Houdini offers a machine learning–based character deformer that artists can train inside the software to produce more realistic skin and tissue movement than traditional bone-based animation methods. The tool, introduced with Houdini 20 and expanded in Houdini 20.5 and 21, lets artists train and apply the model entirely within Houdini's node network without writing custom code.

The ML Deformer works by generating random character poses, running a physics simulation on each pose to capture realistic skin deformation, then training a neural network to predict those deformations at playback speed. The trained model is stored in the ONNX format, which Houdini loads at runtime via its built-in ONNX inference engine. In Houdini 21, new nodes (ML Deform SOP and APEX Add ML Deformer SOP) allow the trained model to be applied in both standard geometry and character rig contexts, and SideFX ships a pre-trained deform model trained on its 'Otto' human character.