The latest couple or so version of Sharpen AI have started producing vertical stripe artifacts when using the “Stabilize” mode. The artifacts are not visible when previewing the changes prior to exporting, they only appear in the exported image. This happens with both GPU and CPU processing. I haven’t encountered this behavior with older versions.
Please see the images below. Anyone else experiencing this issue?
Yes, samé here, especially visible in portraits with areas of clean skin. And it happens in default sharpening as well.
Please raise a support request at the Topaz Labs main website.
Confirmed here too. I have raised support ticket 193995 and supplied the images I used.
Just remember when you raise a support request it is important to include all of the following along with the input image and output settings:
I recently ran a stabilize today on a TIF exported from Capture One today and saved it back as a TIF, No Compression, ProPhotoRGB after using OpenVINO/CPU processing and no issues. But I must admit I didn’t try using my NVIDIA GTX 1050/4GB because the CPU option is a little faster.
I did provide a screen shot of Sharpen AI’s preferences, a screen shot of Sharpen AI’s Graphics Information which gives RAM details and details of the graphics card. I also provided details of my CPU and full build details of the Windows Pro version. I also detailed the slider settings, and the save settings.
Along with that I uploaded the original jpg file I used, the jpg file exported from Sharen AI, and a comparison shot that showed the original and the exported zoomed to 400%, which clearly showed the vertical lines or stripes.
Hopefully that should be enough for them to go on. 
I have had a response to my support ticket 193995.
It was pointed out that the source photo I had uploaded was better handled by Sharpen AI’s Focus mode, which I can now confirm does indeed produce a much better result on that particular photo, and without the vertical artifacts that Stabilize mode had produced.
Topaz’s Emily Dworkin said “Stabilize mode will produce artifacts when camera shake is not present. In this case it was an issue of missed focus, not a lack of stabilization”.
IMO that is interesting, in that I have yet to update (my v1.3.0) to the current version but I have found when I used Sharpen AI and run all three modes on an image and bearing in mind the intended “function” of Stabilise…it yielded the best looking compared to the other modes and the subjects were not moving ones!
FWIW I was running all three modes on each image regardless of the subject and making my decision on what the output file looked like. Therefore I hope & trust that as the AI models have been ‘improved’(?) that the other modes may yet yield equally appealing output.
PS I just open Sharpen AI to double check my version number and noted when I declined the update(for now 
) a dialogue came up telling me to be sure to update to v 1.4.3 (did I recall that exact version number right) if I want to receive future updates!!! what does this mean if I don’t update yet??
The problem with that is that the vertical artifacts or lines, what ever you want to call them, don’t seem to appear in the preview, but do appear in the saved file. (Both being viewed at the same zoom percentage.
I’ve done some more tests today. All of the images I put through Stabilize mode and saved have exhibited the vertical artifacts, but to be honest, I can’t see it at 100% zoom so I’m not sure how much of a problem it really is. It is apparent at 200% zoom and above, but is almost uniform, giving a sort of an image on canvas effect, which is arguably not all that unattractive.
Ah! thanks for the clarification…so my older version may have have that effect if I zoom in at the 200% zoom. I surmise that is an academic concern at that ‘level’, as surely at normal viewing distances of a print that would not/never be seen.
Having said that, if then processed in Gigapixel AI as I have done maybe a concern depending on the enlargement increase…but then maybe GAI processes such lines out???
This is not true. I did a test and there are still vertical stripes artifacts in the images even with photos with camera shake. Please see attached below. Older versions didn’t create these artifacts even when there was no camera shake seen in the photo. Stabilize mode has produced the best results for me out of all 3 sharpening methods 90% of the time so it’s a real pity that it doesn’t work properly anymore.
I am baffled why these new versions are being rolled out to customers when there are clearly problems with them and they don’t function properly. I am seriously considering downgrading to an older version that didn’t exhibit these problems.
I agree 100% with Svetlina and Greyfox. I find that Stabilize use to look best on a majority of pictures but now it doesn’t. I may look good when you run the preview but the saved picture is not the same.
Remember that Stabilize is for natural shake reduction, typically shake caused by motion blur from handheld images at slower shutter speeds.
Focus is for sharpening out-of-focus items or to correct issues caused by incorrect in-camera focus whereas Sharpen is for general sharpening.
Agreed about the intended use, and I have no issues with the way Focus is working, but unfortunately for me, Stabilize is producing this odd vertical pattern regardless of what it is given, and if, as has been stated, the earlier versions didn’t then that is a pretty significant regression.
I too would seriously consider going back to an earlier version, if I knew where to source it.
If you want an earlier version raise a support request and they should be able to provide a link to download.
Thanks AiDon, have done
I’ve raised a support ticket for an older version too.
I have found the vertical banding to be an issue as well. Using Sharpen AI 1.4.5, the vertical banding only occurs (for me) when utilizing GPU acceleration and the “Stabilize” mode. It doesn’t seem to matter whether the blur is caused by motion or focus issues. It appears to occur regardless of the photo, but only if GPU acceleration is turned on. Turning off GPU acceleration eliminates the problem. Given that the CPU can get it right on the same photos, I’d have to conclude there is something amiss with the GPU accelerated “Stabilize” mode.
Fortunately I have a relatively fast 8-core Ryzen proc, so using CPU mode is not as much of a slow-down as it might be for others, but it is noticeably slower than GPU mode so it would be nice to see this corrected in future iterations, especially for those with slower processors.
The files below demonstrate the issue. Note that these examples are blown up to roughly 300% to emphasize the problem and make it easier to see, but I am able to begin to see the problem at 100%. It is slight at 100%, but it is there. Not being a guru at programming or compression algorithms, it seems somewhat baffling to me what would cause the GPU to create the banding when the CPU doesn’t seem to do so, given that (I would think) both processors are chewing through the same code.
You should be able to use the upload symbol or paste a JPEG OR PNG now.
Thank you! Pics attached to original post to demonstrate the issue.
