How to make a great AI image: Red apple

I want to show you how to make a great AI image: what goes into it and that it’s not just a matter of pressing a button and something great comes out.

Lately there has been a lot of discussion about AI imagery.

Note I do not call it “art” for several reasons. Not that I don’t think it is art. I don’t actually care whether it’s “art” or not, because I don’t want to be part of that pretentious discussion. Let’s just leave it at that.

I simply enjoy it.

Whatever position you take on MidJourney, one thing I want people to understand. It’s a thing that many people don’t understand: you do NOT press a button and a great image comes out.

Creating a great image requires a lot of time, dedication, creative “vision” (a horribly pretentious word, but I mean, knowing what you want and what would make the image great), knowledge of some pretty weird prompting language. The image for one of the previous posts (with the black and red trees) heavily used the “elliptic curve optimization” parameter. You need to understand things like prompt weights. It’s also not programming because the results are random, but you can build on that randomness and push it until it breaks. That is the point where truly beautiful unique images come out. There are people who are waaay better at this than I am. I watch them in awe, just as I watch visual artists in awe.

Many people would also further process AI imagery, either by drawing over the images (as I’ve done with the Fantabulous Hounds book) or by bashing images together or photoshopping different elements onto it.

The vision and processing is where the magic happens.

Anyway, I had some fun yesterday and thought I’d take you through the process of creating one image, which took me an evening, and so much prompt re-rolling that I would have blown through four times the monthly allocated $10 subscription in one evening (just in case you’re asking if the $10 a month plan is enough and I’m like absolutely not).

I give you: the red apple.

A single red apple on a tree in a desolate colourless landscape. That’s pretty much the prompt I started out with.

There are quite a few of these types of conceptual images on stock photo sites. Many of them will have been created artificially in some way (either AI or with 3D models).

The first lot of images were pretty bad, although they understood the colourless bit, and none of them gave the trees any leaves, but often the apple was just randomly hanging in the sky without being attached to the tree.

The above two sets of preliminary images (MidJourney gives you a grid of four and you choose which one you want to work with), I particularly liked the first one because of the moon. A lot of the images were a bit meh and without much contrast. I also liked the last one because of the shape of the tree and the depth.

Then I got this image where the moon is behind the tree, which I liked, but otherwise this image is too pale and without contrast.

After a few re-rolls, I got this one, which has a much better composition, but now it’s introduced an annoying AI artefact: the drips hanging from everything. They’re a dead giveaway for AI, and although they can have a great effect, I don’t want them.

This is better, but still some drips and now the apple has two stalks.

This one I really like. AI tends to add bodies of water, but cracked earth fits the theme much better.

Although this image is usable out of the box, it also has another annoying AI feature: the focus is smack-bang in the middle of the image.

So I used the pan feature and added language like “desolate landscape with a dead tree” for Midjourney to fill the strip to the right of the image with something I liked. It tried to sell me second moons, additional apples, trees with leaves and other weird stuff, but eventually I got the finished image below:

I might pull this into Photoshop and use some filters to enhance it, but I wanted to show you what the raw MidJourney output looks like.

In January, I made a full colour book with AI-generated pictures of greyhounds doing human things. You can see it here:

Fantabulous Hounds

Comments are closed.