NeuralScaler 4K (DLSS 5)

NeuralScaler 4K is an offline, high-performance 4K video super-resolution workstation developed upon NVIDIA DLSS 5 (Deep Learning Super Sampling 5) neural reconstruction technology. It integrates native DLSS 5 hardware pipelines, temporal-spatial motion vector inference, cross-platform color consistency management, dynamic VRAM safeguard mechanisms, and a frame-locked dual-source comparison player.

Architecture & Key Features

• NVIDIA DLSS 5 Neural Reconstruction: Harnesses native NVNGX runtimes (nvngx_dlss.dll, nvngx_dlssd.dll, nvngx_dlssnr.dll) alongside Tensor Core acceleration and Optical Flow inference to accurately reconstruct low-resolution inputs into ultra-clear 4K videos with minimal temporal artifacts.
• Cross-Platform ColorSync Engine: Eliminates the persistent industry dilemma where 4K upscaled videos look natural on PC monitors but heavily oversaturated and red-tinted on mobile OLED (Display P3) displays. Enforces chroma-plane isolation (planes=1), injects broadcast-grade MP4 colr container atoms (+write_colr), tags SPS VUI h264_metadata, and performs automated ITU-R BT.601 to BT.709 matrix transformation, ensuring 100% truthful, neutral color and skin tones across iPhone, Android, and PC.
• VRAM Safeguard & Chunked Streaming: Robust stream chunking and continuous hardware telemetry prevent CUDA Out-of-Memory faults during intensive 4K reconstruction workloads across NVIDIA GeForce RTX series GPUs.
• Frame-Locked Comparison Player: Phase-locked loop synchronization dynamically compensates for browser decoding disparities between original footage and high-bitrate 4K exports, guaranteeing strict $dx=0, dy=0$ spatial alignment and zero temporal drift.

System Requirements

• Operating System: Windows 10 / Windows 11 (64-bit)
• GPU: NVIDIA GeForce RTX Series / AMD Radeon (Dedicated VRAM >= 2GB)
• Driver: NVIDIA Driver >= 535.00
• Webview Engine: Microsoft Edge (Built-in)

Quick Start

Method A: npm CLI (Headless Super-Resolution)

# Run immediately via npx without manual installation:
npx neuralscaler <input.mp4> --target 4K

# Or install globally and use the shorthand 'ns' command:
npm install -g neuralscaler
ns input.mp4 -o ./output_4k/ -t 4K -q FAITHFUL

# Or invoke directly via batch script:
ns.bat input.mp4 -t 4K

Method B: Standalone Installer (Recommended for Windows)

1. Download the latest NeuralScaler-4K-Setup-v*.exe from Releases.
2. Run the wizard setup to get desktop shortcuts and silent Windows Explorer context menu integration (Right-click -> Super-resolve with NeuralScaler 4K).

Method C: Portable Archive

1. Download and extract the latest NeuralScaler-4K-Portable-windows-x64-v*.zip from Releases.
2. Double-click NeuralScaler.bat to launch the engine and desktop interface (or use ns.bat from terminal).

Method D: Build from Source

# 1. Clone repository
git clone https://github.com/justForever17/NeuralScaler-4K.git
cd NeuralScaler-4K

# 2. Install dependencies & build frontend
npm install
npm run build

# 3. Launch server & app window
python server.py

Sponsor & Support

If you find NeuralScaler 4K valuable for your projects or workflows, consider supporting its continuous maintenance and compute resources!

License

This project is licensed under the MIT License.