← All resources

Generative AI model for Global Illumination effects

Auto-discovered

Blog Advanced

SungYe Kim, Wojciech Uss, Wojciech Kalinski, Alexandr Kuznetsov, Harish Anand

Visit resource ↗

Description

Generative AI model for Global Illumination effects is a July 2025 AMD GPUOpen Learn article by SungYe Kim, Wojciech Uss, Wojciech Kalinski, Alexandr Kuznetsov, and Harish Anand on conditional diffusion for specific indirect lighting such as reflective caustics.

It motivates gaps in diffusion generators for multi-bounce high-frequency effects, then describes a latent diffusion model conditioned on albedo, normals, roughness, metallic maps, direct illumination cues, and light pose via positional encoding and spherical harmonics. Results show cardioid-shaped caustics composited with direct lighting against reference GI, with detail deferred to their Eurographics 2025 short paper.

It is for advanced rendering researchers and engine engineers exploring neural GI helpers who already understand path-traced references and G-buffer inputs.

The page is a free HTML overview with framework and result figures; the formal short paper is cited separately.

It is useful as an industry summary of conditioning diffusion on render buffers for a narrow GI effect before reading the full short paper.