Topaz adds local AMD GPU support to Wonder 2 via NeuroStream

• Topaz says Wonder 2 now runs locally on AMD GPUs, extending the local path it introduced with NeuroStream beyond NVIDIA hardware Topaz announcement.
• The model is positioned as a no-tune enhancer for sharpening, denoising, and upscaling low resolution images, with Topaz showing face and eye crops as examples Topaz announcement Comparison examples.
• Topaz says Wonder 2 is already available across Topaz Photo, Topaz Gigapixel, Bloom, and the API Availability thread.
• In its March Precision Update, Topaz also said more image and video models will be optimized for AMD platforms with NeuroStream built in, which suggests this was a platform move, not a one-off port Topaz announcement.

Topaz's own links are the useful ones here: the original NeuroStream launch post framed local execution as a VRAM optimization play, the Precision Update quietly added AMD support for Wonder 2, and the AMD partner page leans on local acceleration for denoising and preview workloads. The API docs make Wonder 2 look less like a generic upscaler and more like a targeted detail-recovery model for hair, skin, fur, and feathers.

The announcement tweet is short, but the attached quote card says more: Topaz argues that larger image and video models have become hard to run on local hardware, and that NeuroStream plus AMD GPUs is how it plans to bring more of that work back to the desktop. That lines up with the earlier NeuroStream launch, where Topaz described it as proprietary VRAM optimization for running larger models locally.

AMD's own Topaz partner page focuses on AI offload for photo enhancement, including denoising previews on Ryzen AI systems. The new part is that Wonder 2 now joins that local AMD path.

Topaz says Wonder 2 can now be found in four places:

• Topaz Photo
• Topaz Gigapixel
• Bloom
• The Topaz API docs

That matters because it spreads the same model across desktop editing, creative upscaling, and programmatic workflows. For a creator, this is less a single app update than a shared enhancement model showing up across Topaz's stack.

Topaz's API docs describe Wonder 2 as a precision sharpening model for high resolution organic textures, specifically calling out hair, skin, fur, and feathers. The older Wonder docs describe the broader family as automatic repair for small or medium low resolution images with softness, noise, or compression artifacts.

That matches the examples Topaz chose for the AMD launch: tight crops of a face, a cat's eye, and a portrait with glasses. The pitch is simple, a single model that cleans up messy low-res detail without asking the user to stack separate sharpen, denoise, and upscale steps.