NVVFR Video Processor

The name sounds like an AI model, but NVVFR isn't one. It's a thin wrapper that shells out to rigaya's NVEncC - the same command-line encoder that powers a chunk of the Windows encoding world - and strings three jobs together in a single pass: NVIDIA's hardware super-resolution, frame-rate doubling, and H.264/H.265 re-encoding. No model download, no Python deps, no VRAM budget to think about. It uses your GPU's fixed-function silicon and leaves the tensor cores for the stuff that actually needs them.

You'd reach for it at the end of a generation pipeline. You render a clip with Wan, LTX-Video or AnimateDiff, save it, and this node is the "finish the file" step: upscale it, smooth out the motion, and hand you back a clean .mp4 in ComfyUI's output folder. If you've read our upscaling notes, the mental model is "closer to Lanczos than to a diffusion model" - NVIDIA's NVVFX super-res cleans up pixelation and compression fast, but it won't invent pores, eyelashes, or fabric weave, and it can't repair a genuinely broken source. For an already-clean render that's a feature, not a bug: you get more pixels and you get them in seconds, per frame, for free.

How it works

Under the hood the node does three things. It finds nvenc/NVEncC64.exe inside its own folder, checks your input file exists, then builds an NVEncC command line:

NVEncC64.exe -c hevc -u P4 --output-res 1920x1080 --output-depth 10
  --vpp-resize algo=nvvfx-superres,superres-mode=1,superres-strength=1.0
  --vpp-fruc double -i input.mp4 --output output.mp4

--vpp-resize with nvvfx-superres is the hardware super-res, --vpp-fruc double is the frame interpolation, and everything else is plain encoding. Both effects come from NVIDIA's Video Effects SDK, which is why the install has two extra manual steps and why it needs an RTX 20-series card or newer. One thing worth knowing before you fiddle with settings: the quality dropdown is buggy as shipped. The intended high → preset P7 branch tests for the string "quality" instead of "high", so it's dead code - as it stands, high and low both map to the fast P1 preset and medium maps to P4. Output still looks fine; just don't expect high to buy you anything yet.

Inputs and outputs that matter

Most of the eleven inputs are self-explanatory. The ones a beginner actually sets:

• video_path - a plain text field, not a file picker. Paste the full absolute path (backslashes and all, on Windows).
• enable_superres and superres_strength - the upscaling, strength 0.0–1.0. Turn it off and the node becomes a plain resizer + re-encoder.
• enable_double_frame - hardware frame-rate doubling via --vpp-fruc double.
• width / height - target resolution (64–8192).

The output is output_path: a STRING you can wire anywhere, plus a small in-node video preview. Files land in ComfyUI/output/ as {output_prefix}_00001.mp4, auto-incrementing so you never overwrite. Note the encoder/depth interaction: pick H.264 with 10-bit and it silently drops you to 8-bit, because H.264 doesn't carry 10-bit here.

Install

There's no requirements.txt - nothing to pip install. The real setup, per the README:

cd ComfyUI/custom_nodes
git clone https://github.com/SaturMars/ComfyUI-NVVFR

Then two manual downloads: the NVIDIA Video Effects SDK for your card, and the x64 build from the NVEnc releases page, extracted so NVEncC64.exe sits at ComfyUI/custom_nodes/ComfyUI-NVVFR/nvenc/NVEncC64.exe. Restart ComfyUI and you'll find "NVVFR Video Processor" under the video/processing category.

Troubleshooting

The error messages are unusually honest, which helps. "NVEncC64.exe not found" means the NVEnc extraction step didn't land in the right folder. "Input video file not found" means the path in the text box is wrong - check for trailing spaces and full absolute paths. If encoding itself fails, watch the console for NVEncC's stderr, and remember the hard requirements: Windows (the node spawns the .exe and uses a Windows-only subprocess flag), and an RTX 20-series or newer for the VFX effects. This one will not run on Linux, and it won't upscale on a GTX card. If you'd rather have a more flexible, maintained take on the same hardware super-res, Comfy-Org/Nvidia_RTX_Nodes_ComfyUI is the better-known sibling; NVVFR's edge is doing upscale + interpolation + encode in one pass with no VRAM cost. It's rough around the edges, but for a "pixels now" finish on a clean clip it works.