Topaz Labs drops faster 4K video models

If you have generated video with the latest AI tools lately, you know the telltale signs: skin that looks like polished plastic, textile textures that blur into smudges, and garbled background text that instantly ruins the illusion.

Topaz Labs wants to be the tool that cleans all that up. On March 31, the company dropped its Precision Update, introducing several new AI model families across Topaz Video and its Astra cloud platform. The headline upgrades deliver up to 4K enhancement both locally and in the cloud, with processing speeds jumping up to 5x faster than previous iterations.

Taming plastic skin and hungry GPUs

Instead of offering a one-size-fits-all button, Topaz split this release into targeted engines designed for very specific headaches.

The standout for creators working with GenAI footage is Starlight Precise 2.5. It focuses on taking clips from the newest generative video models up to 4K while actively removing the artificial, waxy sheen that plagues synthetic characters. It restores natural facial features, fabric weaves, and legible text, though it is currently locked to the Astra cloud platform due to compute demands.

For power users who refuse to leave their local desktop workflow, there are two new diffusion models inside Topaz Video Pro. Starlight Fast 2 cuts processing times down dramatically, running 3 to 5 times faster than Starlight Mini. The catch is hardware: you will need an NVIDIA GPU with at least 16GB of VRAM, and Topaz recommends a whopping 32GB. Cloud export remains an alternative if your rig cannot handle it.

The second local option is Starlight HQ, which finally lifts the 1080p input restriction found on older models like Starlight Mini and Starlight Sharp. It accepts higher-resolution footage to balance sharpness and micro-contrast without altering the original creative intent. This one is somewhat gentler on hardware, asking for a 10GB minimum NVIDIA card and recommending 12GB.

Traditional animators and motion graphics artists were not left out either. The new Gaia 2 model brings a restructured frame-by-frame architecture to prevent detail bleeding during 2x upscaling, specifically targeting aliased edges in 2D animation and stop-motion. Best of all, it runs locally on both Windows and Mac.

Rounding out the release, Topaz Photo gained True Resolution Detection to spot images that were upscaled or resampled without real pixel data, while Topaz Video added native vertical and horizontal mirroring alongside a modular installer to stop models from eating your entire hard drive.

Why it matters

Topaz is carving out a lucrative role as the industry's digital dry cleaner.

While the rest of the tech world races to build the next foundation video generator, raw outputs from tools like Gen-3 or Sora still struggle with uncanny valleys and resolution limits. By tuning Starlight Precise 2.5 specifically against synthetic artifacts, Topaz is giving production houses a realistic bridge to make generative clips usable in broadcast-ready 4K projects.

That said, the hardware barrier here is brutal. Asking local users for 32GB of VRAM to get the best performance out of Starlight Fast 2 keeps high-end rendering in workstation territory, effectively pushing everyday freelancers toward Astra's cloud billing.

We think Topaz made a sharp strategic move by leaning hard into diffusion-based reconstruction rather than simple edge sharpening. It turns video enhancement from a mechanical filter into an intelligent finishing pass, ensuring that even as generative video floods the market, Topaz remains indispensable in the delivery pipeline.