ComfyUI’s all-you-can-interpolate buffet

What it does This ComfyUI extension packages roughly a dozen video frame interpolation algorithms—RIFE, FILM, AMT, GMFSS Fortuna, and others—into workflow nodes that generate intermediate frames from an image sequence. The README concedes that about 99% of the underlying code is lifted from an existing VSGAN project, so the real value here is plumbing: turning disparate research repos into interchangeable ComfyUI nodes.

The interesting bit Rather than forcing you to pick one model, it treats interpolation as a buffet. You can swap between RIFE 4.9, FILM, or STMFNet mid-workflow to see which handles your motion best. It also layers on scheduling multipliers and a clear_cache_after_n_frames toggle to keep VRAM usage in check, plus an experimental Taichi backend for non-NVIDIA hardware.

Key highlights

• Bundles 12+ VFI models (RIFE 4.0–4.9, FILM, AMT, IFRNet, GMFSS Fortuna, etc.) as individual ComfyUI nodes.
• Supports scheduling multipliers for controlled interpolation strength.
• Includes a memory-management safety valve (clear_cache_after_n_frames) to reduce RAM/VRAM pressure.
• Experimental non-CUDA support via Taichi Lang backend.
• Designed to plug into ComfyUI-VideoHelperSuite and Advanced-ControlNet workflows.

Caveats

• Several nodes carry hard constraints: STMFNet and FLAVR require at least four input frames, while ATM-VFI and MoMo are limited to 2× interpolation.
• Non-CUDA support is experimental; if a node throws NotImplementedError, it simply isn’t supported by the Taichi backend yet.
• The project is marked WIP, and the README warns that some options (like RIFE’s fast_mode post-v4.5) are effectively no-ops.

Verdict Worth a look if you already live in ComfyUI and want to benchmark multiple interpolation algorithms without installing a dozen separate repos. If you need a standalone, polished CLI tool or a single model with guaranteed long-term support, this isn’t it.