Simplifying and Improving On-Policy Distillation for Few-Step Diffusion

arXiv:2610.18472 · cs.CV, cs.LG, eess.IV · Submitted 2026-10-13 · Read on arXiv

cs.CV, cs.LG, eess.IV

Submitted: 2026-10-13

Updated: 2026-10-13

Comments: Project Page: https://zjwfufu.github.io/SMD-Page/

Code: https://github.com/black-forest-labs/flux

Project page: https://zjwfufu.github.io/SMD-Page

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Distillation — 1× in this paper · explained in 2 episodes
• On-Policy Distillation — 1× in this paper · explained in 3 episodes

Sources

• Parallel Decoding Distillation for Fast Image and Video Generation
• Qwen-Image Technical Report
• Wan: Open and Advanced Large-Scale Video Generative Models
• Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation
• Continuous-Time Distribution Matching for Few-Step Diffusion Distillation
• ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment
• OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation
• ImgEdit: A Unified Image Editing Dataset and Benchmark
• Step1X-Edit: A Practical Framework for General Image Editing
• Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material
• AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation
• Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models
• Consistency Models Made Easy
• Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference
• LCM-LoRA: A Universal Stable-Diffusion Acceleration Module
• Multistep Consistency Models
• Dual-Expert Consistency Model for Efficient and High-Quality Video Generation
• SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation
• Target-Driven Distillation: Consistency Distillation with Target Timestep Selection and Decoupled Guidance
• Few-Step Distillation for Text-to-Image Generation: A Practical Guide

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