D2PO: Optimizing Diffusion Samplers via Dynamic Preference

📰 ArXiv cs.AI

Optimize diffusion samplers using D2PO, a framework that improves timestep schedules and classifier-free guidance weights

advanced Published 9 Jul 2026

Action Steps

Implement D2PO to optimize diffusion sampling policies
Configure timestep schedules using dynamic preference optimization
Apply classifier-free guidance weights to improve texture fidelity
Test the optimized sampler on a benchmark dataset
Compare the results with existing student-teacher regression frameworks

Who Needs to Know This

ML researchers and engineers can benefit from this framework to improve the efficiency and quality of their diffusion-based models

Key Insight

💡 D2PO can improve the efficiency and quality of diffusion-based models by optimizing timestep schedules and classifier-free guidance weights

Key Takeaways

Optimize diffusion samplers using D2PO, a framework that improves timestep schedules and classifier-free guidance weights

Full Article

Title: D2PO: Optimizing Diffusion Samplers via Dynamic Preference

Abstract:
arXiv:2607.06609v1 Announce Type: cross Abstract: We propose D2PO (Dynamic Direct Preference Optimization), a principled framework for optimizing diffusion sampling policies with respect to timestep schedules and classifier-free guidance (CFG) weights. Our work is motivated by a fundamental limitation of existing student-teacher regression frameworks; low-NFE student samplers are trained to mimic high-NFEteachers, often sacrificing high-frequency texture fidelity while preserving coarse global s

Read full paper → ← Back to Reads