Topaz Labs, known for its photo and video editing software, has introduced Project Starlight, an AI model designed to enhance old or degraded footage from home video collections and archives. According to the company, this is the first diffusion model created for such restoration, requiring no manual corrections.
Advanced AI-Powered Restoration
Built with over 6 billion parameters, Project Starlight is powered by NVIDIA accelerators. For comparison, OpenAI’s GPT-4o, released in May 2024, originally had 8 billion parameters and processes text, audio, images, and video. Topaz Labs claims the model “accurately restores detail” while maintaining temporal coherence, ensuring high-quality results without motion artifacts or inconsistencies between frames.
The AI automatically removes noise, deblurs, upscales, and smoothes frames, making it accessible to users without video editing expertise. While existing AI tools perform individual restoration tasks like upscaling, color correction, and frame interpolation, they often require human oversight to manage the entire process effectively.
Users can restore videos up to 10 seconds long for free, while clips up to 5 minutes long require credits and are limited to 1080p resolution, adds NIXsolutions. The enterprise version supports longer, higher-resolution videos. It remains unclear whether Project Starlight will run locally or integrate into other Topaz Labs applications, yet we’ll keep you updated as more details emerge.