AI BriefingNeural Render Proxies for Interactive and Differentiable Lighting
·2026.07.02 06:10
Key point
A proposal for a neural rendering proxy technology that reconstructs lighting in static scenes in real time and enables differentiable control.
Details
Lighting work in traditional CG animation pipelines takes a very long rendering time, incurring significant costs for iterative work. To address this, researchers at Disney Research and ETH Zurich introduced Neural Render Proxy (NRP).
NRP separates the traditional rendering process into Path Sampling and Emission Computation. After collecting light transport data with just a single rendering pass, a lightweight neural network trained on this data can simulate changes in lighting in real time under any lighting condition.
The key features are as follows:
- Real-time performance: Provides interactive frame rates of about 30–60 Hz for lighting reconstruction.
- Scalability: Regardless of scene complexity, computation cost is determined by resolution, number of light sources, and number of parameters.
- Differentiable: Supports a Gradient-based inverse workflow that can inversely compute lighting parameters from image-space edits or generative targets.
- Compatibility: Can also be used alongside existing non-differentiable production renderers.
