I've been experimenting with Midjourney, an AI text-to-image program that takes prompts and generates artwork. It's... pretty amazing.

Checkout the Midjourney site to get started. Once signed up, you'll need to head on over to discord to start using Midjourney.

You'll use a discord channel to generate your artwork. The bot will introduce itself, and direct you to accept the Terms of Service, and to jump into a newbie channel, to start creating. Once in a channel, simply type in /imagine followed by a description of the artwork you'd like to generate.

Midjourney will generate a grid of 4 images for you. You can then hone-in on the one you like, and have the option to either create a variation on one of them, or upscale one of them. I think upscale generates the image again, at a higher resolution, in more detail.

Since it uses Discord, it has a low barrier to entry. You can change the settings of variations to 'Remix' mode, by typing /prefer remix.

In remix mode, creating a variation brings the prompt up again, so you can edit it. This is great, as you can further refine your idea.

To test it out, I decided to try and recreate an artwork similar to those awesome Galaxy of Horrors posters that NASA created, and let you download and print for free.

So I entered the prompt:

/imagine colorful galaxy getting swallowed by a black hole

It generated 4 images that looked really cool. So I iterated on it, making variations, picking the one that looked a little more visually interesting.

The posters have beautiful typography, that reminds me of B-movie or sci-fi/horror movie posters from the 50's and 60's. So I tried a prompt that included adding text.

This didn't really work. The text wasn't real text.. it just kinda looked like characters. And, it was swirling around along with all the other colors that were getting sucked into the black hole.

So I went back to a previous version. I tried the 'Beta Upscale Redo' button - not really sure what that meant. After reading their docs briefly, it looks like one thing that it does, is create an image that is 2048px x 2048px.

After googling whether Midjourney can add accurate text to it's images, a few reddit posts suggested that you can't reliably do that. They suggested combining your artwork with text in another program. So I decided to export my black hole, and finish it in Figma.

After having a play in Figma with text, experimenting with fonts and grunge textures, and trying out some options for warping the text, this is my final artwork. I decided to put some shocked 1950's comic style people in there as well (also generated with Midjourney).

Here is my final result.

I'm pretty happy with it.

The ethics of AI generated art

With the rise of AI and machine learning tools like Midjourney, as well as other tools like ChatGPT, there certainly are lots of questions that arise. These tools use machine learning, and have trained their models on hundreds of thousands of images, documents, etc. found on the internet. It's like that much of this material not copyright free, and that the artists of this material have a reasonable claim to compensation from companies like Midjourney, or OpenAI. I believe there are lawsuits already underway, and there's a growing movement of creators aiming to boycott AI generated art.

It will be interesting to see how this struggle plays out. I for one am really interested and excited at the possibilities that these technologies will enable.

I've only scratched the surface with this experiment. Midjourney and ChatGPT both make api's available to developers, and it could be a really interesting project to see what kinds of creative service or app you could make with them.

There are also plenty of people on the internet at the moment that are quick to tell you how you can make thousands of dollars with these tools. I'm still a bit skeptical. You might be able to generate artworks with Midjourney, and then sell them on print websites, but... would people really want these things? Some of the artworks are pretty lifeless.

I dunno. What do you think? Is this art?

# Stable diffusion

Create AI generated images with Midjourney