Gigapixel 7.0.4


This week we have a big change in regards to RAW handling, and another batch of optimizations which should make batch processing feel even faster. More detail about these two below, but if you just want to see the full change log it should be below the download links.

For RAW files, we have decided to move away from DNG exports so we can better support color accuracy, consistent results, and support optimizations that are not possible while also supporting DNG exports. The benefits of this for RAW inputs should include:

  • Preview times should be drastically reduced
  • Memory and CPU usage should be lower
  • Exports should match preview in all cases

This does mean if you are using RAW in your workflow, you will need to output to a different format. By default, the “preserve” option on file format will default any RAW file to TIFF.

In optimization news, importing, selection, and general batch operations should be on average much faster than 7.0.3. Here are some personal benchmarks from my development machine:

Importing 100,000 images:

  • 7.0.3: about 2 minutes 20 seconds
  • 7.0.4: about 11 seconds

We will continue trying to keep optimizations on this level in mind in future development so that batching large inputs is possible.

As a note, all pre-7.0.0 versions of Gigapixel will upgrade through 7.0.0. This means if you’re on 6.3.3, the in-app upgrade flow will take you to 7.0.0, and then you can see 7.0.4 from that version. You may optionally also download the installers directly from this thread. They should automatically uninstall Gigapixel 6 for you before installing the updated version.

Released March 13th, 2024
Mac: Download
Windows: Download

Change log:

  • Reworked RAW flow to support more optimizations
    • DNG export is no longer supported
  • Optimized image loading and selection flows
    • Should drastically speed up time to load images
    • Personal benchmark loading 100,000 images:
      • 7.0.3 - 2 minutes 20 seconds
      • 7.0.4 - 11 seconds
  • Optimized exports with alpha channel in some cases
    • If alpha is all the same value, it will no longer process via model
  • Added native right click menus for Mac
  • Fixed issue where crop values weren’t rounded correctly
  • Fixed some preferences page’s reset button behavior
  • Fixed switching scale modes in crop causing preview issues
  • Fixed file list resizing when images are added/removed
  • Fixed file names sticking in the open file dialog
  • Fixed scale being allowed to go below minimum value (0.2x)

Most issues and bugs were Fixed,
Now I report an issue that mentioned in 7.03,name it “100%” issue:
Load a new image, should zoom 100% preview and preview the “center”.
Now,only the first image is 100% and Center every started the programe ,so this issue not occur in Photoshop . 7.00, 7.01 no this issue.
BTW, Paste from Clipboard is an useful method. Menu >>Edit >>Paste from Clipboard

1 Like

@adam.mains If you guys are going to “promote” batch processing, you really need to add a timer to these things that shows some kind of time intervals for time remaining… I just loaded 317 67M NEF files and they loaded fairly quickly… That 67M NEF file translates to a 3.17G TIF file and about about 6 or 7 minutes to process and another 2 minutes to write the TIF file… Needless to say, I ain’t waiting for GIGA to finish that job without some clue as to when it might finish… :sweat_smile:


I am aware of a couple of zoom related bugs but we are waiting for some other updates before touching those due to potential conflicts. It is in our issue tracker and will be addressed eventually but I cannot give you a timeframe, sorry.

A timer is something I want to add, but due to the nature of estimation it may not be super accurate, especially for batches with different file sizes and scales, or different amounts of faces, etc.

The current optimizations for batch are mainly so that people can reasonably import images without excess memory usage, freezing for long periods, or crashing. For instance, my largest test was 1 million images imported (took ~45 seconds), but I don’t think it’s reasonable to actually output that many currently. There will likely be a separate optimization pass for outputs at a later date.

Also I’m curious, how high are you upscaling these NEFs to?

1 Like

Huge bug to install this new version. Impossible to install. It asks me this: “Your processor must support AVX instructions to install this program.” But I have AVX. My processor is an Intel I9 9900K.
Why can’t I install it when I know how to install Topaz Video AI 4.x.x, which also requires AVX?
It’s really absurd to be stuck like this for a bug like this.
Knowing that I have version 7.0.3 installed. So why don’t I want 7.0.4?

