Topaz Photo AI v1.4.2


This week introduces a more well defined shortcut structure as well as several fixes. For shortcuts we’ve added and refined some of the interactions around the app. For instance, the classic Ctrl+S/Cmd+S to open the save panel or to start saving when in plugin mode, Enter/Esc interactions for various menus (Cropping, Masking, Exporting), new hotkeys for entering modes (e.g., M for subject masking), etc. All available hotkeys should be shown either in the dropdown menus at the top or in the new Shortcuts panel in Preferences.

Along with these we’ve also added several fixes which are listed in the change log below.

Please give us any feedback or report issues with this release. We’ll be updating TPAI regularly to address those pieces of feedback and issue reports. If there’s a specific image you’d like us to see, you can send it us at this dropbox link.

Released July 20th, 2023
Mac: Download
Windows: Download

Note for MacOS users: The new minimum supported OS version is now 11.0 (Big Sur).

Changes since 1.4.1:

  • Added TensorRT models for Preserve Text
    • Should fix some crashes on Nvidia cards
  • Added and improved shortcut interactions
    • Added a shortcuts panel in Preferences to display shortcuts
      • Can access via Help > Keyboard Shortcuts or the shortcut Ctrl + / on Windows, Cmd + / on Mac
    • Shortcuts are now disabled when the image is loading
  • Preview will now switch to imported image if it’s already imported
  • Minor UI improvements
  • Show a more direct error when importing images with ? in the path on Mac
    • This is due to a limitation with one of our libraries
  • Fixed issue where Z8/Z9 special handling was running on converted Nikon files
  • Fixed width and height based auto-upscaling being off by a few pixels
  • Fixed manually running face detection turning off Sharpen
  • Fixed issue where changing lens correction in plugin mode would close the app
  • Fixed resolution unit metadata being stored incorrectly in some cases
  • Fixed Face Recovery hair and neck selection saying “hair or necks”
  • Fixed issue where text on upscale label would push toggle out of bounds
  • Fixed issue where Face Recovery label would push toggle out of bounds

Same artifacts when enabling “Removing Noise” as in the last Beta. This time i have to enable remove noise manually and the auto pilot had not enabled it by themselve like in the beta:

I uploaded this page to the dropbox.

1 Like

Thanks for reporting this. I was able to reproduce and documented it. The Extreme models seems not to cause this artifact.

Another alternative is to use Upscale at 1x and increase Suppress Noise.


It seems that the initial state of selecting text to preserve has a bug. The Red brush for Selecting has the Black Subtract overlay there as well:

1 Like

Just tested TensorRT text preserve model - it’s INCREDIBLY fast now, compared to v1.4.1 on RTX 3090. Gotta test the stability. :slight_smile:

Ooops, spoke too soon, two minutes pass and the text model decided to balk at the images I’ve thrown at it, and TPAI switched to CPU. I’ll send the log to the dropbox.

[cudaResources.cpp::nvinfer1::ScopedCudaEvent::~ScopedCudaEvent::24] Error Code 1: Cuda Runtime (an illegal memory access was encountered) - this line perseveres in the log. Aha. I don’t have any issues with the GPU though. :slight_smile:

As far as recent versions go, I have first noticed in this version over-sharpening artefacts (white peaking where the image is already sharp), with Standard and Strong sharpening and normal and strong remove noise affected most. Only upscale and extreme noise reduction don’t seem to be affected by this bug.

Edit: after taking a screenshot, the bug does not appear to be showing. Perhaps it’s a previewing bug, I’ll edit post shortly after looking at output :+1: .

Edit: Yes, the final output is flawless and free of the bug :smiling_face: .

