enhancr is an elegant and easy to use GUI for Video Frame Interpolation and Video Upscaling which takes advantage of artificial intelligence - built using node.js and Electron. It was created to enhance the user experience for anyone interested in enhancing video footage using artificial intelligence. The GUI was designed to provide a stunning experience powered by state-of-the-art technologies without feeling clunky and outdated like other alternatives.

It features blazing-fast TensorRT inference by NVIDIA, which can speed up AI processes significantly. Pre-packaged, without the need to install Docker or WSL (Windows Subsystem for Linux) - and NCNN inference by Tencent which is lightweight and runs on NVIDIA, AMD and even Apple Silicon - in contrast to the mammoth of an inference PyTorch is, which only runs on NVIDA GPUs.

• Encodes video on the fly and reads frames from source video, without the need of extracting frames or loading into memory
• Queue for batch processing
• Live Preview of output media
• Allows chaining of interpolation, upscaling & restoration
• Offers the possibility to trim videos before processing
• Can load custom ESRGAN models in onnx format
• Has Scene Detection built-in, to skip interpolation on scene change frames & mitigate artifacts
• Color Themes for user customization
• Discord Rich Presence, to show all your friends progress, current speed & what you're currently enhancing
• ... and much more

RIFE (NCNN) - megvii-research/ECCV2022-RIFE - powered by styler00dollar/VapourSynth-RIFE-ncnn-Vulkan

RIFE (TensorRT) - megvii-research/ECCV2022-RIFE - powered by AmusementClub/vs-mlrt & styler00dollar/VSGAN-tensorrt-docker

CAIN (NCNN) - myungsub/CAIN - powered by mafiosnik/vsynth-cain-ncnn-vulkan (unreleased)

CAIN (TensorRT) - myungsub/CAIN - powered by HubertSotnowski/cain-TensorRT

Thanks to HubertSontowski and styler00dollar for helping out with implementing CAIN.

waifu2x (NCNN) - nagadomi/waifu2x - powered by nlzy/vapoursynth-waifu2x-ncnn-vulkan

RealESRGAN (TensorRT) - xinntao/Real-ESRGAN - powered by AmusementClub/vs-mlrt

AnimeSR (TensorRT) - TencentARC/AnimeSR - powered by mafiosnik/AnimeSR-TensorRT

Thanks to HubertSontowski for helping out with implementing AnimeSR

DPIR (TensorRT) - cszn/DPIR - powered by AmusementClub/vs-mlrt

• Dual Core CPU with Hyperthreading enabled
• Vulkan-capable graphics processor
• Windows 10

• Quad Core Intel Kaby Lake/AMD Ryzen or newer with Hyperthreading enabled
• 16 GB RAM
• NVIDIA 1000 Series (Pascal) for TensorRT/NCNN or AMD Polaris for NCNN
• Windows 11

Download the installer on Patreon for newest pre-packaged builds with all dependencies. This repository does not provide builds & the embeddable python environment that's included on Patreon.

There probably will be free versions down the line, after a backlog of versions has released.

The GUI was built with cross-platform in mind and references both operating systems already. The current focus lies on delivering a stable & working solution for Windows first, but support for Linux and macOS will follow very soon.

Support for Apple Silicon is planned as well, but I currently only have an Intel Macbook Pro available for testing.

Input size: 1920x1080 @ 2x

¹ _{Ryzen 5 3600X - Gainward RTX 2060 Super @ Stock}

² _{Ryzen 7 3800X - Gigabyte RTX 3070 Eagle OC @ Stock}

³ _{Ryzen 9 5950X - ASUS RTX 4090 Strix OC - @ ~3100MHz with curve to achieve maximum performance}

Why do I have to convert engines for CAIN (TensorRT) multiple times?

This is a workaround to make CAIN work on TensorRT inference. It doesn't support dynamic shapes, and you have to convert an engine file for every new video resolution. It's a bit inconvenient at first, but definitely worth the 4x-5x speed boost compared to NCNN, especially from the point on once all the engines for your most used resolutions are converted.

Where can I get custom upscaling models?

I can't include most models in enhancr because of Licensing. Custom models can be obtained here and loaded inside the models tab.

I'm running out of memory when trying to run a TensorRT inference, what should I do?

Either lower concurrent streams in settings or set up tiling in combination with a smaller engine resolution to decrease Video Memory usage.

I get "Python exception: operator (): no valid optimization profile found"

This means that shapes are not fitting, make sure you are using custom shapes in settings if you go over 1080p input or use correct tiling settings if you use tiling.

TensorRT is a highly optimized AI inference runtime for NVIDIA GPUs. It uses benchmarking to find the optimal kernel to use for your specific GPU, and there is an extra step to build an engine on the machine you are going to run the AI on. However, the resulting performance is also typically much much better than any PyTorch or NCNN implementation.

ncnn is a high-performance neural network inference computing framework optimized for mobile platforms. Ncnn does not have any third party dependencies. It is cross-platform, and runs faster than all known open source frameworks on most major platforms. It supports NVIDIA, AMD, Intel Graphics and even Apple Silicon. Ncnn is currently being used in many Tencent applications, such as QQ, Qzone, WeChat, Pitu and so on.

You can support this project over on Patreon or donate via PayPal to speed up development and encourage more releases and/or contribute yourself by opening a pull request. I appreciate every form of support towards this project and it means a lot!

The code is somewhat messy at some points right now, due to time pressure I stood under, but I'll rewrite affected files one by one with time. Of course there will be updates with new features and bugfixes regularly and also implementations of new interesting technologies and models when they release.

To interact with the community, share your results or to get help when encountering any problems visit our discord. Previews of upcoming versions are gonna be showcased on there as well.