experiment with generative fill

[DanK]

Today I played with this image in part to test out Photoshop's new "generative fill" as a technique for cloning out distractions. I'm not sure the image itself has a great deal of potential (comments welcome on that, of course), but I was very pleased with generative fill. It didn't work 100% of the time, but most of the time, if I simply selected the area with a not-too-loose lasso and told it to generate, it removed the object and filled the area well, with no further touch up needed. There were signs, light fixtures, and other such in this one.

I'm not at all an enthusiast for using AI to pepper my own photos with images Adobe has scraped from other people's work, but to the extent that it simplifies tedious tasks like cloning, I'm a happy camper.

[Manfred M]

Thanks for these comments Dan.

In my view, Adobe has been slowly moving in this direction over the past number of years and while I have used the Content Aware Fill in some of my work, that technology does have its limits. From an "ethical" standpoint, it uses my own work to create new content, rather than generating content based on data that has been used to create training a training set. The ethics of how this software was trained is still being vigorously debated, as are some of the legal issues like copyright / owenership.

That being said, it can be a real time-saver when compared to the alternatives of manually building up the elements in the image that need to be repaired, as you have mentioned.

I suspect that future iterations of this technology will continue to improve, but like you, I feel a "light touch" is probably going to be the way I use it in my own images.