LTXVModalityGuidance - ComfyUI Built-in Node Documentation

This node applies cross-modal (audio-video) guidance to an LTXV-AV model. During sampling, it runs one extra forward pass per step with the audio-to-video and video-to-audio cross-attention connections disabled, then pushes the result toward the coupled prediction. This strengthens audio-visual synchronization, such as lip-sync. The reference default for modality_scale is 3.0; setting it to 1.0 disables the extra pass.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The base model to which modality guidance will be applied. It is cloned internally, leaving the original model unchanged.	MODEL	Yes	-
`modality_scale`	Strength of the audio-video coupling guidance. Default is 3.0. Set to 1.0 to disable the extra forward pass.	FLOAT	Yes	1.0 to 100.0 (default: 3.0)
`start_percent`	The point in the sampling process, as a percentage from 0.0 to 1.0, at which modality guidance begins. Default is 0.0.	FLOAT	Yes	0.0 to 1.0 (default: 0.0)
`end_percent`	The point in the sampling process, as a percentage from 0.0 to 1.0, at which modality guidance ends. Default is 1.0.	FLOAT	Yes	0.0 to 1.0 (default: 1.0)

Guidance is applied only for sampling steps whose sigma values fall within the range defined by start_percent and end_percent. Outside this range, the node returns the denoised result unchanged. A modality_scale of 1.0 also disables the extra forward pass entirely.

Outputs

Output Name	Description	Data Type
`model`	The cloned model with a post-CFG guidance function attached. This modified model applies modality guidance during sampling.	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): 038be607c42e626a8a8f5fe336ee466d0847d43835edb71e20ff38f668069cfb

​Inputs

​Outputs

Inputs

Outputs