On this page, you can see the characteristics of my CPU. And it does have the AVX2
Intel Core i99900K Processor 16M Cache up to 5.00 GHz Caractéristiques techniques

And here’s a screenshot of my CPU in the device manager.

And in the CPU-Z, I confirm that I have the AVX and AVX2.

1 Like

They were scaled to 4.3something by Auto pilot. They’re a serious of images that were captured one night as star trails… of course, noise was really the only issue that needed to be corrected… and didn’t need to be scaled, I just let Giga do it’s thing to test the import process, and test some minor performance issues… I never “batch” process images as a general rule except in the above instance and I’d never use GIGa to “batch” process images just for Noise.

But it sure highlighted the need for some type of process completion calculator…

Thanks for the response Adam. I’ll be Zooming with Lingyu later today on another issue.


Although I’m a big fan of this software, I’m not sure that moving away from DNG format is a good idea. Specialy when it comes to removing noise, the DNG format often give better results than tiff format.

So, please bring it back.


So much this! I also For the utmost quality Pictures really would love the program to preserve the image format and continue exporting in dng should still be an available option.

I truly appreciate the efforts of the Topaz developers. The loading speed has indeed been drastically improved. However, I do have one concern. During the beta testing phase, we had the opportunity to test the GHQ 192 model version. Now, when I modify the ghq v1 options file to load the model from one of the previous beta versions, I achieve significantly better results in this case with ghq v1 compared to the regular ghq v2 and even the standard model. I would like to see newer versions deliver not only faster loading times but also improved models, rather than compressed and reduced quality ones. So that it doesn’t seem like I’m talking “in vain”, here are a few examples.

Yeah also for me I noticed HQv1 retains more details then HQv2.
I use HQv1 on all my photos. never bothered with v2, as it was just not that good. to much details are lost.
The real competition to HQv1 in v7.x, is the HQ AI in v6.3.3.
I noticed that the HQ in v6.3.3 retains bit more details and it sharpens better the image then HQv1 in v7.x. I have yet to decide which one I like better , the HQ in v6.3.3 or the HQv1 of 7.x…
I use Sandboxie-Plus to run the v6.3.3 in a Sandbox. that way I can have both versions installed on the same computer without them conflicting each other.

You should give it a shot and tell me which one you prefer… HQ in 6.3.3 or HQv1 in v7.x

I also prefer the results from 6.3.3. Unfortunately, GPx 6.3.3 compresses to 95% JPG quality. Regarding the model itself: you might have misunderstood me. I’ve edited the model’s .json file (ghq v1) to avoid using the 96x96 model ( …) and use “ghqv2-v1-fp16-192x192-2x-ox.tz2” (the older Gigapixel doesn’t support .tz2 models, but you can try 2x and 4x adequates .tz). The 192px models were in final beta testing and it’s a shame they weren’t kept. They show its best potential when upscaling a photo with good denoising applied beforehand, or even better, with a higher quality input image. Thanks for the Sandboxie suggestion. I’ll try running 6.3.3 alongside 7.0.4 in it. Regards.

1 Like

how do you edit those? they are binary files.

you can use a text editor, like notepad++. I wouldn’t dwell on this any further, it might be against the forum rules. greetings

6.10 standard is a GOOD MODEL
this image in G7 all models have a terrible result(horse skin)

Hi Fidelity v2 produced very good result. Mac Studio Ultra, Sonoma 14.3.1

Windows 10 ,I run 6.1 and 7.04 at the same time, compare all models,I can’t say v6.1 is better than 7.04,but sometime ,V6.1 this model is a Good Model.

Your result is not much, but still better. PAI has that problem, from time to time. More than a year ego, I posted a picture of mountains, where trees had “baldie” spots, and compared that with GP result. It shows on people clothing as well. PAI can’t reproduce texture as good as any version of GP.

My examples with Gigapixel 7.0.4 on an M2 Mac Mini show different sections of a recent iPhone 12 image, performing 6X enlargement on auto-pilot:

Please try reduce the denoise strength of the model in gp 7. New model has more sensitive control of sliders, but the auto slider selection can sometimes be inaccurate or not to the user’s taste.