Topaz Gigapixel v1.1.0-b0

[esther.huynh (esther.h)]

Welcome to the Open Beta Forum for Topaz Gigapixel!

This beta version is packed full of bug fixes, a few UI changes, and most excitingly, brand new model updates.

We’d love to get your feedback by dropping your thoughts in this survey.

Please feel free to share with us your big wow moments here in our forum post as well.

Thank you!

Esther H.
Associate Product Manager
Topaz Labs

---

Topaz Gigapixel v1.1.0-b0
Windows: [Download]
Mac: [Download]
Snapdragon: [Download]
Released on January 16, 2026

---

Wonder 2: Better than before!

This new and improved version of our latest model, Wonder, gets an uplift. This new precision-focused model aims to restore fine details in faces and textures without the feeling of being “too artificial”. We’ve worked hard to also remove artifacts while also improving performance.

This model will be available for Cloud render only.

Wonder 2 operates as a single step model with no preview controls or exposed model settings. You might see that faces have more natural skin textures and clothing looks less like an AI pattern. Even phone images get more detail from far distances. Check out this example for a phone image taken at Huntington Beach, California:

Wonder 2 isn’t recommended for phone images with low-light with high noise grains. For the sake of testing, Face Recovery toggle and Grain does not work.

---

Introducing Recover 3!

This is our most natural enhancement model to date, designed to produce softer, more realistic results and significantly better detail in fur and feathers for animal images.

For this beta, the model is available in the cloud. There are a few known issues—please review the Known Issues section below for details.

Optimal image size is 1024 × 1024 or smaller. Larger images are supported but may result in slower rendering times. Highly structured, high-grain images are not recommended for this version, and some images may exhibit background artifacts. For testing purposes, the Face Recovery toggle and Grain controls have been disabled.

---

File List Updated

We’ve tweaked the File list to support our generative models displayed. Before, we only showed model settings for our Core models making it easy to review model settings for denoising, debluring, face recovery, and so on.

Our file list now cleanly shows our generative models and includes their model settings (if there is one) and also provides an easy hover access to prompts for Redefine.

---

Cloud Queue Improvements

Alongside several bug fixes related to cloud rendering, we’ve added some quality of life improvements to the cloud render workflow.

Cloud Queue added to Import Page - You can now find the Cloud queue after opening the app. Look for the Cloud queue button above the Recent images section.

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 to cloud
• Added Recover 3 to cloud
• File list improved for generative models
• Added Cloud queue button at import page
• Cloud queue download all button added
• Added error state for Tiff files using the Cloud Compare button
• WHoverTooltip sizing based on new designs
• Fix HEIC not loading, add test for OIIO missing format support
• Fix plugin dialog not showing up
• Automatic Lensfun Update
• Cloud fixes and improvements
• Fix potential crash in model downloader
• Fix cloud sizes and settings not being shown on refresh

---

Known Issues:

• Crop is incorrect in output with Wonder 2 or Recover 3
• Face Recovery toggle does not work on Wonder 2 or Recover 3
• Grain toggle does not work on Wonder 2 or Recover 3
• Preview entire image on Recover 3 does not work
• File list shows “Creativity strength” when it should be just “Strength”

[jo.vo]

Hmm, „cloud only“, „Cloud only“ „only cloud for now“ - seems like I have to skip this one :-/

[bas.evers]

Hi all,

Here’s me again. The julia fan.

Although (stand alone) version 7.1.3 was my al time favourite, I have to admit Topaz made some serious progress. I tested the Wonder 2-preset and the Recovery 3-preset. As much as possible with the same settings. Although in 7.1.3 the 0-100 detail setting might slightly different from the nowadays setting 0-100 “strength“..? For face recovery I used 25 for all images – and for fair comparison.

Here are the results:

Wonder V2: Face recovery 25

The High res image.

There is much detail in the skin. Maybe even a little too much. Can it be? Nothing that can’t be solved in Photoshop. Well done Topaz!

Recovery V3: Strength 100 (maybe too much?) Face recovery 25

The High res image.

Better result than my all time favorite 7.1.3, but it has a few small “hickups”. The birthmark under her right eye is a little too much. And there are a couple of new birthmarks. Where did those come from? Nevermind, it can be fixed.

Overall there is much more realistic detail in both results. Now you can see pores and skin lines. So congratiolations Topaz. I hope you will be as good as Magnific some day (which PiXimperfect tested). Keep up the good work!

