Topaz Labs drops Gigapixel 8.3 with 5x faster Recover v2

Anyone who has used generative AI to upscale an old photo knows the uncanny horror of hallucinated faces and gibberish text. Instead of sharpening what was there, the AI just makes things up.

Topaz Labs wants to fix that with Gigapixel 8.3, introducing its new Recover v2 model. The core philosophy here is simple: restore, don’t recreate.

Restoring reality instead of hallucinating

While many modern upscalers act like generative artists sketching over your photo, Recover v2 uses diffusion technology specifically tuned to extract and rebuild true details without creating artificial junk.

Topaz claims the new model is over 5x faster than its predecessor. On an Apple M3 Pro, it delivers a speed boost of over 500% alongside significantly cleaner output with far fewer artifacts.

It is specifically trained on tricky jobs: damaged film prints, low-res scans under 1,000 pixels, fine architectural lines, and text signage that usually turns to mush.

Running it locally still demands at least 8GB of VRAM. For users without a beefy GPU, Topaz offers cloud rendering, noting that the new model has also made cloud render costs significantly cheaper.

UI tweaks and a nod to local purists

The update also fixes several day-to-day workflow headaches. The model selection menu has been reorganized to cut down on endless scrolling.

Zooming into an already processed area in the navigator no longer triggers an annoying re-render, a smart preview feature finally ported over from Topaz Photo AI.

For print jobs, the Gigaprint tool now includes a dedicated crop tool with rotation controls, plus an optimization slider to tune brightness and contrast before output.

Topaz also listened to privacy-conscious power users. Even on standard personal licenses, you can now toggle off cloud rendering entirely in preferences to keep everything strictly on your local machine.

Gigapixel 8.3 is currently discounted by $15 through April 9, bringing a new purchase to $84 and upgrades to $64.

Why it matters

For archivists, photographers, and prepress designers, this is one of the most practical upscaling updates in recent memory. Generative upscale tools are everywhere, but professional workflows demand accuracy rather than creative hallucination.

Topaz’s approach looks very smart: optimizing diffusion to run 500% faster locally gives creators the quality benefits of modern AI without forcing them into slow rendering loops or expensive cloud queues.

Giving non-pro users the ability to completely hide cloud features is another great move that builds trust with the local-first crowd.

The only real catch remains hardware requirements. An 8GB VRAM baseline still excludes older machines, and while Topaz insists it restores rather than invents, diffusion models inherently involve some predictive reconstruction that pros will still need to verify.