I couldn’t resist trying the AI’s ability to generate images and illustrations from some user-provided text. In particular, I’m talking about Generative AI, a machine learning system capable of generating new content or output that has not explicitly been programmed into the system.

So, I set up an account on Midjourney and paid for a month of the most expensive package, the Pro Plan, at $60 + VAT. I chose this plan because it allows me to run the ideas in stealth mode, and the results are visible only to me, not to other system users.

I hope to have several hundred images and illustrations to upload to Microstock at the end of a month, as Adobe Stock and Dreamstime accept content created with AI tools.

At the end of two weeks, I ran a thousand text-to-image time prompts on Midjorney, and I am disappointed with what can be done with this AI.

The most annoying thing is that it doesn’t generate realistic images. Instead, Midjourney’s creations are Frankenstein-ish: they combine a bit of everything without keeping some basic rules in mind.

For example, every insect has six legs. Midjourney didn’t know that and came up with all sorts of nonsense.

A green shiny fly on a white background

Spiders have 8 legs, and everyone knows they don’t have wings. Unfortunately, Midjourney has yet to hear that and comes up with stuff that makes you laugh and cry after wasting over twenty minutes without generating a usable result.

A spider sitting on a spider web woven between trees in a forest

And it’s the same with people – it puts too many limbs or too few.

The image below should have been a witch flying on a broom. What came out resembles the fruit of love between a stork and a bat.

This was supposed to be a witch flying on a broom.

Creature symmetry of any kind is also optional at Midjourney. Eyes of different shapes or colors. Different numbers of fingers between left and right hands, such are the things I’ve come across so far.

Easter rabbit in a kitchen. Poor rabbit’s missing a paw.

Another problem is keeping the right proportions – some objects in the images are exaggeratedly large.

I collected many such misses this month using Midjourney heavily; you can see some in the gallery below.

Sometimes such problems can be easily fixed in Photoshop. However, other times the mistakes are so big that the time it takes to correct them becomes too long, considering the low sales volume of this art genre.

Another annoying thing is the resolution of the images created, of a maximum of 2048 x 2048 pixels. To get something of a decent size, I have to use Adobe Bridge to convert the .png files created by Midjourney to .dng files. The latter can be opened in Camera Raw in Photoshop, and I can use the Super Resolution feature.

There are some good things too. Sometimes I didn’t get what I wanted, but I ended up with some creations that I would never have thought of and that aesthetically look great.

Regarding the acceptance rate of my creations, I am doing fine. Everything I sent was approved. But I sent a very medium number of images. This is because I prefer to spend my remaining subscription time generating as many ideas as possible, and at the end, I will process them and upload them to Microstock.

So it could be better, but it’s alright, too. It takes a lot of patience to generate something usable. Sometimes I ran the same text-to-image prompt ten times until I got what I wanted.

What is certain is that at the end of these 30 days, I will not renew my subscription. I’m not fully satisfied with what I can get now. I think more time is needed for AI-based tools to evolve in such a way as to produce worthwhile images.

The only thing worth it is that it allowed me to add illustrations to my portfolio, as I was previously only focused on photography and video. It remains to be seen if my creations will sell and if I will ever cover the costs of this little adventure.