What is it about?
Every cloud has a silver lining: When you look at a cloud in the sky, its bottom will look dark but the parts on the rim, where the sun can reach, are brightly lit. That's because the cloud casts shadows onto itself. When you want to make an image of a virtual 3D world with clouds or smoke in it, you have to compute how much each part is shadowed. This work presents a new way to do that accurately and efficiently.
Featured Image
Photo by Chris Linnett on Unsplash
Why is it important?
Many industries rely on rendering to make images of virtual scenes: Games, movies, TV, advertising, product design, virtual reality and many more. Nowadays, this is often done with GPUs. Previous ways to compute shadows from clouds were either inaccurate or took a lot of computation time on GPUs. Our new method is almost perfectly accurate, while being as fast as inaccurate alternatives. That makes it viable to use it in applications like games, where such shadow tests may need to be performed billions of times per second for different points.
Perspectives
After working on this problem for more than half a year, I had a breakthrough when I found a statistics book from 1993 in the library. It turns out that math theorems that date back to the 70s provide a compelling solution to this seemingly unrelated problem in computer graphics. There are many hidden treasures like that slumbering in libraries and I enjoy hunting for them.
Christoph Peters
Technische Universiteit Delft
Read the Original
This page is a summary of: Jackknife Transmittance and MIS Weight Estimation, ACM Transactions on Graphics, December 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3763273.
You can read the full text:
Resources
Project webpage
Provides additional materials, such as source code, images, slides and a presentation video.
Presentation video (10:49)
A recording of the presentation that I held at SIGGRAPH Asia 2025 in Hong Kong on December 16.
Source code
Complete open source code for a renderer that uses this new technique.
Contributors
The following have contributed to this page







