasserstein distanceover 14 encoders that is independent of the training loss and resists gaming. Combining the preferred choices yields improved RDM (iRDM): it sets the one-step state of the art on ImageNet atSW_r141.30, corroborated byPickScore, a human-preference proxy our objective never optimizes, which prefers it over the prior best one-step generator on 71.2% of matched samples. The same recipe post-trains the four-stepFLUX.2[klein] into a one-step generator, surpassing the four-step version onGenEval, 0.826 to 0.794, and onPickScore, 22.76 to 22.58, in 90 H200 GPU-hours. Project page: https://alan-lanfeng.github.io/rdm/.

View arXiv pageView PDFProject pageGitHub8Add to collection

Get this paper in your agent:

hf papers read 2607\.02375

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.02375 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.02375 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.02375 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Reinforcing Few-step Generators via Reward-Tilted Distribution Matching

Hugging Face Daily Papers

RTDMD is a two-stage framework combining distribution matching distillation with reward-guided reinforcement learning to improve few-step image generation alignment with human preferences. It achieves state-of-the-art results on multiple models with only 4 inference steps.

Optimizing Visual Generative Models via Distribution-wise Rewards

Hugging Face Daily Papers

This paper presents a reinforcement learning framework for visual generative models that uses distribution-wise rewards, with a subset-replace strategy for efficiency, improving image diversity and quality while addressing mode collapse and reward hacking.

Representation Fréchet Loss for Visual Generation

Papers with Code Trending

This paper introduces FD-loss, a method to optimize Fréchet Distance as a training objective for visual generation by decoupling population and batch sizes. It demonstrates that this approach improves generator quality and suggests FID may not always accurately reflect visual quality.