New Procedural Generation Sample and AI Remeshing

Hi everyone 👋,

We have a great new learning resource and Nvidia has made progress with AI generated topology.

‍

When Procedural Content Generation (PCG) was announced in 2023, it introduced a new way to procedural create environments in Engine similar to other programs like Houdini. You can watch this video on the channel to learn why it is significant.

But the possibilities of the tool fell short initially, mostly being used to spawn plants and create biomes. While this is a great feature, it did not show off the true potential of the tool until now.

Epic Games has released the Cassini Sample, the first official use of PCG to generate buildings and complex artificial environments. It makes use of a lot of the new features introduced in UE5.5 and UE5.4 including grammar, GPU Instancing, recursion and more…

In the sample you can see how to generate a space station with PCG and spawn over 5 million meshes to create a gas giant’s rings. This opens up a whole new potential making the tool not just useful for natural environments, but also for man-made environments like buildings and cities.

You can watch this talk at Unreal Fest covering this project in depth to learn how to make use of it. I believe a goal of Epic Games is for PCG to be able to generate a city similar to how Houdini was used to create the Matrix City sample.

‍

If you have created 3D models before then you know retopology, sometimes called remeshing, is one of the most time consuming part of asset creation. It is the process of recreating a model’s geometry so it has less polygons than the original but still keeps the form of the model. This is essential for 3D objects that are scanned or sculpted, which often times have way more polygons than is necessary and a high poly count makes animation deformation difficult.

There has always been algorithms that automatically remesh an object, the most popular of which being ZBrush’s ZRemesher. But it can often times mess up or the user has to manually define polygon loops.

Nvidia’s research team has announced Meshtron, which uses AI generative algorithms to quickly and accurately retopologize an object. Unlike previous algorithms, Meshtron understands how to keep loop consistency and distinguish between key features of a mesh.

You can read the paper and see samples of it in action here. Hopefully the code is released so it can be implemented in our 3D programs.