ZImageFunControlnet - ComfyUI Built-in Node Documentation

ZImageFunControlnet applies a control network patch to a base model so it can guide the image generation or editing process. It combines a model, a model patch, and a VAE, and lets you control how strongly the control effect influences the result. Optional image, inpainting image, and mask inputs allow more targeted edits. The node works with Z-Image ControlNet patches and with Qwen Image 2.1 Fun ControlNet patches loaded through the Load Model Patch node.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The base model used for the generation process.	MODEL	Yes	-
`model_patch`	A specialized patch model that applies the control network’s guidance.	MODEL_PATCH	Yes	-
`vae`	The Variational Autoencoder used for encoding and decoding images.	VAE	Yes	-
`strength`	The strength of the control network’s influence. Positive values apply the effect, while negative values can invert it (default: 1.0).	FLOAT	Yes	-10.0 to 10.0 (step 0.01)
`image`	An optional base image to guide the generation process.	IMAGE	No	-
`inpaint_image`	An optional image used specifically for inpainting areas defined by a mask.	IMAGE	No	-
`mask`	An optional mask that defines which areas of an image should be edited or inpainted.	MASK	No	-
`start_percent`	The point in the denoising process, as a fraction of the total sampling steps, where the control network starts taking effect (default: 0.0).	FLOAT	No	0.0 to 1.0 (step 0.001)
`end_percent`	The point in the denoising process where the control network stops taking effect (default: 1.0).	FLOAT	No	0.0 to 1.0 (step 0.001)

Note: The inpaint_image parameter is typically used in conjunction with a mask to specify the content for inpainting. The node’s behavior may change based on which optional inputs are provided (e.g., using image for guidance or using image, mask, and inpaint_image for inpainting). The start_percent and end_percent values restrict the control network to a window of the denoising process, and outside that window the model is sampled without the patch. If strength is 0, or if none of image, inpaint_image, and mask is connected, the node returns the base model unchanged. For Z-Image Control patches a provided mask is inverted (1.0 - mask) before use, while a Qwen Image 2.1 Fun ControlNet patch uses the mask as given. This node is flagged as experimental.

Outputs

Output Name	Description	Data Type
`model`	The model with the control network patch applied, ready for use in a sampling pipeline.	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): 9673b8b6e091713bcc93fe5fd1cfed12e6941571d1017e10ac94c19e1afd4ca1

​Inputs

​Outputs

Inputs

Outputs