Bas

[bas.evers]

And I almost forgot to test the one thing I was pretty critical about in the past. Now Topaz managed to improve the level of detail in the clothes as well. Although I have different preferences for the outcomes in all of the results (trousers and shoes: Wonder V2, sweater: Recovery V3), you can always combine these outcomes of course. Yes, that would mean more work, but sometimes that’s worth the effort…

[bas.evers]

And one final comparison for today: Recovery V3 with strength 100 (left) and strength 50.

The latter one is more subtle but gives minor strange new details that weren’t there before (a white spot one her forehead and some weird marks on the left of her pink). It probably recuires some more testing to find an optimal result that can be further processed in Photoshop.

Have a good day all!

[andymagee-52287]

They said face recovery is not turned on yet for Wonder 2 and Recovery 3, so how did you get it to work?

[bas.evers]

Uh… just turn it on, I suppose.

[devenpsc]

Wonder 2 is not working for me. The cloud keeps failing the render.

Also, what’s with the cloud only options. Topaz does realize professionals can’t upload private client work to the cloud, right?

So I’m trying to see how Recover 3 works, but there’s no local rendering so I’m forced to use cloud, but oh gee, I’m stuck in a que and still waiting minutes later. So much for fast online rendering. LOL!

Still queued….

Still queued….

Four minutes later, still…

Maybe we can get up to ten minutes waiting in que. Still waiting… With my RTX 3090 I would have had this new model tested already locally.

And still waiting… I’m shutting this down. I have better things to do.

[andymagee-52287]

Original

Wonder

Wonder 2

Wonder 1 adds some moire patterns over the bird.

Wonder 2 is better.

Recover 1

Recover 2

Recover 3

I’d forgotten how slow Recover 1 was.

I like Recover 2, except for the sharpening halos.

Recover 3 created a bunch of artifacts in the sky.

[bas.evers]

Why..? It’s fast.

[TPX (Thomas K.)]

I think my image is too big - so no tesing.

And cloud only realy scares me.

---

---

It did display this only after chooseing hifi.

[plugsnpixels]

My first attempt with a random iPhone photo, since the description above says “Even phone images get more detail from far distances”…

Description also says, “Optimal image size is 1024 × 1024 or smaller.” So that means pretty much every image needs to be cropped or otherwise prepped before processing.

So I went into PS and cropped/saved a smaller section out of my digital zoom iPhone image, and it’s still bigger than the recommended size. Here is the full frame (reduced):

The unprocessed crop (other than levels):

The successfully processed result at 2X with Wonder 2:

Wonder 1:

Standard Max:

Recover v3:

Redefine Realistic:

Redefine Creative:

Local Low-res:

Local High fidelity:

None of these are what I would consider traditional Gigapixel restorative quality…

ON1 Resize didn’t do any better, interesting…

Back to the Gigapixel Beta, cranking Redefine Creativity to Max, Texture = 3 – THIS is what I was expecting all along (except for the artifacts):

Again but at 6X:

I asked before, but WE NEED A WAY TO CLEAR OUT OUR CLOUD QUEUES. After rendering these tests, I don’t want to see them clogging up the list.

Update: Bloom Realism with Max Creativity, 6X:

[andymagee-52287]

This was the best fidelity relative to the original I could get. Redefine Creative Medium with the prompt “Sun setting through Southern California Smog” at 1X on a slightly different crop.

[plugsnpixels]

Thanks, I just finished my “artsy” version…

[TPX (Thomas K.)]

@esther.huynh

This model will be available for Cloud render only.

Will wonder 2 make it into local rendering?

[jo.vo]

Hmm, I’d say that only the Bloom rendering is acceptable here.

This is Nano Banana btw.

[alanhusband-677420]

Still can’t permanently clear the cloud queue of old renders that are no longer available for download. They keep returning to the queue when refreshing the queue or restarting the app.

[bas.evers]

Another thing I would like to address is that slower steps might give different (interesting) results. I did a 100% Wonder V2 scale to the original image. After that I upscaled 2x – and reused the upscaled image to proceed – until I reached the 6x upscale. Wasn’t too fond of that result, but I started over with a 100% scale (Wonder V2) and after that I upscaled that particular image to 600%, like I always do with this image. This time I had some interesting results – partly. See for yourself.

Left is the original image upscaled too 600%, on the right: first upscaled 100% and thát result upscaled to 600%, which gave me more details…

[bas.evers]

Here is another example of upscaling in multiple steps (Wonder V2). See my earlier post.

The original.

Upscaled 100% (so no actual upscaling there…)

Left 600% of the original, right 600% of the 100% upscale – all upscaled with Wonder V2.

The right version is not entirely acurate everywhere but some elements have improved better than direct upscaling. It might be worth looking into if you have a real difficult (lores) image to upscale…