H22 - Gaussian Splats and Machine Learning | Jakob Ringler | Houdini 22 HIVE

Ever wanted to build smart, self-healing textures or generate 3D worlds instantly from flat images? Jakob Ringler dives into the cutting-edge Machine Learning additions shaking up Copernicus and Solaris in Houdini 22. He breaks down the massive performance upgrades coming to Neural Cellular Automata (NCA), showing how a new split architecture allows you to easily train, upscale, and procedurally deform tileable patterns right within COPs. The talk also highlights seamless mesh mapping using adjacency rasterization, intelligent alpha masking via Meta SAM 2, depth map extraction using Microsoft MOGI 2, and an extensive native toolkit to train, rasterize, and dynamically relight 3D Gaussian Splats inside Houdini. 00:00 - Intro & Neural Cellular Automata (NCA) Basics 05:30 - Solving Performance with Split Architecture 06:40 - How to Train Custom NCA Textures in COPs 10:43 - Driving Pattern Growth and Deformations 13:02 - Seamless Mesh Mapping via Adjacency Rasterization 14:18 - Latent Space Blending Between Patterns 16:47 - Next-Gen Alpha Masking with Meta SAM 2 18:56 - Extracting Utility Passes and PBR Textures with MOGI 2 20:17 - Native Gaussian Splatting Training & Relighting 28:38 - Audience Q&A: Animated Splats & Future Tech #Houdini22 #H22 #H22HIVE #SideFX #MachineLearning #Copernicus #GaussianSplatting #NCA #NeuralCellularAutomata #SAM2 #MOGI2 #TextureSynthesis #3DGS #VFXPipeline