RifeTensorrt Interpolation using Tensortcore and upscaler.

- ALL-ONE-CLICK install compatible with comfyu desktop and comfyui_portable:

Choose your CUDA version first in the version list up top then Download, Extract then click on INSTALL_RIFE.bat Drag your comfyui folder EX: (Like just G:\ComfyUI_windows_portable1) Drag the comfyui folder in the command prompt then they will ask you to install what wheel you want. Choose your python CP38 to CP313(this is your python version). Thats it! Your set and ready to interpolate like a maniac. Note that once its installed you can also install the comfyui-upscaler-tensorrt from the manager.

Once fully installed you can delete the install no worries.

https://github.com/yuvraj108c/ComfyUI-Rife-Tensorrt?tab=readme-ov-file

https://github.com/yuvraj108c/ComfyUI-Upscaler-Tensorrt

Thanks to https://github.com/yuvraj108c/ComfyUI-Rife-Tensorrt for making the Node.

I wanted to share this cause i was using VF1 film interpolation and rife interpolation but they were kinda slow when interpolating a 20 sec video. So i made a Tutorial on how to make Rife tensorrt work. It is super fast. There also an upscaler using tensortcore it is quite fast aswell.

Rife tensorrt : https://github.com/yuvraj108c/ComfyUI-Rife-Tensorrt is a interpolator that is super quick i recently discovered it. But to make it work is quite tricky but i think it is worth the time. with rife tensorrt I can interpolate a video of like 20 sec in like 5 or 10 sec it is super quick. I will help you to get it installed step -by -step.

- MANUAL METHOD:

-Step 1: Go to the NVIDIA TensorRT download page: https://developer.nvidia.com/tensorrt

-Step 2: You will need an account, so log in or subscribe to nvidia

-Step 3: once your logged in, download the right file, for me it was TensorRT 10.14.1 GA for Windows 10, 11, Server 2022 and CUDA 13.0 ZIP Package since i downloaded the latest comfyui portable, it is Cuda130.

-Step 4: Once downloaded, extract it somewhere in a folder. Then click on the folder open it then go in the PYTHON folder, inside you will see: tensorrt-10.14.1.48-cp313-none-win_amd64.whl. CP313 is the python version.

-step 5: Now you need to install that previous wheel from step 4. Go to your comfyui folder EX:(G:\ComfyUI_windows_portable\python_embeded) Type CMD in the search BAR it will open your command window type: python.exe -m pip install ...\...\python\tensorrt-10.14.1.48-cp313-none-win_amd64.whl, Where is located your downloaded file tensorrt-10.14.1.48-cp313-none-win_amd64.whl. Install it.

-step 6: Go to your python_embeded FOLDER EX:(G:\ComfyUI_windows_portable\python_embeded\Lib\site-packages INSIDE the site-packages CREATE a new folder named ¨tensorrt_libs¨. Inside that folder you will copy everything from here: https://huggingface.co/pitou69/tensorrt_libs/tree/main/tensorrt_libs. (NOTE: Those files can be use with python 3.13 or 3.12 and any cuda). Download everything and copy inside the tensorrt_libs folder.

-step 7: Now you will need to create a new folder inside you model folder EX: (G:\ComfyUI_windows_portable\ComfyUI\models). Name the folder tensorrt. Inside that folder create a other folder named rife. then download the engine: https://huggingface.co/pitou69/tensorrt_libs/tree/main/tensorrt/rife. Place rife47_custom.engine inside it.

Here are all the rife engine(model) if you want to test it all: https://huggingface.co/pitou69/tensorrt_libs/tree/main/rife%20engine. since rife 49 was recommended i built the engine for rife 49 and 48.

• MANUAL DOWNLOAD : ( Only for python 3.13 and 3.12 CU130, You can still take the tensorrt_libs but you will have to do step 1 to 4 from the tutorial. Build the wheel if you have another CUDA like CU124 etc...).

If you decide to trust my archive and download my file it is all the file from the python_embeded Here what to do.

-step 1: download the archive then extract it in a folder.

-step 2: Copy all three file tensorrt, tensorrt_libs and tensorrt-10.14.1.48.dist-info into : G:\ComfyUI_windows_portable\python_embeded\Lib\site-packages.

-step 3: go to your model folder EX: (G:\ComfyUI_windows_portable\ComfyUI\models) create a folder name tensorrt. Inside that folder create a folder name rife. Inside the rife folder copy the rife47_custom.engine file and thats it.

This quick install is only if you trust other download. Otherwise do the Tutorial i made. And Also that quick install is only for the new comfyui portable so Python 3.13 or 3.12 CU130(cuda130).

Thats it! your ready to go and have Super quick interpolation.

EDIT: You can install The tensort upscaler aswell when everything is done. https://github.com/yuvraj108c/ComfyUI-Upscaler-Tensorrt you can download both the interpolator and the upscaler from the manager. The upscaler will work once you done the tutorial or download the archive. NOTE that the upscaler take a bit of VRAM when upscaling from 480p or even more if 720p. I didnt find a solution to reduce the vram usage. I would suggest to put a Clean VRAM Used node after the interpolation rife tensorrt node it can help.