This project provides a command-line interface (CLI) for AI-powered image and video upscaling, targeting users who need a programmable, low-cost, and locally runnable solution. It consolidates various open-source upscaling models, offering a unified interface for batch processing and integration into custom workflows.

How It Works

The upscaler utilizes a polyglot script approach, orchestrating pre-compiled binaries of AI upscaling models. It dynamically indexes available models and their parameters, allowing users to select specific algorithms and scaling factors via CLI arguments. For video processing, it integrates with FFmpeg to handle frame extraction, upscaling, and reassembly, employing a streaming approach for resource management.

Quick Start & Requirements

• Install: Download the latest release zip from releases.
• Prerequisites: FFmpeg and ffprobe (for video), Microsoft Visual C++ Redistributable (Windows). A compatible GPU is strongly recommended for significant speed improvements.
• Setup: Unzip the release package.
• Usage: Run via start.cmd (Windows) or a shell script wrapper (Linux/macOS). Example: ./start.cmd --model ultrasharp --scale 4 --input my-image.jpg.
• Docs: User Manual

Highlighted Details

• CLI-friendly interface for both image and video upscaling.
• Supports local execution, reducing reliance on cloud services.
• Video upscaling handles large files with a focus on resumeable operations.
• Benchmarks provided for various hardware configurations.

Maintenance & Community

Contributors include Joly0 (Windows support), Cory Galyna (Repo/CI), and Jean Shuralyov (Documentation). Contributions are accepted via issue tickets and direct repository cloning/cherry-picking.

Licensing & Compatibility

Licensed under BSD-3-Clause. Compatible with commercial use and closed-source linking as per the upstream licenses.

Limitations & Caveats

macOS support is noted as FAILED due to issues with Intel iGPU and CPU utilization. The binary for macOS is unsigned. Video upscaling currently only supports constant bit rate (CBR) videos. Some AI-generated images may fail due to internal decoder bugs.