LumaGuide: Distribution Shaping for Training-Free HDR Generation in Diffusion Models

2026-07-28

AI Summary

This work addresses the limitation of pre-trained diffusion models in generating high dynamic range (HDR) images due to statistical biases in their training data. The authors propose a fine-tuning-free distribution shaping framework that directly modulates the output luminance distribution during sampling via differentiable energy-based guidance to achieve HDR synthesis. Their approach uniquely aligns the target luminance histogram in the perceptually uniform PQ color space, enabling flexible specification of the desired distribution through predefined profiles, reference images, or text prompts. Furthermore, it naturally extends to temporally coherent HDR video generation. Experiments demonstrate that the method produces HDR content with coherent highlights and preserved shadow details while maintaining semantic fidelity, showcasing its versatility and effectiveness across both image and video domains.

Abstract

Pretrained diffusion models generate realistic images but are constrained by the statistical biases of their training data, limiting their ability to produce high dynamic range (HDR) content. In this work, we introduce LumaGuide, a training-free framework for distribution shaping in diffusion models. Instead of modifying model parameters, LumaGuide steers the sampling process to match target feature distributions via differentiable energy-based guidance. We instantiate this framework for HDR generation by controlling luminance distributions in perceptually uniform PQ space. Our results show that aligning luminance histograms is sufficient to induce HDR-consistent behavior, including coherent highlights and preserved shadow detail, while maintaining semantic fidelity. Beyond HDR, LumaGuide enables flexible specification of target distributions through data-driven presets, reference images, or text-driven predictors, and extends naturally to video generation with temporal consistency constraints. More broadly, our work demonstrates that controllable generation can be achieved by directly shaping output distributions at sampling time, without retraining diffusion models.

Problem

• HDR generation
• diffusion models
• distribution shaping
• training-free
• luminance distribution

Innovation

• distribution shaping
• training-free
• HDR generation
• diffusion models
• energy-based guidance