Intrinsic Lora Sampling (Intrinsic_lora_sampling) - ComfyUI-KJNodes

The Intrinsic Lora Sampling node, whose technical core is the integration and application of the "Intrinsic LoRA" series of lightweight models.

Intrinsic Lora Sampling is a specialized sampling node for processing intrinsic properties of images. By loading specific LoRA models, it transforms input ordinary images into intrinsic property maps such as depth maps and normal maps. This node simplifies the process of extracting physical properties from images, eliminating the need for complex multi-step workflows.

Node Functionality

The technical core of this node is the integration and application of the "Intrinsic LoRA" series of lightweight models. It receives an IMAGE or LATENT input, combines it with a specified task type (e.g., depth estimation) and a text prompt, and generates the corresponding intrinsic property image through a single sampling step. The node supports batch processing for efficiency and allows direct LATENT input to skip the encoding step.

Node Parameter Description - Intrinsic Lora Sampling

Connection Inputs (Inputs)

Parameter Name	Data Type	Required	Default Value	Range/Options	Description
`model`	MODEL	Yes	-	-	Connect the base diffusion model used for denoising and sampling.
`lora_name`	FOLDER_PATHS	Yes	-	-	Specify the Intrinsic LoRA model file to load and apply from the built-in list.
`clip`	CLIP	Yes	-	-	Connect the CLIP text encoder for processing text prompts.
`vae`	VAE	Yes	-	-	Connect the Variational Autoencoder for encoding input images into latent space representations.
`image`	IMAGE	No	-	-	Input the original image to be processed. When this parameter is empty, `optional_latent` will be used.
`optional_latent`	LATENT	No	-	-	Provide an optional latent variable input to skip the image encoding step.

Control Parameters (Parameters)

Parameter Name	Data Type	Required	Default Value	Range/Options	Description
`task`	COMBO	Yes	depth map	"depth map", "surface normals", "albedo", "shad...	Specifies the target intrinsic property to generate. Options include `depth map`, `surface normals`, `albedo`, and `shading`.
`text`	STRING	Yes	-	-	Provide additional text prompts to assist in guiding the property generation process.
`per_batch`	INT	Yes	16	1 - 4096 (Step: 1)	Sets the number of images processed per batch to control memory usage. The range is 1 to 4096, with a step of 1.

Output

Parameter Name	Data Type	Description
IMAGE	IMAGE	Outputs the generated intrinsic property image.
LATENT	LATENT	Outputs the latent variable representation obtained after sampling.

Usage Scenarios

In a material analysis or 3D reconstruction workflow, you can connect a real photograph to this node, select the "depth map" task, and quickly obtain a depth information map of the scene. The generated depth map can be further used for synthesizing new viewpoints, adjusting depth of field, or as input for other 3D software.

Notes

Please note that this node relies on a specific built-in LoRA model library, and its effectiveness is limited by the capabilities of these pre-trained models. If the input image does not match the task type (e.g., generating a normal map from a solid color image), meaningful results may not be obtained.

Intrinsic Lora Sampling Node Source Code Link

The Intrinsic Lora Sampling node is from the ComfyUI-KJNodes node package.

Intrinsic Lora Sampling(Intrinsic_lora_sampling)