SPEED Segment Interpolate

The SPEED segment node for interpolating long video with a diffusion model, memory intact

By ethanfel·Created 7 months ago·Updated about a month ago· 17

images
model
settings

images
model
elapsed_seconds

◄multiplier2►

◄clear_cache_after_n_frames10►

◄keep_devicetrue►

◄all_on_gpufalse►

◄batch_size1►

◄chunk_size0►

◄source_fps0.00►

◄target_fps0.00►

◄seed0►

◄segment_index0►

◄segment_size500►

SPEED is the pack's modern default interpolator, but it's also a diffusion model, which means every midpoint it produces started life as random noise. Interpolate a short clip and that's fine. Interpolate a long one and you've got two problems: the assembled output blows past your RAM, and the diffusion's stochastic noise wants to stay reproducible. SPEED Segment Interpolate solves both - it processes one slice at a time with a per-segment seed.

It's the same SPEED pipeline as the plain Interpolate node: multiplier (2x/4x/8x, recursive midpoints), seed, batch_size, chunk_size, keep_device, all_on_gpu, clear_cache_after_n_frames, source_fps/target_fps - plus the segment pair:

segment_index (0-based) - which slice to process.
segment_size (default 500) - input frames per segment, one-frame overlap for seamless stitching.

The pattern. One node per segment, chained by feeding each node's model output into the next's model input to force sequential execution. Each segment ends in a Video Combine save, freeing RAM before the next segment starts; Tween Concat Videos rejoins the segment_*.mp4 files afterward. Give each segment a seed you can track if reproducibility across runs matters - the same seed per segment plus identical settings reproduces each segment's noise, but changing batching or chunking can still shift how noise is assigned, so don't chase pixel-perfect reruns across a refactor.

Install. Search "Tween" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Ethanfel/ComfyUI-Tween
pip install -r requirements.txt

Restart, let the SPEED loader pull the checkpoint and runtime snapshot, and keep VideoHelperSuite installed for load/save.

Common issues. Same trio as every segment node: don't wire two segments to the same source in parallel (chain the model output or the RAM wall returns), keep loader FPS, source_fps/target_fps, and Video Combine frame_rate synchronized, and remember SPEED is stochastic - a "why does this segment look different on rerun" question usually has "you changed a boundary" as the answer.

Categoryvideo/SPEED

Inputs (14)

Name	Type	Default	Description
images	IMAGE		Input image batch. Output frame count: 2x=(2N-1), 4x=(4N-3), 8x=(8N-7).
model	SPEED_VFI_MODEL		SPEED model from the Load SPEED Model node.
multiplier	COMBO	2	SPEED is midpoint-only: 4x and 8x use recursive midpoint passes.
clear_cache_after_n_frames	INT	101–100	Clear CUDA cache every N frame pairs to prevent VRAM buildup. Lower = less VRAM but slower.
keep_device	BOOLEAN	true	Keep the ~235M-parameter model on GPU between batches. Faster, but uses substantial VRAM.
all_on_gpu	BOOLEAN	false	Store all intermediate frames on GPU instead of CPU. Much faster (no transfers) but requires enough VRAM for all frames. Recommended for 48GB+ cards.
batch_size	INT	11–64	Number of frame pairs to process simultaneously. Higher = faster but uses more VRAM. Start with 1, increase until VRAM is full. Recommended: 1 for 8GB, 2-4 for 24GB, 4-16 for 48GB+. SPEED is stochastic, so changing batch size can change how seeded noise is assigned to pairs.
chunk_size	INT	00–10000	Process the source in overlapping chunks to bound VRAM. SPEED remains repeatable for the same seed and settings, but changing chunk or batch boundaries can change its stochastic result.
source_fps	FLOAT	0.000–1000	Input frame rate. Required when target_fps > 0.
target_fps	FLOAT	0.000–1000	Target output FPS. When > 0, overrides multiplier and auto-computes a power-of-2 oversample up to 8x, then selects frames. 0 = use multiplier.
seed	INT	00–2147483647	SPEED starts from random pixel noise. The seed makes the same execution settings repeatable without reloading the model.
segment_index	INT	00–10000	Zero-based segment to process. Adjacent segments overlap by one source frame.
segment_size	INT	5002–10000	Input frames per segment. Save each segment before processing the next to bound RAM.
settingsopt	VFI_SETTINGS		Auto-tuned settings from VFI Optimizer. Overrides batch_size, chunk_size, keep_device, all_on_gpu, clear_cache_after_n_frames.

Outputs (3)

Name	Type	Description
images	IMAGE	—
model	SPEED_VFI_MODEL	—
elapsed_seconds	FLOAT	—