Guided Path Sampling: Steering Diffusion Models Back on Track with Principled Path Guidance

2025-12-28

AI Summary

Standard classifier-free guidance (CFG) suffers from systematic extrapolation during iterative denoising and inversion-based refinement, causing sampling trajectories to deviate from the data manifold and leading to error divergence—severely degrading both generation quality and control fidelity. To address this, we propose Manifold-Constrained Guidance Path Sampling (GPS), a novel paradigm that theoretically characterizes how CFG-induced extrapolation destabilizes sampling paths and proves GPS ensures strictly bounded cumulative error—contrasting with CFG’s unbounded error amplification. We further design a semantic-aware dynamic guidance strength scheduler to achieve spatiotemporal alignment between generation dynamics and semantic injection. GPS is architecture-agnostic and integrates seamlessly with主流 diffusion models including SDXL and Hunyuan-DiT. Experiments demonstrate state-of-the-art performance: ImageReward of 0.79 and HPS v2 of 0.2995 on SDXL, and GenEval semantic alignment accuracy improved to 57.45%, significantly outperforming existing methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Iterative refinement methods based on a denoising-inversion cycle are powerful tools for enhancing the quality and control of diffusion models. However, their effectiveness is critically limited when combined with standard Classifier-Free Guidance (CFG). We identify a fundamental limitation: CFG's extrapolative nature systematically pushes the sampling path off the data manifold, causing the approximation error to diverge and undermining the refinement process. To address this, we propose Guided Path Sampling (GPS), a new paradigm for iterative refinement. GPS replaces unstable extrapolation with a principled, manifold-constrained interpolation, ensuring the sampling path remains on the data manifold. We theoretically prove that this correction transforms the error series from unbounded amplification to strictly bounded, guaranteeing stability. Furthermore, we devise an optimal scheduling strategy that dynamically adjusts guidance strength, aligning semantic injection with the model's natural coarse-to-fine generation process. Extensive experiments on modern backbones like SDXL and Hunyuan-DiT show that GPS outperforms existing methods in both perceptual quality and complex prompt adherence. For instance, GPS achieves a superior ImageReward of 0.79 and HPS v2 of 0.2995 on SDXL, while improving overall semantic alignment accuracy on GenEval to 57.45%. Our work establishes that path stability is a prerequisite for effective iterative refinement, and GPS provides a robust framework to achieve it.

Problem

• Addresses instability in diffusion model refinement with Classifier-Free Guidance
• Proposes Guided Path Sampling to keep sampling on the data manifold
• Enhances image quality and semantic alignment through stable iterative refinement

Innovation

• Manifold-constrained interpolation replaces unstable extrapolation
• Optimal scheduling aligns guidance with coarse-to-fine generation
• Path stability ensures bounded error and robust refinement