Topaz Gigapixel v1.1.0 - v1.1.1

Hello!

Today we’re excited to release Topaz Gigapixel v1.1.0, an update focused on expanding our generative model lineup, refining the Cloud rendering workflow, and delivering a number of important fixes and quality-of-life improvements.

Thanks again for being part of the Topaz Gigapixel community :blue_heart:

As always, we’d love to hear your feedback.

Esther H.
Topaz Labs


Topaz Gigapixel v1.1.1
Windows: [Download]
Mac: [Download]
Snapdragon: [Download]
Released on January 23, 2026


Topaz Gigapixel v1.1.0
Windows: [Download]
Mac: [Download]
Snapdragon: [Download]
Released on January 22, 2026


Wonder 2 — Precision Detail Without the “AI Look” (Cloud Only)

Wonder 2 is an evolution of the original Wonder model, redesigned to focus on precision and realism—especially in faces, skin textures, and fine surface detail.

The goal with Wonder 2 is to aim for:

  • Restore fine facial and texture detail
  • Recovers text, logos, and signage
  • Reduce artificial-looking patterns
  • Minimize generative artifacts
  • Improve overall quality compared to Wonder 1

Key things to know

  • Cloud render only (for now)
  • Single-step model with no preview controls
  • Best results when upscaling
  • Not recommended for low-light phone images with heavy noise

Wonder 1 vs Wonder 2 results & false resolution
If you’ve experienced getting good results with Wonder 1 but not with Wonder 2, there is a chance your image may have false resolution. This is common with images from scanners, screen captures, or online sources. False resolution happens when an image is upscaled—its pixel dimensions are increased without adding real detail. The image looks large, but the extra pixels are interpolated and don’t contain true information.

You can spot false resolution by viewing the image at 100% (actual size). If you see compression blocks, pixel stepping, or artifacts in areas that should be smooth, the image likely has false resolution.

Workaround:
Try downscaling the image first. Downscaling can be done in a different image editor. Downscaling removes excess fake pixels and condenses the real detail, which often leads to clearer results when processing the image in Topaz Gigapixel with Wonder 2.


Recover 3 — Our Most Natural Enhancement Model Yet

Recover 3 is built for users who want soft, realistic enhancement rather than aggressive detail reconstruction. It shines especially with:

  • Fur and feathers
  • Natural textures that benefit from subtle recovery

Unlike Wonder 2, Recover 3 is designed to be a general-purpose enhancement model rather than a precision generative tool.

Availability

  • Windows: Local + Cloud
  • macOS: Cloud only (local support coming later)

How to Use

  1. Select the Recover model
  2. Selec the version from the dropdown menu and select “Recover 3”
  3. Choose a strength mode (Low/Med/High) - the app defaults the Medium option as the best balanced enhancement. The strength mode controls the intensity of the detail enhanced.

Recommended usage

  • Optimal input size: 1024 × 1024 or smaller
  • Larger images are supported, but renders more slowly
  • Not recommended for highly structured or high-grain images

Updated File List

We’ve updated the File List to better reflect generative workflows.

You’ll now see:

  • Which generative model was used
  • Any associated model settings
  • Quick hover access to Redefine prompts

This makes it much easier to review how an image was processed—especially when mixing Core and Generative models.


Cloud Queue — Small Workflow Improvements

In this release, we focused on workflow refinements and stability improvements for Cloud rendering—not a full overhaul.

What’s new

  • Cloud Queue access directly from the Import page
  • Download All button for completed Cloud renders
  • Improved error handling and status updates

These changes are meant to reduce friction and make Cloud renders easier to manage, while we continue improving reliability behind the scenes.

Download all - Access your downloads with one single button instead of manually downloading each image. The app will take any “Ready for download” status images and download them in the background.


