This project provides a TensorRT implementation of RIFE for ultra fast frame interpolation inside ComfyUI
This project is licensed under CC BY-NC-SA, everyone is FREE to access, use, modify and redistribute with the same license.
If you like the project, please give me a star! ⭐
Note: The following results were benchmarked on FP16 engines inside ComfyUI, using 2000 frames consisting of 2 alternating similar frames, averaged 2-3 times
| Device | Rife Engine | Resolution | Multiplier | FPS |
|---|---|---|---|---|
| H100 | rife49_ensemble_True_scale_1_sim | 512 x 512 | 2 | 45 |
| H100 | rife49_ensemble_True_scale_1_sim | 512 x 512 | 4 | 57 |
| H100 | rife49_ensemble_True_scale_1_sim | 1280 x 1280 | 2 | 21 |
Navigate to the ComfyUI /custom_nodes directory
git clone https://github.com/huchukato/ComfyUI-RIFE-TensorRT-Auto
cd ./ComfyUI-RIFE-TensorRT-Auto
pip install -r requirements.txtThis node defaults to CUDA 13 (RTX 50 series, driver 580+).
For CUDA 12 (RTX 30/40 series):
pip install -r requirements_cu12.txtFor CUDA 13 (Default):
pip install -r requirements.txtThe node automatically detects your CUDA installation via CUDA_PATH or CUDA_HOME environment variables.
If CUDA is not detected, download from: https://developer.nvidia.com/cuda-13-0-2-download-archive
Important: The TensorRT engine supports different resolution ranges based on the selected profile:
- small profile: 384-1080px
- medium profile: 672-1312px
- large profile: 720-1920px (perfect for 1440x960 and higher resolutions)
For images larger than your selected profile's maximum, resize them before using the RIFE node or select a higher profile.
The node supports resolution profiles to optimize VRAM usage:
- small: 384-1080px (recommended for most video generation)
- medium: 672-1312px (for higher resolution videos)
- large: 720-1920px (for 4K and high-resolution content)
- custom: Connect a "RIFE Custom Resolution Config" node for manual control
The following RIFE models are supported and will be automatically downloaded and built:
- rife49_ensemble_True_scale_1_sim (default) - Latest and most accurate
- rife48_ensemble_True_scale_1_sim - Good balance of speed and quality
- rife47_ensemble_True_scale_1_sim - Fastest option
Models are automatically downloaded from HuggingFace and TensorRT engines are built on first use.
-
Load Model: Insert
Right Click -> Add Node -> tensorrt -> Load Rife Tensorrt Model- Choose your preferred RIFE model (rife47, rife48, or rife49)
- Select precision (fp16 recommended for speed, fp32 for maximum accuracy)
- Select resolution profile (small, medium, or custom)
- The model will be automatically downloaded and TensorRT engine built on first use
-
Process Frames: Insert
Right Click -> Add Node -> tensorrt -> Rife Tensorrt- Connect the loaded model from step 1
- Input your video frames
- Configure interpolation settings (multiplier, CUDA graph, etc.)
- Windows 11, CUDA 13.0, TensorRT 10.15.1.29, Python 3.12, RTX 5090
- WSL Ubuntu 24.04.03 LTS, CUDA 12.9, TensorRT 10.13.3.9, Python 3.12.11, RTX 5080
- Fixed Dependencies: Updated TensorRT to 10.15.1.29 to resolve installation conflicts
- RTX 5090 Support: Tested and confirmed compatibility with RTX 5090
- Resolution Documentation: Added clear guidance on resolution limits and preprocessing
- CUDA 13 Default: Updated to CUDA 13.0 and TensorRT 10.14.1.48
- Auto CUDA Detection: Automatically finds CUDA toolkit and DLL paths
- Resolution Profiles: Added small/medium/custom profiles to reduce VRAM usage
- Automatic Model Management: No more manual downloads! Models are automatically downloaded from HuggingFace and TensorRT engines are built on demand
- Improved Workflow: New two-node system with
Load Rife Tensorrt Model+Rife Tensorrtfor better organization - Updated Dependencies: TensorRT updated to 10.13.3.9 for better performance and compatibility
- https://github.com/styler00dollar/VSGAN-tensorrt-docker
- https://github.com/Fannovel16/ComfyUI-Frame-Interpolation
- https://github.com/hzwer/ECCV2022-RIFE
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)