DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation
cs.CV, cs.AI
Submitted: 2026-10-01
Updated: 2026-10-13
Comments: 28 pages, 15 figures. Project page: https://yzmblog.github.io/projects/DMAD
Project page: https://yzmblog.github.io/projects/DMAD
License: http://creativecommons.org/licenses/by/4.0/
The gist: Distribution Matching Distillation (DMD) trains a few-step student from the difference between separately estimated target and student scores, so it must keep an auxiliary diffusion model fitted to
Terminology
Abstract
Distribution Matching Distillation (DMD) trains a few-step student from the difference between separately estimated target and student scores, so it must keep an auxiliary diffusion model fitted to the student's evolving distribution at extra memory and computation cost. We introduce DMAD, Distribution Matching as Adversarial Distillation, which recasts distribution matching as classification and learns the required log-density ratios directly. Two discriminator heads on a shared backbone distinguish real data and teacher samples from the student's, and linear losses on their logits train the student without auxiliary score fitting. We prove that at the discriminator optimum these losses recover the distribution-matching gradient underlying DMD, through the classical identity linking discriminator logits to log-density ratios. We further introduce gap-based reweighting, which adapts teacher supervision across noise levels from the real-data head's empirical logit gap between real and teacher samples. DMAD reaches a Fréchet Inception Distance (FID) of 1.04 with one-step generation on ImageNet-64x64, 14.47 with four-step SDXL on COCO-10K, and a VBench total score of 85.15 with four-step Wan2.1-T2V-14B, the best values among the compared few-step methods and the multi-step teachers. On MiniMax-H3-33B, our four-step student achieves overall human preference rates of 79.1% over DMD2 and 84.6% over rCM for joint audio-video generation, excluding ties. Our code, models and demos are available at https://yzmblog.github.io/projects/DMAD.
Sources
- Optimizing Few-Step Generation with Adaptive Matching Distillation
- SenseFlow: Scaling Distribution Matching for Flow-based Text-to-Image Distillation
- Mean Flows for One-step Generative Modeling
- AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation
- Distribution Matching Distillation without Fake Score Network
- SDXL-Lightning: Progressive Adversarial Diffusion Distillation
- Diffusion Adversarial Post-Training for One-Step Video Generation
- Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generation
- Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed
- LCM-LoRA: A Universal Stable-Diffusion Acceleration Module
- SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis
- Adversarial Diffusion Distillation
- Parallel Decoding Distillation for Fast Image and Video Generation
- Score-Based Generative Modeling through Stochastic Differential Equations
- Wan: Open and Advanced Large-Scale Video Generative Models
- Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis
- SGMD: Score Gradient Matching Distillation for Few-Step Video Diffusion Distillation
- One-step Diffusion Models with $f$-Divergence Distribution Matching
- UltraVideo: High-Quality UHD Video Dataset with Comprehensive Captions
- VideoGen-Eval: Agent-based System for Video Generation Evaluation
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models