Notes from my first hour…
-
The software was easy to install and easy to use after a few minutes getting used to it.
-
Photo AI 2.2.1 crashed once after toggling several options and then sliding toward the Feedback button. I’m not sure whether I clicked that button before the crash. After restarting a longer session had no issues.
-
A Manta Ray photo with backscatter is somewhat worse after the recommended noise reduction. Sensor noise is improved, but the backscatter spots seem brighter than before, so they are arguably more objectionable.
-
I never found a way to address backscatter, which is often the dominant flaw which needs to be fixed in underwater photos.
*Balance color was not recommended for the Manta, but I tried it. The photo came out much redder than in real life. Perhaps an underwater setting could be offered, or the AI could detect scenes which are naturally low in red light. -
A White-capped Albatross was upscaled. IMHO, an image should never be upscaled unless requested.
-
For the Albatross, upscaling seemed to have a noise reduction effect, but noise reduction was not selected. When I reversed these settings, the results seemed better.
-
A Cormorant looked rather artificial with the recommended standard subject sharpening. Changing strength from 44 to 33 provided useful sharpening while still looking natural.
-
The Cormorant really needed noise reduction, which was not recommended. Look at the difference in the eye with sharpening at 33 and normal noise reduction at 28. The improvement was not as good with noise reduction or sharpening alone.
-
An already-sharp shot of another photographer was given subject sharpening, standard, strength 37. Sharpening was not noticeable, which is good. Shadows seem to be brightened, which wasn’t necessarily bad, but unexpected from a sharpening operation.
-
A badly backlit bird captured 1 stop overexposed (sometimes this is the only shot you can get) was improved a little by the recommended sharpening. I tried “Adjust Lighting BETA” but it looks a little worse. The sky is still washed out (which may be impossible to fix) but shadow detail is lost.
-
A gull standing on a mossy rock was nicely improved, though the selection missed the legs and feet.
-
Subject detection of a group of Albatross in flight was pretty impressive. Raw and JPEG results were slightly different. In both cases five overlapping birds were correctly detected as the subject. Part of one tail and the body of one bird were missed. A couple of more distant birds were also missed.
-
I don’t understand why there’s a message telling me that subject must be selected before only the subject can be sharpened. That is already happening automatically and there is no obvious way to change the selection.
-
A rare shot with two straight-on human faces was interesting. We look a little younger, but detail is lost. More oddly, a knit wool hat which was part of the subject but not of the faces became quite blurred, losing the ribbing of the knit. A shirt, jacket, and some text were improved.
I’m sorry if this seems overly critical, but give me a Feedback button and you are going to get feedback. 
As you can tell I’m mainly a wildlife photographer, strictly amateur. Perhaps the AI has been mostly trained on other types of photography, which makes it less likely to choose the right settings here. I am a scientist, and while I’m not a “postprocessing is lying” purist, I will choose accuracy over a big “Wow” factor, even for images not used as data.
Background: Images were a mix of OM-1 and Canon G7X files, mostly JPEG, but some RAW. I’m running on Windows 11 Pro with 24 cores and 32 GB of RAM.
Added few minutes later:
- I have now found options in Preferences which will help with some of my issues listed above.
- Face recovery was amazingly good on some scans from old photographs.
- I know that this is not general editing software, but a simple left/right/180 rotation option would be very handy for scans.
- One more oddity. The face on the left was detected and improved. The one on the right was only detected at low confidence and was oddly distorted by the noise reduction.
- After manually adding the second boy to the selection, his face was improved.
One question: Is the AI learning locally from my actions? For example if I frequently turn down the noise reduction will it start to do that on its own?
