USOStyleReference - ComfyUI Built-in Node Documentation

The USOStyleReference node applies a style reference to a model by combining CLIP vision features with a model patch, and returns a patched copy of the input model. The visual style information is combined with the model’s text conditioning so it can influence generation. This node is intended for Flux models and is marked as experimental.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The base model to which the style reference patch is applied.	MODEL	Yes	-
`model_patch`	The model patch containing the projection model used to encode the reference image features.	MODEL_PATCH	Yes	-
`clip_vision_output`	The encoded visual features extracted from CLIP vision processing of the reference image.	CLIP_VISION_OUTPUT	Yes	-

Note: The clip_vision_output must come from a CLIP vision model that provides the full hidden states and the penultimate hidden state. The node combines the 20th-from-last, 11th-from-last, and penultimate hidden states into the style embedding. The model_patch must expose a projection model through its model attribute that converts these image features into the style embedding. During sampling, the style embedding is prepended to the text conditioning so it can influence generation, and matching zero-position text IDs are prepended to the text IDs so the identifier sequence stays aligned with the extended conditioning.

Outputs

Output Name	Description	Data Type
`model`	The modified model with the style reference patch applied.	MODEL

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): 9033dddb76fafb388c67dcd09d96102a7ab3e5bc416cec61bf18d088da37a0f0

​Inputs

​Outputs

Inputs

Outputs