advanced/guidancegenerated

Skip Layer Guidance WanVideo(SkipLayerGuidanceWanVideo)

This node technically implements a simplified skip-layer guidance, with its core mechanism being that it only skips the processing of unconditional prompts in selected denoising blocks, while the processing of conditional prompts proceeds normally.

Skip Layer Guidance Wan Video

model
MODEL
blocks
10
start_percent
0.200
end_percent
1.000
KJNodes

The Skip Layer Guidance WanVideo node technically implements a simplified skip layer guidance mechanism. Its core mechanism is to skip the processing of unconditional prompts only in the selected denoising blocks, while the processing of conditional prompts proceeds normally.

Skip Layer Guidance WanVideo is a guidance node for video generation workflows. It optimizes the model's behavior during specific denoising stages by simplifying the skip layer mechanism. Its core function is to selectively skip the processing of unconditional (negative) prompts in specified denoising blocks, aiming to improve generation efficiency or achieve specific effects.

Node Function

This node technically implements a simplified skip layer guidance mechanism. Its core mechanism is to skip the processing of unconditional prompts only in the selected denoising blocks, while the processing of conditional prompts proceeds normally. It determines the scope of effect by receiving a denoising progress percentage range (start_percent to end_percent) and a block identifier string (blocks). It is important to note that this node is currently only compatible with specific caching mechanisms (such as TeaCacheKJ) because it relies on the current_percent information within transformer_options to determine the current processing stage.

Node Parameter Description - Skip Layer Guidance WanVideo

Connection Inputs (Inputs)

Parameter NameData TypeRequiredDefault ValueValue Range/OptionsDescription
modelMODELYes--Receives a base model, which serves as the object for this node's skip layer guidance processing.

Control Parameters (Parameters)

Parameter NameData TypeRequiredDefault ValueValue Range/OptionsDescription
blocksSTRINGYes10-Specifies the block identifier(s) to apply skip layer guidance to, input as a string, e.g., "10".
start_percentFLOATYes0.20.0 - 1.0 (Step: 0.001)Sets the starting denoising progress percentage point where skip layer guidance begins to take effect, ranging from 0.0 to 1.0 with a step of 0.001.
end_percentFLOATYes1.00.0 - 1.0 (Step: 0.001)Sets the ending denoising progress percentage point where skip layer guidance stops taking effect, ranging from 0.0 to 1.0 with a step of 0.001.

Output

Parameter NameData TypeDescription
MODELMODELOutputs the new model after skip layer guidance processing.

Usage Scenarios

In a video generation workflow, when you want the model to focus more on conditional prompts and reduce interference from unconditional prompts during the mid-to-late stages of the denoising process (e.g., from 20% to 100% progress), you can insert this node after the model loading node. This helps maintain visual coherence while making the video content adhere more closely to your text description. It is commonly used in scenarios requiring fine-grained control over the dynamic changes of generated content.

Notes

This node has been marked as experimental (EXPERIMENTAL) and deprecated (DEPRECATED), meaning its functionality may be unstable or it will be removed in future versions. When using it, you must ensure the workflow includes a compatible caching node (such as TeaCacheKJ); otherwise, an error will occur.

The Skip Layer Guidance WanVideo node is from the ComfyUI-KJNodes node package.

Comments

Sign in with GitHub to join the discussion.

Loading comments…