REAL Video Enhancer
Introduction
The REAL Video Enhancer is an improved and reimagined version of the Rife ESRGAN App, catering to users across different operating systems like Windows, Linux, and MacOS. It provides easy access to advanced video processing capabilities such as frame interpolation and video upscaling, positioning itself as a modern alternative to older software like Flowframes or enhancr. This tool is designed to enhance video quality, making it smooth and visually appealing by increasing frame rates and resolution.
Features
- Windows Support: Newly supported, although users may encounter false positive Trojan alerts due to the use of PyInstaller.
- MacOS Compatibility: Support available up to version 1.2, with deprecated features in version 2.0.
- Linux Support: Compatible with Ubuntu 20.04+ through both Executable and Flatpak versions.
- Discord Integration: Offers Rich Presence support for Discord native applications and its Flatpak version.
- Scene Detection: Identifies scene changes to maintain sharp transitions in videos.
- Frame Preview: Allows users to view the latest processed frame.
- Inference Efficiency: Utilizes TensorRT and NCNN technologies for high-efficiency video processing across various GPUs.
Hardware/Software Requirements
To run the REAL Video Enhancer efficiently, certain hardware and software configurations are recommended:
- Minimum: Dual-Core 64-bit CPU, Vulkan 1.3 capable GPU, 8 GB RAM, and 1 GB storage (NCNN install).
- Recommended: Quad-Core 64-bit CPU, Nvidia RTX GPU (20 series or newer), 16 GB RAM, and 10 GB storage (TensorRT install).
- Operating Systems: Compatible with Windows 10/11 64-bit or modern Linux distributions like Ubuntu 20.04+.
Benchmarks
The performance of the REAL Video Enhancer is demonstrated using 1920x1080 video with default settings. For example, using RIFE NCNN, a GPU like the RTX 3080 can process video at up to 81 frames per second, while using RIFE TensorRT 10.3 at higher speeds, such as 270 frames per second for certain video versions.
Cloning and Building
Users can clone the project repository using the following command:
git clone https://github.com/TNTwise/REAL-Video-Enhancer
To build the project, execute the command:
python3 build.py --build_exe
For access to canary build versions, they can visit the pre-release section on GitHub.
Credits
The development of the REAL Video Enhancer is attributed to contributions from several individuals and software tools:
- People: Key contributors include developers like NevermindNilas and Styler00dollar for backend code and model contributions.
- Software: The project utilizes powerful software tools such as FFmpeg for media handling, PyTorch for neural network inference, and NCNN for Vulkan-based inference, among others.
Custom Models
The REAL Video Enhancer supports a range of custom models created by various authors to provide enhanced video processing capabilities. Some examples include:
- 4x-SPANkendata by Crustaceous D
- 4x-ClearRealityV1 by Kim2091
Frequently Asked Questions (FAQ)
General Application Usage:
- The program is designed to facilitate quick access to video interpolation and upscaling capabilities.
- For modern Nvidia GPUs (20 series and up), TensorRT is recommended for optimal performance.
TensorRT Related Queries:
- Inference might take time initially due to advanced optimization.
- Optimization failures often occur due to limited VRAM availability.
ROCm and NCNN Queries:
- Various errors might arise from using ROCm or Vulkan, often related to hardware resources.
In summary, REAL Video Enhancer stands as a robust tool for enhancing video quality with its advanced features and support for multiple platforms and hardware configurations. It brings high-performance video processing capabilities to users seeking to improve their video content effortlessly.