Sharpen AI v3.1

Released June 9th, 2021

Windows: Online | Full (Win 10) | Full Legacy (Win 7/8)
Mac: Online | Full


  • Fixed some pixel shifting between the original and preview
  • Fixed errors related to installing plugins
  • Fixed errors opening Sharpen AI on Windows 7 machines
  • Fixed Find Objects button being removed

Released June 3rd, 2021

Windows: Online | Full (Win 10) | Full Legacy (Win 7/8)
Mac: Online | Full


  • Added License Info dialog to Account menu
  • Added clarification to mac installer about installing plugins
  • Improved ETA display


  • Fixed previously proced image appearing in background of fully zoomed out image
  • Fixed a crash occurring when memory usage preference was set to an invalid value
  • Fixed crashes on certain older hardware
  • Fixed sliders leaving a buffer on either end even at their extreme values
  • Fixed mask brush strokes being the wrong size on non-100% zoom, on certain display configurations

Known Issues

  • Find Objects button no longer appears

Released May 26th, 2021

Windows: Online | Full (Win 10) | Full Legacy (Win 7/8)
Mac: Online | Full

Major Changes

  • “Silent mode” is now respected when opening Sharpen from a host app
    • This means it won’t require user input to start applying changes
  • Plugins now work on Apple Silicon (M1) devices running Photoshop (or any other host program) both natively or through Rosetta
  • Updated AI engine
  • Added “Submit Feature Request” under the Help menu
  • Increased accuracy of auto-model detection


  • Fixed processing status not showing when running in plugin mode
  • Fixed various JPEG loading issues
  • Fixed some EXIF data crashes on CR2 files
  • Fixed writing of fractional values in metadata for TIFF/DNG (Windows only)
  • Fixed inaccurate count with file saving panel
  • Fixed file saving panel staying open when opening an image
  • Fixed some color desaturation on DNG saves for some RAW types
  • Fixed Select All being left checked if you double clicked a checkbox quickly
  • Fixed long save paths overflowing the save dialog
  • Fixed crash when zooming in very close
  • Fixed pixelated preview issue
  • Fixed login password field not accepting spaces in middle of passwords
  • Fixed processing time overlapping controls when opening preview after saving
  • Changed some EXIF data so it trims ending spaces
  • Changed how external editor files are saved
    • Should no longer leave an 8kb file if the save fails

Minor Changes

  • Changed buy button so that it shows the login screen as well as opening the website
  • Changed spacing around the app name and version
  • Added login buttons to watermark prompts to hopefully reduce confusion when not signed in

Hi Anthony,

Is this an older version? It says released May 12? Or, was this just released on May 26?


I’m sorry, that was a typo on my end. This version was released today, May 26th.


How do I install Sharpen AI as a plug-in in photoshop, on Mac M1?

Great! Could I get an expedited assist from you to Tech Spt today (5/26), pls?

I just installed this 3.1.0 rel. - it says my license expired and I’ll either get the ginormous watermark or I have to roll back to the 3.0.3 rel. I just installed over.

I just got on the PC to do a bunch of Ps & SAI/DAI work this second.

Is there someone I can call right now (3pm - PDT) to get my acct updated to show I should have an active license (ideally for at least 6+ months…)?

I’m on a data capped plan with slower internet. Upgrading applications is not very fast. In Topaz’s case, it seemingly takes forever. I don’t understand how a sharpening program can be the same size as Photoshop (3.5gb). That just doesn’t make sense. It seems all the Topaz program file sizes are quite large.

Germane to the 3.1 update from 3.0.3, why isn’t there a delta update? At the very least, the download size would be much smaller. Also, I noticed the full installer is 2.5gb, but the in-app version is 3.5 gb. What’s the extra gb for?


I am in the same situation. Topaz models or whatever is using is way larger than it should be in my opinion. if DXO can be 200+ MB for same AI Denosing plus everything else and Adobe Lightroom is 2GB and so is Topaz Denoise, something is strange. Also Topaz releases frequent huge updates which are usually buggy and need another update two days later. The cost / benefit scenario is turning against Topaz for some users. If this keeps up the Topaz programs will be larger than OS. I suggest to Topaz product managers to starting thinking about how to reverse this process before its totally out of control.


Installation was slow (about an hour) but I was running other apps at the same time. Can’t say I am concerned about the size. HD and even SSD space is cheap these days.
From what I have seen so far it’s running fine, although there is still the issue of jittery scrolling in batch view (on a MacBook at any rate), as with other Topaz apps.

I’m having exactly the same problem. Scrolling within the batch list is almost impossible to use. A small change of the scroll wheel results in large jumps of the list.

I’ve installed it but noticed an odd issue. I have it set to use the CPU for processing but whenever it creates previews or processes the main image it is using the GPU. I assume it isn’t supposed to do that?

I upgraded to Sharpen 3.1 last night and noticed several odd / not good changes:

  1. I do automotive racing photography. I often just want to sharpen the lead car or two. I like to mask out the remainder of the field using a Fit View screen. When attempting this with the new release version, used the “subtract” function, but no matter the size of selection area or how many times I would go over a spot, it would not remove portions of the masked image. The only work around was to zoom in to remove the items.
  2. The selection size slider bar changed in how it worked, as though the scale changed. Irritating, but got used to it after a few edits.
  3. Dang this is one slowwwwwww program. It would be nice if it took more advantage of my processor and available memory.

Not all bad news - the auto select for masking is much improved on this version.

1 Like

I second that about the speed. One image has just taken me 5 minutes to process. Before updating
the same size image took about 2.

Is this in CPU or GPU mode? Thanks.

How about the halos in GPU mode? Thanks.


In the new version, vertical banding continues?, very visible when enlarging the image a bit in previous versions!


Good question. It was observed in GPU mode only though, and a similar flaw was present in Denoise AI.
Am very hesitant to replace old well-working versions (which I run in CPU mode for the cost of speed) with these new ones (both Denoise and Sharpen) unless there is such clear advantage as non-inferior GPU mode compared to CPU mode.

Anyone else found a bug with masking (MacOS). Paint a mask and it appears to the left of where you drew it. Does anyone test these releases before they go out?

Yes of course it was beta tested. I’m using OS Big Sur on a 2019 MacBook Pro.
I am afraid I am unable to duplicate your bug, although where the chair and dining table are in this example, goodness only knows!

ivkuzmin - Since I had a similar issue to Marbles999 I thought that I would respond. I’m working CPU mode, running AMD Ryzen 7 3700X processor, SSD hard drives, 128GB memory available. Images took between 2-3 minutes, roughly, in new 3.1 version. I didn’t time them before this version, but the processing time seems about the same - pretty slow.

Thanks. Sounds way too long but obviously depends on the speed of your system, image size, and the SAI mode (sharpen vs stabilize vs focus, or whatever new names were invented in the recent SAI versions).
I’m mainly interested in the IQ in GPU mode.

BTW, I had a sort of response to your claims above but did not interfere as it is not directly related to your problem. Just meant to say that I duplicate layer in Photoshop and apply SAI to the entire layer copy. Then, layer mask in Photoshop, and filter whatever I want with desired opacity. Much more convenient and non-destructive way than applying a mask within SAI, IMHO.