A few months ago I became aware that Adobe was trialing some new AI features in a Beta version of Photoshop. Numerous YouTube videos started popping up on YouTube from photographers and content creators singing its praises and proclaiming it as a "game-changer". Although I was a bit sceptical, I decided it was worth downloading the Beta version to try it for myself. I am a rare user of Photoshop as I have never had enough use for it to justify spending all the time needed to use it proficiently. 99% of the time Lightroom has all that I need.
For those who don't know, I should just briefly explain what Generate AI does and how it differs from other Photoshop techniques and tools. I'll then report a few of my experiences and observations and offer a few tips on how to get the most out of it.
What makes the generative AI tool in Photoshop different as that it generates entirely new images based on text prompts you give it and using parts of the surrounding image. When it works, the results can be amazing, it will judge where the light is coming from and adjust shadows accordingly, as in the example image below. However, more often than not, the results can be pretty terrible.
Does this mean that the AI generative fill is a waste of time ? Absolutely not. It has some limitations, but it also has some big possibilities and has been constantly improving since it was first released. It is important to understand what it is good for and what it doesn't do well. There are also a few tricks and tips that can help get better results, but I won't go into those too deeply here and there are plenty of YouTube videos out there that will explain it better than I can. However, I will mention a few tips I have picked up through my experiences.
What it does poorly
I don't think it's a secret to say that it doesn't do people well at all, especially if you try to generate entirely new ones from scratch. They will genuinely look like a bunch of weird, malformed aliens. Also attempting to change many basic human features - hands, eyes, legs etc, will get pretty horrible results. So when you are selecting areas to change, avoid hands in particular if you can.
Another thing it doesn't do too well is create objects that fit the lighting of the scene. Often they will be too bright and oversaturated. Of course, you can do various things to adjust thes, but it becomes so fiddly that you might as well have just found the image yourself and created a composite image. This is particularly the case where you have images that are dark, moody or misty. My advice is, just try generating things that are going to blend in with the existing lighting of your starting image.
What it does well and how it can improve your creative photography
The first thing it does well is to remove objects seemlessly. It does this generally a lot better and a lot more cleanly than things like the Remove Tool.
In fact I can show an example to illustrate this. In this particular image there is a very bright coloured light on a metal fence behind the model. It was meant to be a backlight, but in this shot I let it get into the shot, which kind of ruined it with a blob of overexposed light in the middle of the shot. Impossible to remove cleanly in Lightroom. In addition, it's fairly obvious that the cigarette she was "smoking" was not real, so I was able to add just a whisp of smoke to improve the image that I wouldn't have been able to do naturally.
Original image
Enhanced image
If you want to extend of expand a background of an image, the tool can also be very effective and can help create a better composition by altering how your subject is framed. It's a similar concept to cropping, but in the opposite direction.
In the example below, I had originally taken the photo with a long lens to compres the distance between the subject and background. However, I wanted the grafitti wall to fill up more of the background as well, and the Generative Fill feature allowed me to do that and create a more dramatic image ( in my humble opinion).
Sometimes you may take a photo where you may feel the main subject looks pretty good, but the composition is unbalanced. You may want to add background elements that highlight or draw attention to your main subject, or you may just want to add some kind of framing elements. For this kind of thing, I think Generative fill can really help stimulate your creativity and improve your compositions. I wouldn't necessarily use the results as a final photo, but it can offer you ideas to improve or adjust your compositions in future shoots. For me, it's also helped me focus on story-telling in an image. Your start to think about what each object or element tells you, and how it effects the overall composition.
This is especially effective for portaits I've found. So many portaits are taken in a studio with a black ground with nothing but head and shoulders and facial features softened by lighting or Photoshopping. It makes for a very generic portrait and gives you nothing about the person. I feel that a good portrait is one which the viewer engages with.
Let me try and give a few examples.
The original photo below was taken at a multi-storey car park and Vaughan was just leaning over a railing. I thought the railing could be replaced with a bar. The resulting image then suggested to me a guy who was sort of drinking his sorrows away, and so I added a few other elements and adjusted the lighting a bit. And voila ! The whole mood and story of the image is changed, and to be honest, it's a much more interesting picture than the original.
Here is another example. The photo was taken in my living room with one LED light, and the light and shadowy look suggested to me street lighting in a dark alley, so I selected the background and typed in a few different things like "dark alley at night" or "misty city street", until I eventually got something that looked reasonably ok. But there was a bit too much empty space and so I added figures in the background and a bit more mist, and suddenly it became a scene from a film noir movie. So of course, I then worked on the main subject, adding the hat and trenchcoat. I have found that if you want to do this kind of thing, it's best to start from changing the background first and then adjust the main image according to that and then you tend to get better results.
I don't want to bore you with too many explanations of what I've done with every photo, but I'll include a couple more below for illustrative purposes and maybe it will give you some ideas yourself.
In conclusion, I'd say that Generative AI Fill is a useful tool, if used selectively. It can certainly be used in a gimmicky way to create both amusing and unrealistic pictures ( see examples below), although sometimes that can be a legitimate creative method too.