Sharpen AI v3.1

There was a 2.4.2 for Denoise but none for Sharpen.

“Sharpen AI 3.1 does a worse job on removing motion blur than version 1.9 (the 1’st version I used) !
2.2.2 was better and 2.4.2 was the best.” - raw_edge00.
So there was 2.2.4 , yes?
So is it worth upgrading to 2.2.4 having 2.2.2?

So there was mistake in the message.

Are old versions available for download? I need to get in the habit of downloading updates manually (rather than letting the program download and install them for me) and tucking them away in case I need to go back to an earlier version.

If you type Topaz AI changelog you will find all programms and some version:

My first impressions of Sharpen 3.1.1:

  • The object detection is missing
  • Overall performance (whether using CPU or GPU) seems noticeably slower than even 3.1

Yes, I found 2.4.2 for Denoise, not Sharpen.

Yes, as you said, it’s in the list.

I couldn’t possibly say as I’m a happy 3.1.1 user. Only you can answer that.

1 Like

Ok. A change in numbers.

I am just afraid after reading a lot of posts that 3.x.x versions having just have questionable output quality I shall have worse output quality. I do not mean speed.

Perhaps someone who has the current version, but who wants to go back to an earlier one, could process an image in both and show the result here as a side by side comparison.

1 Like

I have tested a number of images using 2.2.4 as the gold standard against 3.1.1. (A pain to reinstall, uninstall and the reinstall. Wish we could use two versions) I saved 4 pairs which is 5 Gig of files. My opinion is like others. 2.2.4 is very good and slow while 3.1.1 has a number of problems, all of which I have seen noted on this forum. My question. Why don’t the developers do these A/B comparisons themselves before they release the updates. The differences are obvious even at 100% on a good monitor.

For all the Topaz products - older versions are here:

The older Topaz product releases are here:

2.2.4 was very good. The models could have done with more training - small (distant) human faces and/or text sometimes got hideously distorted, especially in slightly noisy images and using the auto settings.

Versions 3+ are just too unpredictable for my liking. Contrasty diagonals get halos on their upper edges. Strange random patches of low sharpening appear. Etc. etc.

Presumably speeding up of the program was the main goal in v3 and corners had to be cut …


I’ve largely avoided this halo discussion because I’ve never seen any at all on my images (though of course I can force them to appear if I over sharpen) but I’m now going to throw a grenade into the discussion (and then beat a hasty retreat!). I felt there must be something to the subject given the number of people complaining about it, so I picked an image which I thought might be prone to haloing and put it through Sharpen AI v3.1.1 in standalone mode. When viewing the output TIFF file I get no halos in Faststone Image Viewer (which is where I normally view my output files), FastRaw Viewer, Affinity Photo, ON1 2021, Capture One or Luminar AI. But boy do I get them when viewing in Photoshop or Lightroom, along with strong colour changes.

Furthermore, I reinstalled v 2.2.4 and processed the same image with the same settings. Comparing them in Faststone I can’t see a scrap of difference. Stacking them in layers in Photoshop shows 2.2.4 as being a clear winner, but doing the same in Affinity Photo shows 3.1.1 as being much better.

What all this means I have no idea - you pays your money and you takes your choice I guess.


Interesting! :bomb:

1 Like

I just want to show an image I’ve used before that was processed as a RAW file in SAI 3.1.1. Some additional work was done in Affinity Photo (Saturation and Levels). The same RAW image was developed in Affinity but not sharpened. You can compare the difference. The camera focused on the boy so the drummer was out of focus but both greatly improved. These are clips at 32%. One had a optics correction but not sure which one.


Sharpened in SAI 3.1.1:

Which mode was used?


1 Like