TorchCompileControlNet

TorchCompileControlNet node, which internally uses torch.compile.

TorchCompileControlNet is a utility node designed to optimize the inference performance of ControlNet models. It leverages PyTorch's just-in-time compilation technology to convert ControlNet models into a more efficient execution format, aiming to improve the processing speed of image generation workflows.

Node Functionality

This node internally uses the torch.compile function to compile the input ControlNet model. It allows users to select different compilation backends (such as inductor or cudagraphs) and optimization modes (such as max-autotune), and supports enabling full graph mode for more aggressive optimization. The compilation process is one-time, and the node internally tracks the compilation state to avoid redundant operations.

Node Parameter Description - TorchCompileControlNet

Connection Inputs (Inputs)

Parameter Name	Data Type	Required	Default Value	Value Range/Options	Description
`controlnet`	CONTROL_NET	Yes	-	-	: Input the ControlNet model to be compiled and optimized.

Control Parameters (Parameters)

Parameter Name	Data Type	Required	Default Value	Value Range/Options	Description
`backend`	COMBO	Yes	inductor	"inductor", "cudagraphs"	: Select the backend engine used for PyTorch compilation. Options are `inductor` or `cudagraphs`.
`mode`	COMBO	Yes	default	"default", "max-autotune", "max-autotune-no-cud...	: Select the compilation optimization mode. Options are `default`, `max-autotune`, `max-autotune-no-cudagraphs`, or `reduce-overhead`.
`fullgraph`	BOOLEAN	Yes	False	-	Enable full graph mode

Output

Parameter Name	Data Type	Description
CONTROL_NET	CONTROL_NET	: Outputs the compiled and optimized ControlNet model.

Usage Scenarios

In workflows requiring rapid iteration or batch image generation, you can connect this node after the ControlNet loader and before the sampler. By compiling the ControlNet, you can significantly reduce the time consumed per sampling or multiple samplings while maintaining generation quality. This is particularly suitable for real-time applications or large-scale content creation scenarios that have requirements for generation speed.

Notes

This node is marked as an experimental feature. Its compilation effectiveness and stability may vary depending on the PyTorch version, hardware drivers, and specific models. If compilation fails, the node will throw a runtime error. You need to check environment compatibility or try different compilation parameter combinations.