LTXVDurationPredictor - ComfyUI Built-in Node Documentation - ComfyUI

This node predicts the natural duration of a shot for a prompt using an LTX 2.4 duration head. It converts the predicted duration into a frame count that fits the VAE’s frame grid, using the supplied frame rate and minimum/maximum duration limits.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The model used to preprocess the text embeddings and run the duration head.	MODEL	Yes	N/A
`positive`	The conditioning that provides the prompt’s text embeddings and metadata for the duration prediction.	CONDITIONING	Yes	N/A
`duration_head`	LTX 2.4 duration head loaded with ModelPatchLoader. Must be an LTX duration head.	MODEL_PATCH	Yes	N/A
`frame_rate`	Frame rate in frames per second used to convert seconds to frames (default: 24.0).	FLOAT	Yes	1.0 to 120.0
`min_seconds`	Minimum duration in seconds used when converting the prediction to a frame count (default: 1.0).	FLOAT	Yes	0.5 to 120.0
`max_seconds`	Maximum duration in seconds used when converting the prediction to a frame count (default: 20.0).	FLOAT	Yes	0.5 to 120.0

Note: The duration_head input must be an LTX 2.4 duration head loaded with ModelPatchLoader. If the connected model patch is not an LTX duration head, the node raises a ValueError.

Outputs

Output Name	Description	Data Type
`num_frames`	The predicted duration converted to a number of frames and snapped to the VAE’s 8k+1 frame grid.	INT
`seconds`	Raw (unclamped) predicted duration. This is the value before snapping to the frame grid.	FLOAT

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): ebbf6a2601a955122ab9862142aa475524c1f38403f4ef8dc9ffee6456ee8ce5