ModelSamplingLTXV - ComfyUI Built-in Node Documentation

The ModelSamplingLTXV node applies advanced sampling parameters to a model based on token count.

The ModelSamplingLTXV node applies advanced sampling parameters to a model based on token count. It calculates a shift value by linearly interpolating between base_shift and max_shift across a token range, then patches the input model with a specialized sampling configuration. If a latent is provided, its dimensions determine the token count; otherwise 4096 tokens are used.

Inputs

Parameter	Description	Data Type	Required	Range
model	The input model to apply sampling parameters to.	MODEL	Yes	-
max_shift	The maximum shift value used in the linear interpolation calculation (default: 2.05).	FLOAT	Yes	0.0 to 100.0 (step: 0.01)
base_shift	The base shift value used in the linear interpolation calculation (default: 0.95).	FLOAT	Yes	0.0 to 100.0 (step: 0.01)
latent	Optional latent input used to determine the token count for the shift calculation. If not provided, a default token count of 4096 is used.	LATENT	No	-

The shift value is calculated by interpolating between base_shift at 1024 tokens and max_shift at 4096 tokens. When latent is supplied, the token count is the product of all dimensions after the first two in the latent samples (the spatial/temporal dimensions). If no latent is provided, the token count defaults to 4096.

Outputs

Output Name	Description	Data Type
model	The modified model with the applied sampling parameters.	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): aba596c5478e9d6ee821eec1eca15506935bcc765a368087ccc442fc2ed6671b