Geometry-Correct Diffusion Posterior Sampling with Denoiser-Pullback Curvature Guidance and Manifold-Aligned Damping
arXiv:2605.27990v1 Announce Type: new Abstract: Diffusion posterior sampling conditions diffusion priors on measurements, but data-consistency updates are typically scaled by hand-tuned guidance weights and can destabilize sampling under stiff, operator-dependent curvature. We replace scalar guidance with a per-noise-level damped Gauss--Newton correction computed in diffusion-state coordinates. The correction pulls likelihood gradients back through the denoiser, uses a one-sided curvature model that avoids forward denoiser Jacobians, and applies diffusion-calibrated rank-one damping aligned wi
The continuous evolution of AI models, particularly diffusion models, necessitates more robust and stable conditioning methods to improve output quality and reliability.
This research advances the core mechanics of diffusion models, reducing instability and improving sample quality, which is critical for future generative AI applications.
Diffusion posterior sampling becomes more stable and effective, reducing the need for manual tuning and improving the reliability of conditional image generation.
- · AI researchers
- · Generative AI developers
- · Creative industries using AI art
Improved quality and fidelity of AI-generated content based on specific conditions.
Faster development and deployment of more sophisticated generative AI models across various applications.
New creative and industrial applications become viable as AI generation becomes more precise and controllable.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG