Artificial Intelligence is changing the way post processing is made (it is also at work when you snap a photo…). Applications like Luminar and Photolemur rode the wave of AI to create products more advanced than Photoshop (the real benchmark in this area), where more advanced refers to the ease of use for the rest of us (with professionals still using Photoshop).
As one could have expected, Adobe is not resting on their laurels and have come up with a sneak peek at their new release of Photoshop expected to hit the market in the coming months.
As you can see from the video clip they are introducing a “select subject” tool that would allow you to select an object, a person, with just one click.
Actually, if you look at the clip you will notice that the result is not “perfect” yet. The shadow under the girl leg is selected along with the girl, in another example a part of the leg of a person is not selected. All in all it is a significant progress if we compare at what we need to do today.
I have tried similar tools in Photolemur and Luminar and in both cases there are some glitches. The AI provides you with a good starting point but you still need to work on it.
This is both an amazing result and a reality check on the status of AI today. Much better than it used to be, much much better than software without AI but still not “au pair” with our capabilities in image recognition.
In reality the identification of a subject is really difficult. It goes beyond identifying an object, it requires the understanding of “why” you want to identify it. If I have a nice beach with people sunbathing and I want to remove the people it is not enough to identify them and replacing with sand like background (or partly sand and partly sea). It is most likely that these people (some of them) are casting a shadow on the beach and removing them while leaving the shadow would result in a very poor photo clearly showing the manipulation. Hence, the identification of a person and its removal should include the smoothing of the shadow.
It would be interesting to see is the “select subject” feature can learn from experience and get better. This is not mentioned in the Adobe presentation and I did not notice any improvement when using Photolemur or Luminar. I guess it is not there, yet.