Houdini 22.0 Nodes Geometry nodes

Agent Add ML Deformer geometry node

Adds an agent shape which is deformed using an ML model.

On this page
Since 22.0

This node adds shapes to an agent and configures them to use a trained ML model which deforms the skin of a character. Internally, the ML model predicts weights to combine the residual blend shapes to apply a correction to the rest skin geometry.

Note

Agent Add ML Deformer assumes the model was trained on a data set that was created with the help of ML Pose Serialize using its default settings. These settings are Mode set to Subskeleton and Include World Rotation turned on. Also, Transform Group needs to match the joint group used with ML Pose Serialize to create the training data set. These requirements would be met if the ML Train Deformer recipe was used for training.

See Machine Learning documentation for more general information.

Parameters

Group

The agents to be modified.

Layer Name

Creates a new agent layer with the specified name. The layer contains shape bindings to the new shapes.

Channel Name Prefix

Prefix for the blend shape channel names used by the deformer. Any missing blend shape channels will be added to the agent’s rig.

If an agent will be assigned multiple shapes which use the ML deformer, the channel names used by each shape must be unique. Channel names can be reused across shape variations where only one variant will be assigned to each agent instance.

Model File

The trained model in ONNX format that inputs a serialized pose and outputs a tuple of weights for the residual blend poses.

Execution Provider

Determines which ONNX execution provider to use for inference. By default, the node attempts to pick the best available provider and prefers to use GPU acceleration.

Automatic

Chooses the best provider for the current system. This option prioritizes CUDA if installed, DirectML/CoreML as a fallback depending on the platform, and CPU inference if no GPU provider is available.

CPU

Perform inference using on the CPU.

CUDA

Perform inference using CUDA/cuDNN. CUDA and cuDNN must be installed using the packages provided by NVIDIA.

DirectML

Only available on Windows. Performs inference using the Windows Direct Machine Learning library.

CoreML

Only available on macOS. Performs inference using Apple’s Core ML library.

Device ID

This parameter can be used to specify which GPU to use for inference when multiple GPUs are available. It has no effect when using CPU inference.

Transform Group

The transform group to serialize the pose for inclusion into the data set on which Model File was trained.

Tip

The joints in the transform group should match the skeleton joint group used when training the model. The Agent from Rig node’s Point Groups parameter can be used to import joint groups when creating an agent from skeleton geometry.

Blend Shapes Source

Packed Residual

The Residual Blend Shapes input is expected to consist of packed primitives. Each packed primitive is expected to contain a residual blend shape, which is a blend shape that has the capture shape subtracted from it. This is the format generated by the ML Train Deformer recipe.

Shape Geometry

Use blend shapes which already exist on the Shape Geometry input. The blend shapes are expected to be in the format of the Rest Geometry input for the Character Blend Shapes SOP.

Inputs

Agents

Agent primitives to be modified.

Shape Geometry

Input providing the shapes to be imported. The name primitive attribute is used to partition the input geometry into separate agent shapes.

Residual Blend Shapes

Packed primitives containing residual blend shapes. The first residual blend shape always adds a weight of 1. The remaining residual blend shapes are scaled by the weights predicted by the ML model before they are applied.

See also

Geometry nodes