Topaz Labs releases its Speed Update with NeuroStream 2 and two new image models

Topaz Labs releases its Speed Update with NeuroStream 2, accelerating six image and video models by 2x to 4x while launching Face Recovery 3.

Topaz Labs presents its Speed Update on 𝕏, the third release in three weeks for the developer. The core of the update relies on NeuroStream 2, an inference engine that accelerates six recent image and video models by a factor of 2x to 4x depending on the case. For example, a 10 MB image that previously required four minutes of processing now completes in one minute. The accelerated image models (Wonder 3, Denoise Max, Face Recovery 3, Super Focus 3) run on NVIDIA and AMD GPUs, as well as on Mac, provided the configuration allows it, and the video models Starlight Precise 2.5 and Hyperion 2 gain approximately 20% speed for a cleaner output.

Two new image models accompany the update. Face Recovery 3 aims for a fine restoration of faces (structures, textures, accessories, skin details) while preserving the subject's identity; it is accessible in Topaz Photo, Topaz Gigapixel, and via the API, with local or cloud rendering. Noise-Aware Sharpening isolates noise before applying sharpness, then reintroduces it into the final render to preserve the original grain, a feature intended for wildlife images, portraits, analog film, or archives. The model is available in Topaz Image Web (cloud), Topaz Photo, and via the API.

Mac and AMD users finally benefit from local support for Starlight Precise 2.5 and Hyperion 2, two video upscaling models previously exclusive to NVIDIA, and a Nuke plugin completes the release.