Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

Hugging Face Daily Papers Papers
diffusion-models distillation few-step-sampling continuous-time image-generation computer-vision

Summary

This paper introduces Continuous-Time Distribution Matching (CDM), a method for few-step diffusion distillation that migrates from discrete to continuous optimization to improve visual fidelity and preserve fine details.

Step distillation has become a leading technique for accelerating diffusion models, among which Distribution Matching Distillation (DMD) and Consistency Distillation are two representative paradigms. While consistency methods enforce self-consistency along the full PF-ODE trajectory to steer it toward the clean data manifold, vanilla DMD relies on sparse supervision at a few predefined discrete timesteps. This restricted discrete-time formulation and mode-seeking nature of the reverse KL divergence tends to exhibit visual artifacts and over-smoothed outputs, often necessitating complex auxiliary modules -- such as GANs or reward models -- to restore visual fidelity. In this work, we introduce Continuous-Time Distribution Matching (CDM), migrating the DMD framework from discrete anchoring to continuous optimization for the first time. CDM achieves this through two continuous-time designs. First, we replace the fixed discrete schedule with a dynamic continuous schedule of random length, so that distribution matching is enforced at arbitrary points along sampling trajectories rather than only at a few fixed anchors. Second, we propose a continuous-time alignment objective that performs active off-trajectory matching on latents extrapolated via the student's velocity field, improving generalization and preserving fine visual details. Extensive experiments on different architectures, including SD3-Medium and Longcat-Image, demonstrate that CDM provides highly competitive visual fidelity for few-step image generation without relying on complex auxiliary objectives. Code is available at https://github.com/byliutao/cdm.
Original Article
View Cached Full Text

Cached at: 05/08/26, 07:03 AM

Paper page - Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

Source: https://huggingface.co/papers/2605.06376 Published on May 7

·

Submitted byhttps://huggingface.co/byliutao

liutaoon May 8

Authors:

,

,

,

,

,

,

,

,

,

Abstract

Continuous-Time Distribution Matching migrates diffusion model distillation from discrete to continuous optimization, enabling arbitrary points along sampling trajectories and preserving fine visual details through dynamic scheduling and velocity field extrapolation.

Stepdistillationhas become a leading technique for acceleratingdiffusion models, among whichDistribution MatchingDistillation(DMD) andConsistency Distillationare two representative paradigms. While consistency methods enforce self-consistency along the full PF-ODE trajectory to steer it toward the clean data manifold, vanilla DMD relies on sparse supervision at a few predefined discrete timesteps. This restricted discrete-time formulation and mode-seeking nature of thereverse KL divergencetends to exhibit visual artifacts and over-smoothed outputs, often necessitating complex auxiliary modules -- such as GANs or reward models -- to restore visual fidelity. In this work, we introduceContinuous-TimeDistribution Matching(CDM), migrating the DMD framework from discrete anchoring to continuous optimization for the first time. CDM achieves this through twocontinuous-timedesigns. First, we replace the fixed discrete schedule with a dynamic continuous schedule of random length, so thatdistribution matchingis enforced at arbitrary points along sampling trajectories rather than only at a few fixed anchors. Second, we propose acontinuous-timealignment objective that performs active off-trajectory matching on latents extrapolated via the student’svelocity field, improving generalization and preserving fine visual details. Extensive experiments on different architectures, including SD3-Medium and Longcat-Image, demonstrate that CDM provides highly competitive visual fidelity for few-stepimage generationwithout relying on complex auxiliary objectives. Code is available at https://github.com/byliutao/cdm.

View arXiv pageView PDFProject pageGitHub1Add to collection

Get this paper in your agent:

hf papers read 2605\.06376

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

Models citing this paper2

#### byliutao/stable-diffusion-3-medium-turbo Text-to-Image• Updated27 minutes ago • 4 • 1 #### byliutao/Longcat-Image-Turbo Updated26 minutes ago • 1

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.06376 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.06376 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

Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation

Hugging Face Daily Papers

This paper introduces Trajectory-Shaped Discrete Flow Matching (TS-DFM), which replaces blind stochastic jumps with guided navigation to significantly improve text generation efficiency and reduce computational costs. The method achieves superior perplexity and speed compared to traditional multi-step baselines while maintaining unchanged inference costs.

AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation

Hugging Face Daily Papers

AnyFlow introduces a novel any-step video diffusion distillation framework that optimizes full ODE sampling trajectories through flow-map transition learning and backward simulation, achieving performance that matches or surpasses consistency-based counterparts while scaling with sampling step budgets.

Temporal Difference Learning for Diffusion Models

arXiv cs.LG

This paper introduces a temporal difference (TD) learning objective for diffusion models that enforces cross-time consistency along the denoising trajectory. It reformulates denoising as a reinforcement learning policy evaluation problem, showing significant improvements in sample quality (FID), especially for few-step samplers.