GPD: Guided Progressive Distillation for Fast and High-Quality Video Generation

2026-02-02

AI Summary

This work addresses the low sampling efficiency of video diffusion models, where existing acceleration methods often compromise generation quality. The authors propose a guided progressive distillation framework in which a teacher model progressively instructs a student model to denoise with larger step sizes. To balance computational efficiency and detail preservation while reducing optimization difficulty, the approach integrates online target generation and latent-space frequency-domain constraints. By unifying diffusion distillation, online target synthesis, and frequency-domain regularization, the method reduces the sampling steps of the Wan2.1 model from 48 to 6 while maintaining competitive visual quality on the VBench benchmark, significantly outperforming current distillation techniques.

Technology Category

Computer Vision: Diffusion Models for VisionMachine Learning: Deep Generative Models & AutoencodersSearch and Optimization: Sampling/Simulation-based Search

Application Category

Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphsEconomics, Online Markets and Human Computation: Cost models of using LLMs in production systemsWeb Mining and Content Analysis: Large pretrained models with web data

Abstract

Diffusion models have achieved remarkable success in video generation; however, the high computational cost of the denoising process remains a major bottleneck. Existing approaches have shown promise in reducing the number of diffusion steps, but they often suffer from significant quality degradation when applied to video generation. We propose Guided Progressive Distillation (GPD), a framework that accelerates the diffusion process for fast and high-quality video generation. GPD introduces a novel training strategy in which a teacher model progressively guides a student model to operate with larger step sizes. The framework consists of two key components: (1) an online-generated training target that reduces optimization difficulty while improving computational efficiency, and (2) frequency-domain constraints in the latent space that promote the preservation of fine-grained details and temporal dynamics. Applied to the Wan2.1 model, GPD reduces the number of sampling steps from 48 to 6 while maintaining competitive visual quality on VBench. Compared with existing distillation methods, GPD demonstrates clear advantages in both pipeline simplicity and quality preservation.

Problem

• video generation
• diffusion models
• computational cost
• quality degradation
• sampling steps

Innovation

• Guided Progressive Distillation
• video generation
• diffusion acceleration
• frequency-domain constraints
• knowledge distillation