Topaz Gigapixel v1.3.1

[Lingyu]

Hello everyone!

Today we have a patch release for Topaz Gigapixel with a large list of fixes. These changes address issues with Wonder 3, Redefine, batch processing behavior, cloud render, and various miscellaneous issues.

We also shipped a NeuroServer update that brings 2x faster processing on Mac, and up to 4x faster processing on low VRAM GPUs (8GB to 12GB) for Windows!

---

v1.3.1
Released June 11th, 2026
Windows: Download
Snapdragon: Download
Apple Silicon Mac: Download
Intel Mac: Download

---

Neuroserver Speed Improvements for Mac and Windows

In this release are many improvements to NeuroServer which speeds up both Mac and Windows processing in different ways. Overall, NeuroServer runs efficiently compared to previous versions, which benefits all NeuroServer-based models: Wonder 2, Wonder 3, and Recover Faces 3.

Mac

On Mac, two things changed. First, NeuroServer initialization and model loading are significantly faster. The first run after installing should start up noticeably quicker, and subsequent runs load faster too. Second, processing now uses both the GPU and Neural Engine together. Combined, Mac processing is almost 2x faster across all NeuroServer models.

Windows

On GPUs that meet the minimum VRAM requirements (8GB and 12GB cards) NeuroServer was running inefficently due to memory usage. We have reduced requirements to allow faster image processing on these devices which were bottlenecked by memory. For 8GB and 12GB VRAM GPUs, the speed improvement is up to 4x.

in our testing with an NVIDIA 4060 Laptop GPU with 8GB VRAM, we upscaled a 24MP image to 96MP (2x upscale) with the following results. Note that devices with more GPU have lower improvement due to no memory bottleneck in processing.

• 1982s (33m) processing time on v1.3.0
• 450s (7.5m) processing time on v1.3.1, a 4x increase

If you’ve been avoiding these models for larger images or batches because of processing time, this update changes that calculation.

---

Changelog

• Update NeuroServer to improve processing speed 2x on Mac and up to 4x on Windows
• Add color correction to High Fidelity 3 to prevent color shift
• Add double-click to maximize window on Mac
• Block generative models on Windows ARM with tooltips explaining they are not supported on that platform
• Fix face recovery processing greyscale images twice
• Fix Redefine Creative model applying Redefine Realistic instead when used on a batch of images
• Fix Redefine model preview not updating when the image prompt is changed
• Fix Standard Max cloud strength tooltip showing 0 or 1 instead of the correct value
• Fix Wonder 3 grain settings not being included in cloud renders
• Fix Wonder 3 enhancement strength not showing correctly in cloud renders after restarting application
• Fix Wonder 3 batch processing using the same strength for all images when different strengths are selected
• Fix Intel Mac color shift when sending images to cloud render
• Fix legacy models appearing in the model selector menu instead of only the model version dropdown
• Fix output image PPI not matching expected resolution for print when using cloud render
• Fix “…” options menu not appearing in the cloud render queue
• Fix out-of-memory crash on images larger than 40MP on some NVIDIA hardware
• Fix Wonder model download not retrying on failure during Mac installation
• Fix “Don’t show again” checkbox not functioning on Neuroserver local render not available dialog
• Fix face recovery help link
• Fix macOS title bar buttons not centering correctly
• Fix Topaz Labs appearing at incorrect size for Photoshop Automate plugin
• Fix crash from log file corruption
• Fix various stability issues from race conditions and memory handling
• Add color correction to High Fidelity 3 to prevent color shift

---

Lingyu Kong
Technical Product Manager
Image AI

[alexandre.topazlabs]

@unrelent
The issue with the halos when using Wonder has been resolved

[Harald.De.Luca]

I tested the new version. And the high-fidelity v3 model is now working well. No more color mismatch issues.

I also tested Recover v3 and noticed it took longer than last time. I thought, “Ah, that’s probably a good sign; it’s now doing a better calculation.” And indeed, it did a great job. No more honeycomb artifacts.

I love it when the models are perfectly functional .

[Gig]

Love the update. Are there any sorts of benchmarks for the different Mac generations? I have considered upgrading to an M5 Pro but it’s hard to know what kind of speeds I’d get.

[alexandre.topazlabs]

We don’t have a benchmark for Topaz Gigapixel. Some users here can share their experience of upgrading computers. It also depends on what computer you have at the moment that you would like to compare with

[plugsnpixels]

Testing upscaling of 1990s 35mm negative scans at 4X cloud (which Photo wasn’t allowing at that size) using Wonder 3. The “befores” at left are not full frame but only a detail view:

[vidmeste]

what app is that out of interest?

[plugsnpixels]

Not sure what you mean, we are in the Gigapixel thread

[Gig]

I thought the team might be able to share something like they did for a 4060 Laptop.

@Lingyu

in our testing with an NVIDIA 4060 Laptop GPU with 8GB VRAM, we upscaled a 24MP image to 96MP (2x upscale) with the following results. Note that devices with more GPU have lower improvement due to no memory bottleneck in processing.

• 1982s (33m) processing time on v1.3.0
• 450s (7.5m) processing time on v1.3.1, a 4x increase

[plugsnpixels]

Trouble with GP as I was doing some work. App hung and sucked RAM when trying to open a second image then close the first.

I was trying to get some good results from a 2010 digital bird image but it wasn’t getting there. Here is the 4X best result (right) compared to ON1 Resize (foreground):

[Harald.De.Luca]

The one with ON1 Resize is still a little better.

[philip.ocampo]

RE: Benchmarks. Perhaps you could share the image that you upscaled, and the settings, and we can see how long it takes our machines to render it?

[alexandre.topazlabs]

I’ll ask the team to see if they can run some tests

[cyph]

so for the mac improvements, what is the better option for AI Processor Auto or M4 Max? confused overall on the difference since one chip, does m4 max option just pick the gpu path?

@Lingyu

On Mac, two things changed. First, NeuroServer initialization and model loading are significantly faster. The first run after installing should start up noticeably quicker, and subsequent runs load faster too. Second, processing now uses both the GPU and Neural Engine together. Combined, Mac processing is almost 2x faster across all NeuroServer models.

[partha.acharjee]

@cyph

so for the mac improvements, what is the better option for AI Processor Auto or M4 Max?

For Mac, it doesn’t matter. Neuroserver will choose the best devices available. This is mainly for Windows machines with multiple GPUs with different capabilities, so users can choose which one to use for that specific session.