OpticalFlowLoader - ComfyUI Built-in Node Documentation - ComfyUI

Loads an optical flow model from the models/optical_flow/ folder. Currently, only torchvision’s RAFT-large format is supported, which is the model used by the VOIDWarpedNoise node. ComfyUI does not download optical flow weights automatically; you must place the checkpoint file manually in the models/optical_flow/ directory.

Inputs

Parameter	Description	Data Type	Required	Range
`model_name`	Optical flow model to load. Files must be placed in the `optical_flow` folder. Today only torchvision’s `raft_large.pth` is supported.	COMBO	Yes	List of files in the `models/optical_flow/` folder

Note: The selected checkpoint must be a torchvision RAFT-large state dict containing keys prefixed with feature_encoder., context_encoder., and update_block.. If the file does not match this format, the node raises a ValueError.

Outputs

Output Name	Description	Data Type
`OPTICAL_FLOW`	The loaded optical flow model, set to evaluation mode and float32 precision, wrapped in a ModelPatcher for use with other nodes.	OPTICAL_FLOW

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): 5e79551545ad7ee2fd4856a47da29808a404342d1d5e57da0980058db6b11c3b