I’ve now spent a day or so playing around with Adobe’s Generate Fill option and I think I have got a reasonable first impression on it what it can and cannot do. The first thing that I have discovered is that it is really good at generating content to fill an area (the name really gives that away!). The examples from the last two posts show how useful that can be.

The second lesson is that there are some clear ethical as well as copyright rules in what it will and will not generate. This is how it should be but I suspect there are AI generation software out there which has few if any of these rules written into their programming.

The third lesson is that once you start to push software away from ‘fill’ it can very quickly get itself into a mess or the results are worse than some basic cloning of a second image on.top of another. In the above image I asked it change the colour of the hat which it has made a reasonable stab at. It is far from perfect but with a bit of further work in Photoshop I’m sure I could make it look a lot better. However, the next three examples show the limits of the software. I asked for a Woolen hat, Baseball cap and a Fur Hat.

The best thing that can be said about the Baseball Cap generation is that it is period sensitive. Other than that it is not really that good.

So perhaps my concerns about fake photos causing problems by using the Adobe version of AI was, thankfully, a overblown. But I suspect the second or third iteration may well be significantly better at generating content. Also Adobe is a huge corporation so they will be very cautious about how they implement their AI generative software. This being said my YouTube feed is already getting videos on how to hack aspects of Generative Fill so now this software is out their how the software will develop is anyone’s guess.

For me I have no interest in changing the colour of some garment or change a jacket for a t shirt, yes it will do that. Instead I will continue to use the software for what I believe it is really good at, adding slices of background as and where I need it, to make the photo look a lot better as here.