AI generated imagery, is it any good?

AI has suddenly swept in with chat GPT seemingly able to perform any task, we have had our eye on a few other tools, today we are going to look at image generators and AI generated imagery, is it actually any good and how can we use it our advantage?

Article Info

December 5, 2023 10 mins

What we will cover

The recent rise of AI has reached a level that is impossible to ignore and no longer belongs in the realm of the future, feeling something akin to a Terminator film. 

I remember counting down to judgement day as a kid (August 29th 2016) and when it came around, my now adult-mind was somewhat disappointed by the lack of progress made by the robots.  

A similar feeling to the realisation I was never going to see a hoverboard in my youth and escape some goons like Marty Mcfly. 

Already I digress.

AI has been there for a little while, but just recently, it’s improved. It’s actually got good.

We are no longer talking about ground-breaking tech in Silicon Valley that seems worlds away, when we can have our favourite browsers that run off the potential of chat GPT.

Chat GPT is a whole article for another day. Today, I’d like to discuss image generation, and more specifically, Midjourney.

What is an Image Generator and what does it do?

Midjourney is an AI image generator. There are now loads of these, but very few that we have seen that do it this well. Well, assuming you understand how to feed the machine. First, let me start with the principal.

You write a series of words to create a scene and within a couple of minutes it generates an image for you, it will be unique, and yours to use how you wish (more on copyright later). 

The command: paper origami white, white background, forest of trees

With each command that you give Midjourney it will generate 4 images with slightly different styles. You can ask it to have another go, or rework and refine one of the options with more commands until you find the composition you are after – hours of my life are already lost to this process.

How does it really work?

It creates beautiful art all right. Does it understand the thing you are picturing? No, far from it  – and it can be frustrating trying to feed it specific instructions to get to that end.

To understand what is happening, think about it like this, take your command and google it, you will get back a bunch of images in the image search, this is what the AI has to work with, it looks for trends in the commands you give it and goes about combining these trends together.

Which yields exactly as you would expect. Sometimes the results are delightful and sometimes, just plain weird.

You can help yourself by giving it an image as a composition and then tweaking the subject, colours and styles. Meaning that, with time, creativity (and a willingness to understand the commands) you can create visual content that you either couldn’t fathom before or certainly couldn’t budget for.

Getting a budget approved for a space trip photoshoot was always hard to get past the finance team at Vu 🙂

You can allow the AI different levels of creativity by using the command –chaos, and you can quickly change the feel of your composition by using recognised styles like graffiti or pixel art – see how you can reimagine a cat in dozens of ways.

Yeah, it’s pretty good.

So traditional art is dead?

In my opinion quite the opposite, and this is the fundamental challenge we all face with AI. Considering it a threat that is going to take away our jobs is just looking at it from the wrong angle  – AI is going to creep into our lives and take away the repeatable mundane tasks, it already is. 

So we need to ask ourselves “How do I really add value?” – for the copywriter (where AI can produce a blog article in an instant), it is asking: will that copy have the correct tone for my audience? Is it tackling what is unsaid on this topic? 

This highlights that, for all its brilliance and progression, it lacks creativity. It looks at what’s out there on the web and offers a rehash of an image, or a rewording of some old copy. 

The future is already rooted in the past. 

However in Midjourney’s case, if the human visionary can join an existing composition with a style it would never meet, then they may have just created something unique and magical together.

If this has whetted your appetite to have a go, here’s a few first practical findings, if you’re more interested in the legalities of who owns the magical unique thing you just created – scoot to implementation & ownership. 

Some tips for the newbie

This isn’t going to cover all the best ways to create images, between YouTube tutorials and the Midjourney guidance you will find your way to good images with an investment of a few hours. 

What I can bring to the table is how to practically use these images within your digital offering, as the seemingly limitless potential at your fingertips can sometimes lead to a numbing “of where to start?”.

You will likely have a placement in mind, your website banners, social posts, blog images, brochures – the list goes on. 

