FrameInterpolator [🅝-🅢🅤🅘🅣🅔]

Same node as this pack's plain FrameInterpolator, just carrying the circled [N-SUITE] badge the pack switched to after its February 2024 rewrite - the schema is identical field for field. If you landed here confused about why there are two nearly identical pages, that's the whole story: the maintainer renamed the pack's node classes to dodge collisions with other packs, and both the old and new names ended up indexed. Functionally, treat this as the current, actively-registered version.

What it does

It's RIFE (Real-Time Intermediate Flow Estimation), adapted in this pack from hzwer/Practical-RIFE. RIFE isn't a pack-specific trick - it's the standard cheap way people add motion smoothness to a video in the wider ComfyUI ecosystem, the same technique that shows up stacked on top of upscalers like SeedVR2 or FlashVSR once resolution and detail are sorted. The key thing it does not do is duplicate frames to hit a higher nominal frame rate, which is the naive approach and looks stuttery. Instead it estimates motion flow between two adjacent frames and generates genuinely new, plausible frames to sit between them.

The inputs and outputs that matter

images - your existing frame batch. METADATA - the frame-rate object that came from whatever loaded those frames (LoadVideo [n-suite] or LoadFramesFromFolder); it's technically allowed to be blank, but in practice you always want it connected, since the whole point is scaling an existing, known frame rate. multiplier (default 2) sets how much you're multiplying the frame count - and the frame rate recorded in the metadata - by.

Out the other side: IMAGES, the expanded batch (double the frames at the default multiplier), and METADATA, updated with the new, higher frame rate baked in, ready for whatever writes your video out.

Installing the pack

• ComfyUI Manager - search "ComfyUI-N-Nodes," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/Nuked88/ComfyUI-N-Nodes.git, then restart ComfyUI.

Common issues & troubleshooting

Some sections of the interpolated clip look warped or ghosted. That's the flow-estimation model struggling with hard motion - fast pans, limbs crossing, objects entering or leaving frame - not a sign your install is broken. Slow, mostly-static footage interpolates cleanly; fast, chaotic motion is the genuinely hard case for any optical-flow method, not just this one.

It's slow. The cost scales with both your input frame count and the multiplier - a small neural net runs once per adjacent frame pair. If you only need the smoothing over part of a longer clip, trim the batch to that section first instead of interpolating the whole thing and discarding the rest.

No visible change in the output. Double-check METADATA is actually wired in from your source loader and not left on its default blank value - without a real frame rate to scale, there's nothing meaningful for the multiplier to act on downstream.