Differential Diffusion Advanced (DifferentialDiffusionAdvanced) - ComfyUI-KJNodes

The Differential Diffusion Advanced node works at a technical level by cloning the input model and setting a custom denoising mask function for it.

Differential Diffusion Advanced is a testing node for advanced image synthesis, allowing users to control the application intensity of the denoising process in different areas through masks and multipliers. Its core function is to modify the behavior of the diffusion model, achieving differential denoising based on spatial masks, thereby applying varying degrees of editing or influence to different parts of the image in a single generation.

Node Functionality

This node works at a technical level by cloning the input model and setting a custom denoising mask function for it. It receives latent representations, a mask, and a multiplier, and uses timestep and threshold calculations to determine at which stage of the denoising process to apply the effect to the masked area. Its internal mechanism compares the mask value with a dynamically calculated threshold to generate a binary mask that changes over time (or noise level), thereby finely controlling the model's focus on specific areas of the image.

Node Parameter Description - Differential Diffusion Advanced

Connection Inputs (Inputs)

Parameter Name	Data Type	Required	Default Value	Value Range/Options	Description
`model`	MODEL	Yes	-	-	model (input parameter): Input the diffusion model used for generation or editing. The node will apply differential denoising logic based on this model.
`samples`	LATENT	Yes	-	-	samples (input parameter): Input samples containing the latent representation of the image, serving as the base data for differential denoising processing.
`mask`	MASK	Yes	-	-	mask (input parameter): Input a mask image to define the spatial area where the differential denoising effect is applied. White areas typically indicate where the effect is applied.

Control Parameters (Parameters)

Parameter Name	Data Type	Required	Default Value	Value Range/Options	Description
`multiplier`	FLOAT	Yes	1.0	-10.0 - 10.0 (Step: 0.001)	multiplier (input parameter): A multiplier controlling the timing and intensity of the effect application, ranging from -10.0 to 10.0 with an adjustable step of 0.001. A larger value concentrates the effect more towards the later stages of the denoising process.

Output

Parameter Name	Data Type	Description
MODEL	MODEL	A copy of the model integrated with differential denoising functionality.
LATENT	LATENT	LATENT (output parameter): Outputs latent samples with attached mask information for use by subsequent sampling nodes.

Usage Scenarios

In practical applications, this node can be integrated into workflows involving local image editing or stylization. For example, a user can first use a masking tool to precisely select the image area they wish to strongly stylize or modify in detail. Then, by connecting the mask, original latent samples, and model to this node and adjusting the multiplier to control the timing and intensity of the effect application, a modified model and latent samples are output for use in subsequent sampling steps.

Notes

Please note that this node is located under the _for_testing category, meaning its functionality may be experimental, and its stability or final results may vary depending on different models and settings. When using it, be aware that the multiplier parameter has a wide range of possible values, and improper settings may lead to generation results that do not meet expectations.

Differential Diffusion Advanced Node Source Code Link

The Differential Diffusion Advanced node is from the ComfyUI-KJNodes node package.

Differential Diffusion Advanced(DifferentialDiffusionAdvanced)