Parallel Decoding Distillation for Fast Image and Video Generation

2026-07-28

AI Summary

Existing diffusion and flow-matching models suffer from slow generation due to iterative sampling, and existing acceleration methods often encounter optimization difficulties, mode collapse, and insufficient motion diversity. This work proposes Parallel Decoding Distillation (PDD), a trajectory distillation framework that predicts multi-step denoising outcomes in a single forward pass, enabling flexible numbers of function evaluations (NFE) without requiring Jacobian-vector products or finite-difference approximations. By directly learning an average velocity representation and eschewing explicit derivative regression, PDD significantly enhances training stability and output diversity while preserving high sample quality. The method achieves state-of-the-art performance with only 4–8 NFE across diverse generative tasks—including text-to-image, video, and audio—demonstrating strong efficacy on models such as LTX-2.3, Wan 14B, and Qwen-Image.

Abstract

Generation in video diffusion or flow models is computationally expensive due to the slow and iterative sampling process. Current state-of-the-art (SOTA) acceleration methods heavily rely on variational score distillation (VSD) and adversarial losses to distill diffusion models into few-step generators. Albeit achieving high-quality video generation, these training losses are notoriously hard to optimize and suffer from mode collapse, leading to loss of video diversity and lack of motion. In this paper, we introduce Parallel Decoding Distillation (PDD), a simplified and scalable trajectory-based distillation method for fast inference of diffusion and flow matching models. Our architecture and training procedure are compatible with any pre-trained model and support sampling with a varying number of function evaluations (NFE). PDD accelerates generation by predicting multiple denoising steps per network evaluation. Conceptually, it learns a representation of the mean velocity without regressing its derivative using JVPs or finite-difference approximations. Our method achieves SOTA performance with 4-8 NFE on LTX-2.3 Text-to-Video/Audio, Wan 14B Text-to-Video, and Qwen-Image Text-to-Image. Moreover, PDD presents a significant improvement in generated video diversity.

Problem

• video generation
• diffusion models
• model acceleration
• mode collapse
• sampling efficiency

Innovation

• Parallel Decoding Distillation
• trajectory-based distillation
• few-step generation
• video diversity
• flow matching