Changelog:

  • Added Wonder 2 (Cloud only)
  • Enabled Recover 3 local rendering on Windows
  • Added "Download All” added to Cloud Queue
  • Added Cloud Queue button to Import page
  • Updated Image AI
  • File List updated for generative models
  • Updated error handling for TIFF files in Cloud Compare tool
  • Fixed HEIC loading issues in some cases
  • Fixed plugin dialog not appearing
  • Fixed potential crashes in model downloader
  • Fixed longest edge preferences not sticking
  • Fixed Cloud queue naming and refresh issues
  • Fixed cloud thumbnails in most cases
  • Fixed cloud filenames being incorrect on upload
  • Fixed cloud uploads not being deleted in some cases
  • Fixed cloud errors displaying wrong model name<
  • New Features dialog restored with updated slides
  • Cloud rendering stability improvements
  • Automatic Lensfun updates
  • Prevented source file deletion in rare Cloud cases
  • (v1.1.1) Fixed cloud uploads sometimes deleting files incorrectly
  • (v1.1.1) Removed neural engine toggle on legacy models on Mac

Known Issues:

  • Recover 3 local rendering is not yet available on macOS**
  • Face Recovery disabled for Wonder 2
  • Recover 3 does not support Face Recovery on Cloud
  • Tooltip not showing for Wonder 2 & Recover 3 when large images over 128 megapixels may fail to process in the cloud
  • Tooltip not showing for Wonder when an unsupported scale (over 6×) may fail to process in the cloud
7 Likes

Wonder2 isn’t working correctly. The scaling / ratio is the weirdest I’ve seen. Seems to chop loads of the image away, example below between them, scaling the image to 10000px on the longest edge,

2 Likes

You might be experiencing this cloud bug with decimal scale factors:

2 Likes

no idea to be honest, although cloud-wonder worked perfectly and cloud-wonder-2 chopped it. :slight_smile:

1 Like

I get the same thing with wonder and standard max, crops the image after processing

2 Likes

@richardT @jamie-6940

I added you to the bug thread. I can confirm that this is the issue you have

1 Like

A photo enhancement model that is cloud only is an interesting choice. Is it really that resource intensive or is this just a way to get people to consume cloud credits? Are there plans to make this able to run locally?

There are no cloud credits in Topaz Gigapixel, it’s free. The model was not ready to work locally, it will come later.

4 Likes

I’m noticing very ‘artsy’ approach of the Recover v3 model on all three ‘strengths’, is it intended? :slight_smile:

It’s in T-Gigapixel AND also in T-Photo, in GPU mode.

..whereas, Recover v2:

…and v1:

2 Likes

Those are good results for Recover. If you want more details on this image you will need to use Redefine. This is going to generate details that are not there in the image. Keep in mind that this image is 192x192 pixels, lower than the resolution of the first Nintendo.

For the difference between the models, Recover v3 is more about fidelity and less about adding details that aren’t there. The other models will always be there so you have the choice :slight_smile:

4 Likes

Yup, that’s why I used this thumbnail for testing almost since the birth of Gigapixel. :slight_smile:

Recover v3 is also very smeary on low resolution (600x800) camera-taken grayscale text page from MMS message, so I had to use Wonder v2 for it to look great.

Roget that - Recover v3 isn’t for the lowres animated stuff. :saluting_face: :slight_smile: And uploading this thumbnail to Wonder v2 cloud yielded an error during processing.

Now it’s the time to take on old Windows XP wallpapers!

it’s worth a try! But the new models were requested for more realistic images like low-res photos or heavily cropped animals in a photo, etc. Otherwise there was too many artifacts. The more we see users testing the models, the more we can add an accurate description of what they are meant for in our release, this is actually good to see this test.

2 Likes

Trying to upscale a meme, photo-like image this time, with Recover v3:

Can you send us that image? I would like to test and share with the development team.

1 Like

Of course!

A post was merged into an existing topic: Upscale Issue | Undesired Export Crop

Another attempt on Recovery v3 (a raunchy tweet of a fundamentalist). :slight_smile:

Clearly, there’s some issue with this cell-like texture.

I’ll have to check again—if that’s really true.


Wonder 2 - Pixel Dense image as input - 3x upscale



Wonder 2 Native res as input - 2x upscale



Hifi V2 - Pixel Dense image as input - 3x upscale



Standard MAX - Pixel Dense image as input - 3x upscale

1 Like

@mikmod1

Here are the results of my tests:

Wonder V2 gave me the best results:

Then Recover V2:

Recover v3 didn’t work on this image. I shared with the development team.

3 Likes

Thank you! Same result on my RTX 3090 with Recover v2, so we know now, there’s something wrong with Recover v3 output, possibly… :slight_smile:

Wonder v2 works wonders indeed with this image - I had to see to believe it myself. There’s a bug though - if the amount of upscaling exceeds 6x, the cloud processing errors out.