PCM : Picard Consistency Model for Fast Parallel Sampling of Diffusion Models

2025-03-25

AI Summary

Diffusion models suffer from low generation efficiency due to sequential denoising. This paper proposes the Picard Consistency Model (PCM), the first framework to embed consistency learning into the Picard iteration paradigm: PCM directly predicts the fixed-point solution at any stage along the convergence trajectory, enabling single-step approximation of the equilibrium. A model-switching mechanism is introduced to balance sampling acceleration and guaranteed exact convergence. Evaluated on image generation and robot control tasks, PCM achieves 2.71× and 1.77× speedups over sequential sampling and standard Picard iteration, respectively, without sacrificing output quality. The core contributions are: (1) a deep integration of consistency modeling with Picard iteration, unifying iterative refinement and fixed-point prediction; and (2) an efficient parallel sampling paradigm that eliminates multi-step iteration while preserving theoretical convergence guarantees.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Machine Learning: Probabilistic Circuits and Graphical Models

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Recently, diffusion models have achieved significant advances in vision, text, and robotics. However, they still face slow generation speeds due to sequential denoising processes. To address this, a parallel sampling method based on Picard iteration was introduced, effectively reducing sequential steps while ensuring exact convergence to the original output. Nonetheless, Picard iteration does not guarantee faster convergence, which can still result in slow generation in practice. In this work, we propose a new parallelization scheme, the Picard Consistency Model (PCM), which significantly reduces the number of generation steps in Picard iteration. Inspired by the consistency model, PCM is directly trained to predict the fixed-point solution, or the final output, at any stage of the convergence trajectory. Additionally, we introduce a new concept called model switching, which addresses PCM's limitations and ensures exact convergence. Extensive experiments demonstrate that PCM achieves up to a 2.71x speedup over sequential sampling and a 1.77x speedup over Picard iteration across various tasks, including image generation and robotic control.

Problem

• Speed up diffusion models' slow sequential generation
• Improve Picard iteration's convergence for faster sampling
• Ensure exact convergence with new parallelization and switching

Innovation

• Picard Consistency Model for parallel sampling
• Predicts fixed-point solution directly
• Model switching ensures exact convergence