After selecting and fine-tuning the edges of a distinct subject (antlers) a portion of the subject is not processed the same way as the rest of the subject. A distinct line of demarcation is present in the saved image. Other than opening the original (scanned image from film) in the app, the screen capture clearly shows the issue in the right-hand portion of the edit screen.
Correct, Subject Only was initially selected, however the mask was not accurate enough to exactly mask the subject completely, despite the fact that tonally, and with very distinct edges, additional editing of the subject was needed to make the mask accurate using brush tools.
Is it safe to assume that shifting back between Subject Only and Edit Subject may be responsible for the visibly incomplete sharpening effect?
Whatever it identifies as subject using “default” it will suggest a mask. The default subject detection can be controlled in the settings.
The AI simply proposes something based on the data but it might not do what you want hence you have to check or change the default to “none” and do the masking yourself.
Tried starting from scratch with the original scanned-film TIFF file I sent to illustrate the issue I had after processing using Topaz AI. On my second try I experienced the same issue with the automated masking process imprecisely masking the subject using the [limited] setting options.
If having to select the subject manually using the app’s brush tools is the only solution for overcoming the limitations of AI doing it for me what I anticipate is that a great deal of time will be spent manually creating an accurate mask. (I say this because the brush tool in this iteration of the app, is rather tedious to use.)
I feel that the Topaz manual brush masking tool is too crude for quick, critical masking work. Photoshop has multiple options to isolate/mask a subject from other elements in a scene. Are the PS tools perfect? No, not even close, but far more capable and faster to use than the fallback brush tool in Topaz. That said, the question I ask now is whether human intervention in the AI process engine allows AI to take into consideration what the human editor is attempting to do/correct while AI is interpreting the purpose of the task at hand. My gut feeling is that AI is (mostly) ignoring the attempt at human intervention resulting in the incomplete processing cycle that inspired this support issue. Admittedly, my limited familiarity with Topaz idiosyncrasies leads to the concern that AI wants to work by itself.
Additionally, very little of the workflow I’ve experienced using this application has been intuitive. Absolutely, I wish that the tool bar could be moved off my image editing screen and expanded on my other monitor where there is plenty of room to prevent having to open and close multiple editing drop-downs to find tool palates displaying the tool I need.
I definitely see the promise of this application but given the slowness of producing reliable, acceptable results I have serious reservations about how well it will fit into our day-to-day work environment…which currently contains an extensive archive of scanned transparencies waiting to be processed. Being an optimist I’m going to continue to try to use Topaz on other images to get a better feeling for it’s utility in real-world editing.
Hello Konrad,
As requested, here’s a 100% JPEG of the original TIFF file as it was saved after scanning on our film scanner. Hopefully it will help you determine what the issue I’m having is with Topaz AI processing.
Thank you very much for looking into this for us.
From a visual point of view, I agree that low-res is better than high fidelity.
It already removes noise and sharpens the image plus the mask is not bad.
The source has here and there already a few issues. It is hard to tell if this is partly due to the used lens or comes from the scan but some blurry areas are not great.
So, AI cannot fix that, but if you intend to print it even in an A3 size it will look good.
Thanks for looking this example image over. We’ll try a number of other scanned images to become more acquainted with how Topaz interprets them. At present, one of the functions we hope the app can do for us is to enhance marketability of our images for broader potential use.
The incomplete sharpening issue we encountered after changing various settings and and applying different options prior to saving the final file may well have been responsible for the sharpening anomaly. We’ll let you know if we encounter any other issues. Thanks!