Language of compositions

Pick one and think of a composition you would like to see. Let’s take our homepage. The phrase “A sustainably focused digital agency” isn’t going to give us much, in fact, let’s have a look…

Some strange compositions because the instructions were vague 

Not exactly what we’re after, but some interesting ideas, so we need to paint the scene a little. “A planet covered in trees and buildings”, then we refine from there.

This is far more consistent as it can understand the parameters of the command

From here we refined the image further to become our new homepage banner.

Dimensions and language

Now, a square image isn’t much use on the web, banners (famed for being very short and wide shots) are going to take just a slither of your subject matter, so changing the aspect ratio with your commands will help. Ar–3:2 for example. 

Be aware if you create a portrait image with just instructions for a person’s face, then it won’t have any information for the gaps and you may get a weird repetition of faces (the only info it has to go on), so try and describe the outfit of the person or background to shape the scene.

Once you have something you want to use, there’s an option within Midjourney to upscale it. This is the process of taking the image, stretching it onto a bigger canvas and filling the gaps so it doesn’t look pixelated.

This is an interesting and slightly less covered area of producing image assets because there’s a few different ways you can do it, and they will yield varying results.

Regular, light, detailed upscale vs external

The upscaling options within Midjourney will output your image from 500px up to 2000 pixels wide, which is pretty much as wide as a standard screen and passable for web, as long as you don’t need to crop into it. 

One thing to be mindful of is that it doesn’t just make the image larger, these various options will also add textures or styles to the image as they go to work. This guide explains all, but my advice is to keep it regular or light and then look at options outside of Midjourney to raise the size. 

If you want to go for higher res, to use in graphics or print for example, then you will need images that are thousands of pixels wider than that. Have a look at this bit of kit from ARC which will polish up any fuzzy jpegs, it’s not the only choice you have either.

The role of Image Upscalers

As we know, an image it’s made up of pixels, thousands of small coloured squares, and when an image gets stretched the squares get bigger leaving an image looking pixelated.

Image upscalers have to predict what would have been left in the void and fill it with smaller blocks of colour.

As you would expect they vary in quality, we tested a few out, and Vance AI seemed to be a good balance of cost and quality.

Vance AI on the left, look at the detail compared to the image on the right, this is Midjourney’s own upscaling tool.

As we see here, they do a much better job than Midjourney, so our advice would be to take an image at its regular output from Midjourney and then use an image upscaler to bring it up to size. 

Implementation & Ownership

So with all these visual tools at your fingertips, you can begin to develop your brand imagery affordably. Though if you are going to do this, it would be worth knowing a little bit about the legalities of ownership.

If You are not a Paid Member of Midjourney, You don’t own the Assets You create. Instead, Midjourney grants You a licence to the Assets under the Creative Commons licence.

If you pay for an account, you can use your art as you want. This includes selling them, but the company can use your image too. There is also confusion over who would own the image should it ever go to court, with the current consensus saying that although you own the copyright the content creator is Midjourney, not the human. 

Add to that complexity that the CEO did a recent interview with Forbes where he admitted that he hasn’t got a clue to the copyright status of the 100 million odd images that Midjourney uses, so if that was ever traceable, you could have multiple stakes for ownership from anyone who created the images that Midjourney scrapes.

Thinking ahead

There are already loads of forums and artists springing up their own creative communities and Midjourney has its own community showcase where you can get a taste of what art is possible and the commands used to create it.

Obviously there will be a future for this, in the same way we are now seeing spectators and prize pots for egames events both in the millions.

Counter strike has paid out over 166 million dollars in prize money, and free fire has had over 5 million viewers.

And in the same way my parents’ generation would have scoffed at the idea you would be defined a sportsperson by “playing computers”, is there any truth to the title of artist for those that can create gorgeous artwork through a handful of commands to an AI tool?

Whatever side you come down on that argument, we need to embrace bringing AI’s speedy creative and constructive ideas into our businesses, and reframe “what are they taking from us?” to “what can they give us?”.

If the jury is still out on electronic artists, art and authenticity, then if nothing else use AI imagery practically, and consider it as a tool to create very quick sketches of compositions and ideas. 

This can have a myriad of benefits for how we communicate internally with our teams and externally with our customers.

Do you know anyone who may be interested in this project?

Click to share: