Yes Anthony. That image worked better without FaceRestore as the two foreground faces are sufficiently clear. However, on some of my low res JPGs from 20 years ago, Face Restore can help with people who are in focus, but overdoes people in the background. That’s why people are shouting for a face selection tool! Hope you can work it out.
2 Likes
John Petter
Face Restore is not intended for use on portraits where the subject is in the foreground.
Were these faculty or grad students shots taken with the Mavica?
I had an old Kodak digital cam 1st. 2-3 m.p. Think would have been late 90’s. Sometime after the “brick” military looking “cell phones”…
Yes, I bought this with the intention to do large batches of images, doing them one by one and having to wait for the preview as well takes forever. Not efficient at all.
Yes, much slower workflow now you can’t batch everything with the same settings. The upscale defaults to 4x, I had to go through a load of images one by one just to set them to 2x.
He said the Autopilot chose ‘Recover faces’, if it wasn’t meant to, then why did it?
1 Like
Imo, actually I had the Mavica for about a year before this picture was taken so I wasn’t too far behind you! 
By 2005 I was using a Panasonic FZ-20 SuperZoom with 5 megapixel resolution and upgraded my equipment every couple of years since then until the pandemic. I did try Photo AI on my newer high-resolution images and it worked quite well but I would still like to have a details slider available under the sharpening tool so I can be more precise in the details aspect of the image without reducing sharpening overall. I see other users have requested a masking brush and that would work too. When I was trying to restore the details, I tried first to adjust the noise slider but determined that it was the sharpening slider that was just too strong.
1 Like
Yes, these are the Pro Mavica shots.
In the late '90s I also used a Kodak DC265 1MP camera 
Yes, in all of my examples above, I have simply opened an image, and the Autopilot in Photo AI has done the rest, initially.
If I’m not supposed to use ‘Recover faces’ on images with a high quality face, that’s fine by me, but the selling point of the software is that it makes the decisions for you (mostly), and it invariably (in these cases) has used ‘Recover faces’ and set it to 100% automatically.
1 Like
i have noticed the same problem with P ai compared to denoise ai (pictured left). P ai is smearing detail, and this is with the noise slider set to 8. no enlargement either.
Correct, I currently use Topaz products (Gigapixel AI) mostly to upscale pictures that most of the time are not high enough quality / resolution to be send to my lasers that 318 DPI is the native DPI. I can do the calculations to get the DPI from the pixels, but Gigapixel AI does this automatically today.
Also no option to do anything but 1, 2, 4 or 6x, where Gigapixel has the option of going by inch and by DPI to have the proper output for my lasers. So even if I get the DPI (or pixels) calculations, no way to use Photo AI to use those numbers for the native output into LightBurn since the only options are 1, 2, 4 or 6x.
I just tried to manually change the desired width or height and it doesn’t recalculate the other value (assuming they are locked ratio) or allow to change the 2nd variable. If the width is 2000 pixels and I enter 800 then go over to length field, the width field goes back to 2000 pixels. So don’t see that option in Photo AI 1.0.
1 Like
Using two monitors and the CURSER disappears on the monitor that is showing Topaz AI
I to miss the possibility to redefine the select area. Please add this feature to a newer version.
Definitely need to be able to add or remove masked areas as PAI does not differentiate between faces that should be in focus and those outside the focal plane that should be blurred. I have had issues where i wanted facial recovery on a slightly out of focus face but not on a face in focus. The AI took the sharp focused face and blurred it while attempting to recover the blurred face. Hair masking is poor. AI also randomly renders textured areas out of focus (grass).
3 Likes
There are many things AI is not ready for at this point. It cannot know the mind of a user. It appears Topaz is of the mindset that this AI is capable of making all these decisions for us. This may be a good option for a novice user who isn’t overly concerned with optimal quality, but for advanced amateurs and professionals, this move is not ideal. By stepping back from the Standalone apps and diverting focus to this app, it does many a disservice. Unless PAI can bring all the capabilities AND EXISTING OPTIONS that exist in the SA apps, and additionally bring more to the table, then we as users are losing flexibility. I hope that Topaz will start to hear what the users here are saying as it appears they are not right now. If I were to have purchased this product, I would have been thoroughly dissappointed. This is really an extended Beta test and this app really isn’t ready for retail. I know this seems very critical, but with each new release I have tested, I am still frustrated by many recurrent issues and the lack of integration of what should be simple options to port over from the existing apps.
8 Likes
I have been with Topaz just over a year and every time they send a upgrade for one of my program bundles i have a problem with download and now with the so called all dancing Photo A.I. I DOWNLOADED IT AND IT JUST FROZE WHEN I TRIED TO LOAD A PHOTO BUT TOPAZ ARE QUICK TO POINT TO YOUR COMPUTER when there is a problem also to try and post on there community page is a nightmare same with everything with Topaz they expect everyone to be a i.t. expert maybe its time to consider changing !!! having just started to get to grips with the basics of my Bundle they bring another program out that doesnt work on my computer and before Topaz i never had a problem with my p.c.
I had the same issue several versions ago. High frequency detail is obliterated.
Please note that there an error in the decimal point of the jpeg output size on the Queue screen. For example, Original 4.8mb shows as output 61.7mb, which should be 6.17. Maybe noted already?
