LDF-VFI Sequence Interpolate

A ComfyUI node in video/LDF-VFI with 10 inputs and 3 outputs.

By ethanfel·Created 6 months ago·Updated 10 days ago· 14

◄temporal_factor8►

◄sampling_steps16►

◄t_shift8.0►

◄t_cond0.10►

◄seed42►

◄offload_aftertrue►

◄source_fps0.00►

◄target_fps0.00►

Categoryvideo/LDF-VFI

Inputs (10)

Name	Type	Default	Description
images	IMAGE		Ordered source sequence. LDF models it holistically with internal skip-concat chunks.
model	LDF_VFI_MODEL		LDF-VFI model from the Load LDF-VFI Model node.
temporal_factor	INT	82–16	Native temporal upsampling factor. LDF supports every integer from 2x through 16x.
sampling_steps	INT	161–100	Diffusion steps per temporal chunk. Official quick start uses 16; fewer is faster but may reduce quality.
t_shift	FLOAT	8.00.1–20	Diffusion timestep shift. Official quick start uses 8.
t_cond	FLOAT	0.100–1	Noise applied to autoregressive boundary latents during skip-concat sampling.
seed	INT	420–2147483647	Seed for VAE sampling and diffusion noise.
offload_after	BOOLEAN	true	Move the ~6.4 GB model stack back to CPU after generation to release VRAM.
source_fps	FLOAT	0.000–1000	Input FPS. Set with target_fps to produce the requested output cadence.
target_fps	FLOAT	0.000–1000	Optional exact output FPS. Chooses the smallest native integer factor up to 16x, then selects the nearest generated frames.