Search papers, labs, and topics across Lattice.
This paper introduces a unified stochastic differential equation (SDE) formulation that bridges Monte Carlo rendering and diffusion-based generative modeling, enabling physically grounded control over diffusion-generated images. The method models a general SDE for Monte Carlo integration based on the Central Limit Theorem and instantiates it via physically based path tracing to create a physically grounded SDE representation. By analyzing the physical characteristics of path tracing in relation to noise variance, the approach extends these characteristics to existing diffusion models.
Control diffusion models with physically-based rendering parameters by reformulating Monte Carlo integration as a stochastic differential equation.
Diffusion-based image generators excel at producing realistic content from text or image conditions, but they offer only limited explicit control over low-level, physically grounded shading and material properties. In contrast, physically based rendering (PBR) offers fine-grained physical control but lacks prompt-driven flexibility. Although these two paradigms originate from distinct communities, both share a common evolution -- from noisy observations to clean images. In this paper, we propose a unified stochastic formulation that bridges Monte Carlo rendering and diffusion-based generative modeling. First, a general stochastic differential equation (SDE) formulation for Monte Carlo integration under the Central Limit Theorem is modeled. Through instantiation via physically based path tracing, we convert it into a physically grounded SDE representation. Moreover, we provide a systematic analysis of how the physical characteristics of path tracing can be extended to existing diffusion models from the perspective of noise variance. Extensive experiments across multiple tasks show that our method can exert physically grounded control over diffusion-generated results, covering tasks such as rendering and material editing.
