Error as Signal: Stiffness-Aware Diffusion Sampling via Embedded Runge-Kutta Guidance

Researchers propose Embedded Runge-Kutta Guidance (ERK-Guid), a new method that improves diffusion model sampling by using solver-induced errors as guidance signals. The technique addresses stiffness issues in ODE trajectories and demonstrates superior performance over existing methods on ImageNet benchmarks.

• →ERK-Guid leverages solver-induced errors as guidance signals to improve diffusion model sampling quality.
• →The method specifically targets stiff regions where ODE trajectories change sharply and local truncation error degrades sample quality.
• →ERK-Guid outperforms state-of-the-art methods including Classifier-Free Guidance and Autoguidance on synthetic and ImageNet datasets.
• →The approach provides theoretical and empirical analysis of stiffness and eigenvector estimators with solver errors.
• →Code implementation is publicly available on GitHub for reproducibility and adoption.