Topaz Labs drops Project Starlight for video restoration

Topaz Labs just introduced Project Starlight, calling it the first diffusion model designed specifically for video restoration.

The tool is built to take old, low-quality video files and upscale them to high resolution. According to the team, Starlight represents their biggest technological leap since the original launch of Video AI. They are currently offering early access signups through their social channels.

Bringing diffusion to video repair

Most video diffusion models today are trained to generate scenes from thin air, turning text prompts into cinematic clips.

Project Starlight points that underlying tech in a different direction. Instead of creating footage from scratch, it leverages diffusion architectures to intelligently reconstruct missing details in compressed, blurry, or vintage footage.

Topaz claims the approach allows for a level of visual detail that traditional video upscaling algorithms simply couldn't produce.

Why it matters

For video editors, archivists, and anyone looking to restore vintage home movies, conventional upscaling often leaves a waxy, over-smoothed mess. If diffusion can hallucinate plausible high-frequency details while maintaining temporal consistency, it could completely modernize video preservation workflows.

We think this is a smart pivot. While the rest of the AI video ecosystem is racing toward text-to-video generation, Topaz is applying cutting-edge models to a very specific, high-demand utility problem with an existing paying audience.

The real test will be compute. Diffusion models are notoriously resource-heavy. Whether Project Starlight can run reasonably well on local creative rigs without taking hours to render a few seconds of video remains to be seen.