Skip to content
,

ComfyUI Frame Interpolation (2026): Smooth AI Video With RIFE and FILM

6–10 minutes
Smiling woman in a green blazer on a city rooftop at dusk holding a card reading RIFE, with the text Smooth AI Video: 16 to 32 FPS

Most local AI video models produce short clips at a low frame rate. Wan 2.2, for example, outputs 16 frames per second, which can look a little choppy next to normal 24 or 30 fps video. Frame interpolation fixes that. It creates new in-between frames so motion looks smooth, without generating the whole video again.

In ComfyUI, the standard tool for this is the ComfyUI-Frame-Interpolation node pack by Fannovel16. It includes RIFE, FILM and several other interpolation models. RIFE is the one most people use because it is fast and looks good.

This guide shows how to install the pack and build a simple Load Video, RIFE, Save Video workflow.

It covers picking the right model and settings, getting the frame rate maths right, adding interpolation to the end of a Wan workflow, and fixing the common errors. It is checked against the node pack’s README and source as of November 2026.

TL;DR

  • Install: ComfyUI Manager > search ComfyUI-Frame-Interpolation (Fannovel16) > Install > restart.
  • Node: RIFE VFI with rife49 (default) is the best all-round choice.
  • multiplier 2 doubles the frames: 16 fps becomes 32 fps. Set the output frame_rate to fps × multiplier, or the video plays in slow motion.
  • Videos: use VideoHelperSuite’s Load Video and Video Combine nodes.
  • Not NVIDIA? set ops_backend: taichi in the pack’s config.yaml.
  • Model files download automatically into the pack’s ckpts folder on first use.

What Frame Interpolation Does

An interpolation model looks at two neighbouring frames and estimates how every pixel moved between them. It then draws one or more new frames in between.

The result is smoother motion at a higher frame rate. Typical uses:

  • Turning 16 fps AI video into smooth 32 fps.
  • Making a slow-motion version of a clip by keeping the original frame rate.
  • Smoothing jerky animation from image-to-video models.
  • Creating morph transitions between two still images.

Interpolation is not the same as upscaling. It adds frames, not pixels. If you also want sharper video, pair it with a video upscaler such as RecaVSR or RelayVSR x2.

Install ComfyUI-Frame-Interpolation

With ComfyUI Manager (easiest)

  1. Open ComfyUI and click Manager > Custom Nodes Manager.
  2. Search for ComfyUI-Frame-Interpolation by Fannovel16.
  3. Click Install, then restart ComfyUI.
  4. Also install ComfyUI-VideoHelperSuite if you do not have it. It provides the Load Video and Video Combine nodes.

Manual install

From your ComfyUI folder:

cd custom_nodes
git clone https://github.com/Fannovel16/ComfyUI-Frame-Interpolation
cd ComfyUI-Frame-Interpolation

Then run the installer that matches your system:

  • Windows: run install.bat (it uses ComfyUI’s Python).
  • Linux or macOS: run python install.py with the same Python environment ComfyUI uses.

AMD, Intel and Apple GPUs

Some models in the pack use CuPy, which needs an NVIDIA GPU. On other hardware, open config.yaml inside the pack folder and set:

ops_backend: "taichi"

Restart ComfyUI after changing it. RIFE itself runs on plain PyTorch and is the safest choice on non-NVIDIA cards.

New to custom nodes? Our list of must-have ComfyUI custom nodes explains how they work and which ones are worth installing.

RIFE, FILM and the Other Models

The pack ships many interpolation nodes. Each downloads its own checkpoint into custom_nodes/ComfyUI-Frame-Interpolation/ckpts the first time you run it.

Node Speed Quality Notes
RIFE VFI Very fast Very good Best default for AI video
FILM VFI Slow Excellent on big motion Good for large gaps between frames
GMFSS Fortuna VFI Slow Excellent on anime Popular for animation
IFRNet VFI Fast Good Lightweight alternative
AMT VFI Medium Very good Several model sizes
STMFNet VFI / FLAVR VFI Medium Good Need at least 4 input frames

Start with RIFE. Only switch to FILM or GMFSS if you see warping on fast motion that RIFE cannot handle.

Build a Simple Interpolation Workflow

This workflow takes an existing video, doubles its frames and saves a smooth version.

  1. Add Load Video (Upload) from VideoHelperSuite and choose your clip. Note its frame rate.
  2. Add RIFE VFI (under ComfyUI-Frame-Interpolation > VFI). Connect the IMAGE output of Load Video to its frames input.
  3. Set ckpt_name to rife49.pth and multiplier to 2.
  4. Add Video Combine and connect the RIFE output to its images input.
  5. Set Video Combine’s frame_rate to the original fps × 2 (for example 32 for a 16 fps clip).
  6. Pick video/h264-mp4 as the format and click Queue.

The finished file appears in your ComfyUI output folder. Play it next to the original and the difference is usually obvious straight away.

