LoraSave - ComfyUI Built-in Node Documentation

The LoraSave node extracts and saves LoRA (Low-Rank Adaptation) files from model differences. It can process diffusion model differences, text encoder differences, or both, converting them into LoRA format with specified rank and type. The resulting LoRA file is saved to the output directory for later use.

Inputs

Parameter	Description	Data Type	Required	Range
`filename_prefix`	The prefix for the output filename (default: “loras/ComfyUI_extracted_lora”)	STRING	Yes	-
`rank`	The rank value for the LoRA, controlling the size and complexity (default: 8)	INT	Yes	1-4096
`lora_type`	The type of LoRA to create. “standard” extracts low-rank matrices via SVD; “full_diff” saves the full weight difference (default: “standard”)	COMBO	Yes	`"standard"` `"full_diff"`
`bias_diff`	Whether to include bias differences in the LoRA calculation (default: True)	BOOLEAN	Yes	-
`model_diff`	The ModelSubtract output to be converted to a lora	MODEL	No	-
`text_encoder_diff`	The CLIPSubtract output to be converted to a lora	CLIP	No	-

Note: At least one of model_diff or text_encoder_diff must be provided for the node to function. If both are omitted, the node will produce no output.

Outputs

Output Name	Description	Data Type
-	This node saves a LoRA file to the output directory but does not return any data through the workflow	-

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): 642510bfe083130cb2724d7ae38d30d4694c85a43c7ce27230690ddd9fe76b79

​Inputs

​Outputs

Inputs

Outputs