TNTwise/REAL-Video-Enhancer

What it solves

It provides a user-friendly interface for performing high-quality video interpolation (increasing frame rate, e.g., 24 to 48 FPS) and video upscaling (increasing resolution, e.g., 1080p to 4K) across Windows, Linux, and MacOS.

How it works

The application acts as a GUI wrapper for several specialized AI models and backends. It leverages TensorRT, PyTorch (CUDA/ROCm), and NCNN (Vulkan) to run inference on various GPUs. It integrates multiple model architectures including RIFE and GMFSS for interpolation, and Real-ESRGAN and SPAN for upscaling, while using PySceneDetect to ensure sharp transitions during scene changes.

Who it’s for

Users who want to enhance the visual quality and smoothness of their videos without needing to operate command-line tools or complex AI environments.

Highlights

• Cross-platform support: Works on Windows, Ubuntu 22.04+, and MacOS 15+.
• Multiple Backends: Supports NVIDIA RTX GPUs via TensorRT, AMD/NVIDIA GPUs via PyTorch, and a wide range of Vulkan-capable devices via NCNN.
• Diverse Model Library: Includes support for a wide array of interpolation, upscaling, decompression, and denoising models.
• Real-time Preview: Displays the latest rendered frame to monitor progress.
• Scene Detection: Prevents artifacts by detecting scene changes to preserve sharp transitions.