RIFE VFI Settings Explained

Setting What it does Recommended
ckpt_name RIFE version: rife47, rife49, rife417, rife426 and others rife49 (default); try rife426 for newer results
multiplier How many frames per original frame (2 doubles, 3 triples) 2
fast_mode Faster, slightly lower quality On by default; try off for final renders
ensemble Runs extra passes for better quality On (ignored by rife426)
scale_factor Internal flow resolution, from 0.25 to 4 1.0; use 0.5 for 4K input
clear_cache_after_n_frames Frees VRAM every N frames 10; lower if you run out of memory

RIFE needs at least 2 input frames. If you feed it a single image, it will fail.

Get the Frame Rate Right

This is the step most people get wrong. Interpolation adds frames, but the video player only knows how fast to play them from the frame rate you set.

Original Multiplier Set frame_rate to Result
16 fps, 81 frames 2 32 Same length, smoother
16 fps 2 16 Twice as long, slow motion
24 fps 2 48 Same length, very smooth
12 fps 3 36 Same length, smooth

Want a standard 30 fps file? Interpolate 16 fps to 32 fps, then set Video Combine to 30. The clip will be very slightly slower (about 6%), which most viewers will not notice.

Add Interpolation to a Wan Video Workflow

You do not need a separate workflow. Add RIFE straight after the VAE decode in your video workflow.

  1. Open your Wan 2.2 workflow, such as the one in our Wan 2.2 16GB 4-step guide.
  2. Find the VAE Decode node that feeds the video output.
  3. Insert RIFE VFI between VAE Decode and Video Combine.
  4. Set the multiplier to 2 and the Video Combine frame_rate to 32.
  5. Optional: keep a second Video Combine connected directly to VAE Decode so you save both versions.

The same approach works for any image-to-video model. See our ComfyUI image-to-video guide for more options.

Interpolate or Generate More Frames?

You could also ask the video model for more frames in the first place. Here is how the two approaches compare.

Generate more frames Interpolate with RIFE
Speed Slow; generation time grows with frame count Fast; seconds for a short clip
VRAM High; longer clips can run out of memory Low; frames are processed in pairs
New motion Yes; the model can add real movement No; it only fills gaps in existing motion
Best for Long, complex scenes Smoothing short AI clips

For most local setups, the best mix is to generate a short clip at the model’s native frame rate, then double it with RIFE.

Tips for the Best Results

  • Generate fewer frames, then interpolate. Making 41 frames and doubling them is much faster than generating 81.
  • Avoid scene cuts. Interpolation across a hard cut creates a messy blend. Split clips first.
  • Upscale after interpolating if VRAM is tight, so the interpolation model works on smaller frames.
  • Use scale_factor 0.5 for high resolutions; it speeds things up with little quality loss.
  • Do a short test with the Load Video frame_load_cap option before running a long clip.

Troubleshooting

Problem Likely cause Fix
Video plays in slow motion Frame rate not multiplied Set Video Combine frame_rate to fps × multiplier
CuPy or CUDA error on AMD/Intel/Mac CuPy needs NVIDIA Set ops_backend: "taichi" in config.yaml, or use RIFE
Nodes missing after install Dependencies not installed Run install.bat or install.py, check the console, restart
Checkpoint download fails GitHub or Hugging Face blocked Download the file manually into the pack’s ckpts folder
Out of memory on long clips Too many frames held in VRAM Lower clear_cache_after_n_frames, use scale_factor 0.5, or split the clip
Warping or ghosting on fast motion Motion too large between frames Try FILM VFI, or generate at a higher base frame rate
Error with STMFNet or FLAVR Fewer than 4 input frames Use RIFE, or feed more frames

Still stuck with ComfyUI basics? Our ComfyUI guide covers updating, folders and the console.

FAQ

Which RIFE model should I use?

rife49 is the default and a safe choice. rife426 is newer and worth testing, though it ignores the ensemble option.

Can I go from 16 fps to 60 fps?

Yes, with a multiplier of 4 (64 fps) and then setting the output to 60. Higher multipliers increase the risk of artefacts on fast motion.

Does frame interpolation work on images?

Yes. Feed two or more images in a batch and RIFE will create frames between them, which makes a simple morph.

Is RIFE free for commercial use?

The RIFE code is MIT licensed. Check the licence of each model you use in the pack, as some others differ.

Will interpolation fix flicker?

Not really. It smooths motion between frames but can carry flicker through. Fix flicker in the generation step first.

How long does it take?

RIFE is quick. On a modern NVIDIA GPU, doubling a short 480p or 720p clip usually takes seconds, not minutes.

Sources

Free weekly email

Get new AI guides every week

One short weekly email with our newest local AI, ComfyUI and AI tool guides. No spam, no pop-ups, just the good stuff.

Free. Unsubscribe anytime in one click. We only use your email to send this newsletter. See our privacy policy.

Smiling woman in a green blazer on a city rooftop at dusk holding a card reading RIFE, with the text Smooth AI Video: 16 to 32 FPS