Merserk/dlss5-visual-enhancer

What it solves

This project provides a way to apply NVIDIA's DLSS 5 Neural Rendering and DLSS Frame Generation to static images and existing video files. It allows users to enhance visual quality, upscale resolution, and increase frame rates (interpolation) for media that wasn't originally rendered with these technologies in real-time.

How it works

The application runs as a portable Windows tool with a Gradio-based local interface. It leverages NVIDIA RTX GPUs (Ampere, Ada, and Blackwell architectures) to execute a "feature-18" neural rendering pipeline for images and video, and utilizes DLSS Frame Generation for video frame interpolation. It integrates with FFmpeg for video encoding/decoding and supports a wide range of image formats (including RAW and HEIF) and video codecs (H.264, H.265, AV1, and ProRes Proxy).

Who it’s for

Content creators and NVIDIA RTX GPU owners who want to upscale or smooth out video and image content using DLSS 5's neural capabilities outside of a live gaming environment.

Highlights

• Neural Rendering: Supports various upscaling modes from DLAA (1x) up to Ultra Performance (3x) and adjustable neural settings for style, intensity, and structure.
• Frame Interpolation: Can increase video frame rates up to 120 FPS using DLSS Frame Generation on Ada and Blackwell GPUs.
• Broad Format Support: Handles numerous image formats (including camera RAW) and video containers (MP4, MKV, MOV).
• Hardware Acceleration: Separate GPU selection for AI processing and video encoding (via NVIDIA NVENC).
• HDR Support: Opt-in 10-bit output for H.265, AV1, and ProRes Proxy to preserve HDR colorspaces.