ModelPatchLoader - ComfyUI Built-in Node Documentation

The ModelPatchLoader node loads a model patch file from the modelpatches folder and prepares it for use in a workflow.

The ModelPatchLoader node loads a model patch file from the model_patches folder and prepares it for use in a workflow. It automatically detects the type of patch contained in the file, builds the matching architecture, loads the saved weights, and wraps everything in a model patcher so it can be applied to other models. It supports many specialized patch formats, including extra ControlNet branches, feature embedder models, adapters, animation/LLLite guidance modules, and similar modules.

Inputs

Parameter	Description	Data Type	Required	Range
name	The filename of the model patch to load from the model_patches folder. Select one of the available patch files from the list.	COMBO	Yes	Dynamically generated list of all model patch files found in the model_patches folder

Note: This node is marked as experimental. The patch type is detected automatically from the file contents, so no manual type selection is required. The node reads the checkpoint metadata and inspects the weight keys to decide which architecture to build (for example Qwen Image block-wise ControlNet, Z-Image ControlNet, Wan Uni3C ControlNet, MiniMax H3 Fun ControlNet, SigLIP feature projection, Lightricks duration head, Anima LLLite, MultiTalk, or SUPIR). Weights are loaded with safe loading enabled, and the model is placed on the offload device inside a CoreModelPatcher so it can later be applied to another model.

Outputs

Output Name	Description	Data Type
MODEL_PATCH	The loaded model patch wrapped in a model patcher, ready to be applied to a model in the workflow	MODEL_PATCH

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

---

Source fingerprint (SHA-256): 069f40b1f108ecd74fc58c12aa2f74edff07f743aa1ed6352ff7bcf0c39341d4