See the World in a Different Light

Physical Appearance Modeling and Relighting in the Age of Generative AI

June 4th, 2026 (14:00 - 17:30)
Room 108, Colorado Convention Center, CVPR 2026

Overview

Bridging the gap between the synthesized world and the real world to achieve physically accurate relighting, appearance editing, and capture.

Recent advances in generative AI, graphics, and vision have led to remarkable progress in relighting, appearance capture and editing, and inverse rendering—spanning scales from small objects to humans and entire scenes. The next major opportunity lies in bridging the gap between the synthesized world and the real world to achieve physically accurate relighting, appearance editing, and capture.

This workshop brings together researchers across these areas to highlight recent advances, discuss open challenges, and explore future directions. A cross-cutting theme will be the use of generative AI in physical appearance modeling.

New Datasets Release

We are thrilled to introduce datasets from the organizing team: FaceOLAT, HumanOLAT, and OLATVerse. These capture high-fidelity appearance across faces, humans, and objects under precisely controlled illuminations.

Invited Speakers

Shunsuke Saito
Shunsuke Saito

Meta
Digital Humans & Capture

Ira Kemelmacher-Shlizerman
Ira Kemelmacher-Shlizerman

Univ. of Washington
Generative Models

Dor Verbin
Dor Verbin

Google Deepmind
Generative Relighting

Zian Wang
Zian Wang

Nvidia & U of Toronto
Inverse Rendering

Hongzhi Wu
Hongzhi Wu

Zhejiang University
Material Acquisition

Shuang Zhao
Shuang Zhao

UIUC
Physics-based Rendering

Schedule

14:00
Opening Remarks & Dataset Introduction
Overview of FaceOLAT, HumanOLAT, OLATVerse and capture setups.
14:15
Invited Talk 1: Ira Kemelmacher-Shlizerman
Recent advances in data-driven and generative methods for Fashion AI.
14:45
Invited Talk 2: Hongzhi Wu
Differentiable acquisition as a unified framework that jointly optimizes physical capture and computational reconstruction for efficient, high-quality appearance and geometry modeling.
15:15
Invited Talk 3: Dor Verbin
Key bottlenecks in inverse rendering and emerging solutions that integrate generative priors with physically based optimization.
15:45
Coffee Break
16:00
Invited Talk 4: Zian Wang
Video diffusion models for estimating, synthesizing, and editing lighting and appearance in dynamic scenes, bridging inverse rendering, relighting, and controllable world simulation.
16:30
Invited Talk 5: Shuang Zhao
Current developments in physics-based differentiable rendering and inverse rendering.
17:00
Invited Talk 6: Shunsuke Saito
Strategies for moving beyond studio dome capture by leveraging in-the-wild data and pre-/post-training paradigms inspired by large language models.
17:30
Closing Remarks

Organizers

Xilong Zhou
Xilong Zhou
MPI-INF
Marc Habermann
Marc Habermann
MPI-INF
Jianchun Chen
Jianchun Chen
MPI-INF
Valentin Deschaintre
Valentin Deschaintre
Adobe
Zhao Dong
Zhao Dong
Meta Reality Labs
Pramod Rao
Pramod Rao
MPI-INF
Timo Teufel
Timo Teufel
MPI-INF
Christian Theobalt
Christian Theobalt
MPI-INF
Gordon Wetzstein
Gordon Wetzstein
Stanford
Jiajun Wu
Jiajun Wu
Stanford
Lan Xu
Lan Xu
ShanghaiTech
Yingyan Xu
Yingyan Xu
Google

Contact: xzhou@mpi-inf.mpg.de or mhaberma@mpi-inf.mpg.de