What are Chat GPT and Midjourney?

ChatGPT and Midjourney both use generative Artificial Intelligence (AI) to create outputs in response to what a human user asks them for. In the case of ChatGPT, the output is text while for Midjourney, it’s an image.

ChatGPT took the world by storm in November 2022. In tech circles, it may not have been anything that new or exciting but to the average person, it was pretty space-age. And that was where OpenAI (the creators of ChatGPT) had their master stroke: they released it publicly and asked everyone to get in the sandpit and play with it. I’ll leave the ethics of that decision to others but it certainly created a moment, and a lot of media coverage. 

Midjourney is nowhere near as well-known as ChatGPT, perhaps because it needs a paid subscription, or perhaps because images are a bit more niche than the written word. However, it was causing a stir of its own after a state fair in Colorado awarded first prize in the digital art division to a piece of work created using Midjourney (and other tools). The artist did not hide the fact that he used the AI tool but still found a lot of hostility directed at him about whether he was worthy of the prize. 

Why should I care?

Whether you intend using ChatGPT and Midjourney or not, I would argue it’s important for anyone who consumes content (that’s you!) to be aware that AI-generated content is out there. It’s only getting more common and will keep getting harder to tell apart from purely human-generated content. I won’t say one is real and the other isn’t - in the digital world, both can be equally viable. This isn’t such a problem when content is just something pretty to look at but if it’s created to deceive, that is concerning. Most people aren’t visually trained the way an artist or photographer might be. Add to that the speed with which we whiz through social media posts plus our own biases… well, it opens us up to being manipulated without even being aware of it.

The purpose of this blog post

Quite simply, I wanted to share the experience I had using two AI tools to help me in my creative workflow. Hopefully it will spark some ideas or answer some questions for you. I don’t want to convince you either way whether to use or not use these tools. It was an experiment for my own interest but maybe it will interest you as well!

Getting started with ChatGPT

Just go to the ChatGPT login page and click “sign up”. You can use your existing Google, Microsoft or Apple account to get started with a couple of clicks, or enter an email and password. Accounts are free at the time of writing. 

Once you’re in, the interface couldn’t be easier to use. Use their suggested openers or type your own question / instruction into the text bubble. ChatGPT even saves all your conversations for future reference and gives them descriptive titles. 

As the disclaimer says: ChatGPT can make mistakes. Consider checking important information. The program has been criticised for confidently supplying totally false information as though it were true. Make up your own mind about the answers you’re given.

How I used ChatGPT

I used ChatGPT as an idea generation machine, rather than to write paragraphs of text. For this purpose, it’s really quite outstanding. My opening statement was “I'm looking for ideas about which designs sell best on print on demand sites.” Within seconds, ChatGPT gave me ten options. Some of these were styles rather than topics (e.g. abstract, minimalist or geometric) but the topics it gave were spot-on for the designs I’ve seen on various print-on-demand sites. 

It even gave me some helpful general advice, such as reminding me to put my own creative spin on a design to stand out in a competitive market, or to keep an eye on changing trends. Nice!

Option 4 on the list was Nature and Animals, so my next request was a deep dive into this category. I asked for ideas on topics which could be turned into a collection of 10-12 artworks with a theme. Again, ChatGPT gave me a list of 12 really good ideas with a sentence expanding on the heading. 

I was drawn to the final option, Celestial Creatures, which ChatGPT described as: Combine the beauty of animals with the wonder of the cosmos by creating artworks that depict animals against a backdrop of stars, galaxies, and celestial phenomena. A bit more solid advice about how colour palette and style can be used to make a cohesive, appealing collection. Thanks!

Then, a third and final question got me a list of 14 constellations containing animals. A last suggestion too about how to combine elements of mythology, the animal kingdom and the night sky. I was on my way, armed with a list of ideas to choose from.

Pros and cons of using ChatGPT

For me, in this scenario, the pros were overwhelming. I felt like I had a partner who would never tire of supporting me, giving me their best ideas in a cheerful and conversational tone. Nothing was too much trouble for ChatGPT and the advice, while generic, was actually a nice touch.

This was a very basic use case, though. Coming up with lists, topics or funny names is pretty easy. If you’re generating longer content, such as an email or social media post, ChatGPT works best with guidance. For example, provide the main talking points you want included and tell it what tone of voice you want for the audience you have in mind. Then edit the outcome before you use it. ChatGPT should be a starting point, not an end point, otherwise you run the risk of sounding like everyone else who cut and pasted their chats straight from the bot’s mouth.

Keep in mind also that ChatGPT can only rearrange and reformulate the content it’s been trained on, which is what it finds on the internet. If we lived in a completely unbiased, fair and thoughtful world, that would be just fine. We all know that’s not the case, so I suggest you read ChatGPT’s outputs with a critical eye. Be aware of how it might be reflecting aspects of society that you don’t necessarily support. 

Getting started with Midjourney

Midjourney is hosted on an app called Discord. It’s one of a vast number of communities, or servers, where people with similar interests connect via voice, video or text. In this case, you communicate only with the Midjourney bot, providing it with prompts which it turns into images. 