1 Like

Note: having exactly 20 16-bit tif’s open, 3rd batch of images (1, 4, then this 20 images), there seems to be a selection bug of the images at the bottom. I can’t click on the first 4 images (the images don’t highlight and don’t display in the main window) but the 5th onward is clickable. To get to the first 4 images, simply clicking on any one the first 4 images then pressing the right arrow key goes to the first image. Hovering over them shows the correct information I’m sure, the first 4 images just aren’t clickable.

before clicking:

after hover/clicking (mouse pointer not visible in screenshot):

Oh! It’s showing “could not load model” when clicking on the first four images, but when right arrow’ing over to them, the models load fine and the image displays with the preview successful :+1: !


Thank you for adding shortcuts comprehensively.
Could you please add Cmd-2, Cmd-3, Cmd-4 for zoom 200%, 300%, 400% respectively
These are present in prior AI programs and are very helpful.


are there any news on CLI and/or batch processing features in near future releases (Roadmap).
As asked that before by me and others (many times), last time even liked by your CEO, see

I’d finally really like to be able to use the product with medium/large batches, which in its current state is not really viable.



I just downloaded and tried Photo AI v1.4.2 (TPAI). I’m quite pleased with the result on a model portrait, with noise and slightly unsharp. TPAI detected and activated Removing Noise and Sharpening Subject. I seem to have resp. Normal and Standard chosen every time. Could that be the consequence of using the same lens?
“Portrait” doesn’t select the main clothes (and sometimes parts of the body), when using Edit in (TPAI) in LrC. It does when using Proccess with (TPAI).
I don’t understand how the feathering works, as it delects the same areas, whether 0, 50 or 100.
Eventually, I would like having TPAI automatically activate “Portrait”, when faces are predominant.
It would be fantastic to have face selection as an option: instead of sharpening the whole subject, it would concentrate on the facial features, especially the eyes, eyebrows and lips. Optionally, the rest of the face and body could have a lower sharpening done.
Best regards.

Feedback- 21.7.2023 Sent an ORF file straight to 1.4.2 via Lightroom, produced some lovely PINK artifacts in the DNG file - thanks Topaz!! Now how about fixing the issue.

I strongly advise you to read the userguides for the app: User Guides | Topaz Photo AI. It will clarify some important questions you might have.

The user guides can also be accessed from within the app when in the start screen.


I have had this issue, and I don’t use LR. I thought it had been resolved in this latest release, but evidently not. I just ran an image through TPAI standalone and got this artefact when running autopilot:

It is caused by the denoising and ratching up the denoise slider just a few notches gets rid of it:

You need to give more details about your hardware and software. I am on an M1 Mac mini 16GB, Ventura 13.5.

Preserving Text seems to always be on Noisy/Compressed by default, but to my personal experience I most of the time need the Low Resolution option!


Could you reinstall and see if that still happens? I believe it’s due to an error in the installation and reinstalling fixes it.

Thanks for sending the log. I passed it to my team so they can work on this. We have a new AI Engine version in progress which may fix this.


The upgrade broke my models, the models directory only has the small files now. Uninstalled and deleted all directories, reinstalled but it didn’t do the long download process for the models, each filename went through immediately.

Clearly it didn’t install the models.

How do I force it?

I sent support an email with the log files.

In which folder are the models stored?
Which OS?

I’ve just installed and tested this new update. The first thing I tested was the “preserve text” filter, as the 2 previous versions had a bug that crashed the software when hovering over the image or moving the intensity slider.
This new version now seems perfectly stable. I tested it for at least 2 minutes (zooming in, zooming out, hovering left, hovering right, hovering up, hovering down, etc.). In short, I didn’t pull any punches to test its stability. I had activated preserve text in addition to upscaling the autopilot, and there were no crashes in what I tested.

So for now, I’d like to congratulate the developers on this bug fix. :slightly_smiling_face:

My system is as follows:

  • Windows 10 Pro 22H2 build 19045.3208
  • Intel(R) Core™ i9-9900K CPU @ 3.60GHz 3.60 GHz
  • Intel® UHD 630 Graphics Core
  • NVIDIA GeForce MSI Ventus RTX 3070 8 GB
  • 32 GB ram.

Me, too, crashing issue on my Nvidia card + preserve text is resolved. Thanks!

1 Like