Steps for setting up Midjourney

1. Visit the Discord starter page for instructions on how to set up and verify an account, and how to get the app on your desktop or mobile
2. Create a Midjourney account by logging in with your Discord account details
3. Choose a Midjourney subscription plan, starting at US$10 per month. The main difference between plans is how much processor time you purchase, which is what generates your images. I recommend starting on the lowest plan: image generation takes a matter of seconds, and you can always upgrade more easily than downgrade
4. Open Discord and join the Midjourney server. Click the + symbol on the left hand sidebar, click “Join a server”, and paste the direct link http://discord.gg/midjourney 

If you’re like me and want to read up on what’s possible before getting started, go to Step 5. If you prefer to jump in and learn by experimenting, go to Step 6 first and come back to Step 5. 

5. Read how prompts work, and what they offer on the Midjourney support page. Start with Prompts and Explore Prompting, then go through the sections:

• Commands Parameters and Tools
• Writing Prompts
• Styles and Aesthetics
• Image Size and Aspect Ratio
• Advanced Prompting Tools

It’s possible to have a private channel with the Midjourney bot, which is what I did. I wanted all my images kept together so I could refer back to them later. There are some instructions here but a word of caution: I found about 20 accounts all called ‘Midjourney Bot’ or some variation, but only one was the real bot, not some person who set up a fake bot account. The real bot is identified by a blue rectangle with a tick, and also by the fact that it is always online.

How I used Midjourney

Following my brainstorming session with ChatGPT, I decided to create a series based on animal-themed constellations. I came across NASA’s amazing Hubblesite where they generously post the most inspiring pictures of space for anyone to use. 

Luckily, I picked Pegasus to work on first, and found an image of the actual Pegasus constellation. By Googling maps of the constellation, I understood how the horse would need to be positioned to fit the star arrangement. Of course, I didn’t want to use someone else’s Pegasus, even if I planned on changing it and incorporating it into my artwork. That’s where Midjourney came in - a unique image I have full rights to, tailor made to my instructions. Easy, right? Well, it wasn’t quite that straightforward…

My first thought was to use the Hubble image plus word prompts to get a finished product directly from Midjourney. The first attempt was sort of OK style-wise but the positioning of the horse didn’t align with the constellation. 

The second attempt was, frankly, weird.

So, I changed tack and asked Midjourney for the Pegasus on its own instead, which I could then layer into Photoshop. I knew the rough position I needed, I wanted it full-body, in landscape format, with no background, and I wanted it white for blending with other layers. It took 17 attempts before I finally arrived at this:

Now, this image does have a background but the horse’s stance was the best I had which didn’t cut off feet or wingtips. It also had (only) four legs, a realistic anatomy, and no strange fish-like appendages around the mouth (which I think had once been reins in a source image). Getting rid of the background took a lot of time due to the intricate outline and tonal similarities, but I got there in the end. I used Photoshop to layer this with the background and stars, for an end result I was quite pleased with.

The next constellation I worked on was Draco, the dragon. Again, I knew the shape I wanted to fit to, portrait format this time, still no background and plain white. It took another 17 versions of impossibly-knotted freak show attractions with multiple heads and feet coming from everywhere before I settled for this:

The plain background was much easier to remove this time. I used Photoshop’s Warp tool to stretch the dragon into a better shape for the constellation, then layered it with a Hubble photo of an S-shaped nebula and stars to reach a final product:

Pros and cons of using Midjourney

I finished six constellation artworks, using Midjourney and Photoshop. Here are my thoughts after using Midjourney and reading their website.

Pros of using Midjourney

• Images produced are completely unique
• You own the assets Midjourney creates from your prompts, with very few restrictions
• It can produce outcomes in a fraction of the time it would take using traditional methods
• Prompts can be image-based, text-based, or a combination of both

Cons of using Midjourney

• Midjourney retains a royalty-free copyright to use any of your images and prompts
• If you’re aiming for a very specific outcome, Midjourney can be quite frustrating. Once it had produced the first outcome for a prompt, I found it almost impossible to get something significantly different. Looking at the public channels, I don’t think I can be the only one with this problem. Many times I saw people prompting over and again, clearly not getting what they had in mind
• Midjourney is open about the fact it ‘has been trained to produce images that favor artistic color, composition, and forms.’ While there are prompts to reduce this training, it’s hard to break totally. I felt like I’d stumbled into a geek-boy fantasy world sometimes. Women were always young and dreamy, unless you asked for an old woman, in which case you got a Grimm Brothers crone. The prompt for a leaping dog produced a snarling hell-hound looking to rip out a throat
• Midjourney has quite a problem with repeating elements: fingers, legs, or groups of items for example. I understand this is improving all the time but too many or too few elements are very common issues
• It also struggles to provide anatomically correct animals, birds, or even fruit, unless they are very common. Bears were fine. Dogs, pretty good, although a stray fifth leg would appear now and then. Ravens almost got there but had the hooked beak of an eagle. Sperm whales? Not a chance. I have no idea what these are - tuna or shark would be my guess. 

Would I use the same method again?

In short, yes. Working with ChatGPT for idea generation was easy and I still have a long list of ideas to develop. Working with Midjourney was not as easy as I’d hoped but it was still better than the alternatives I saw. Taking my own photographs of these creatures would be basically impossible, even if they aren’t mythological. Finding online images with the right positioning and a decent resolution would have taken a long time, plus editing time on top of that. I don’t have the skills to produce them in Illustrator or Procreate, and I definitely didn’t want to lift someone else’s artwork. Midjourney gave me the raw material I needed to bring my vision to life. 

You can see all of the constellation series here.