NB: This post was written in late July 2022, about a technology that changes daily – as do opinions about the use of that technology.
At date of publishing, conversations are starting about both the legitimacy of AI generated art in competitions, and whether automatically sourced imagery is inherently plagiaristic. These are huge subjects in themselves, so just something to think about. For now I’ve added some additional article links at the end of this post. Enjoy…
It’s Summer 2022 on the internet: NFTs are crashing, The Metaverse is low-res and empty, and Kate Bush controls social media. So where does an out-of-touch designer go for a new-tech creative fix online?
If you’re up for it, we can take a look together. It’s a bit of a long read this one, so strap in.
You, I and AI
Like ‘The Metaverse’, I’m often using ‘AI’ as a loose term to mean ‘computers doing stuff’. Instagram’s uncontrollable algorithm, the automated Amazon refund system, even your Spotify daily playlists – a version of AI infiltrates your daily existence on the internet in some way. But we here at M&C Saatchi London are more interested in creativity, so today I’ll be focussing on Artificial Intelligence in image generation – and whether we’re all about to be out of a job.
Over the last few months, you’ve most likely seen the output of Dall•E Mini popping up on your social feed and group chats. As with any new internet toy, and the opportunity to make weird jokes – public creativity has been unbound. From Mr Blobby in the Bayeux Tapestry, through to courtroom sketches of the titular Alien antagonist, the results have been equal parts unsettling and entertaining.
The speed and accessibility of the tool is unbelievable – even at this limited level, and Open AI are promising a huge amount more from the full-blown DALL-E 2. Ultimately though, outside of a few examples, the mini version available to the public, is mainly in use as an experimental meme maker.
Creativity in the hands of Chatbots (not the ones that fooled your uncle out of £500)
After a couple of weeks of AI-based silliness, new dreamlike landscapes and psychedelic character designs started taking over my social media feed. According to the tags and @s, Midjourney Beta had arrived, and it was clear that many creatives had found a new, more serious tool, with potential way beyond making funny pictures. NB: it’s also really great for making funny pictures.
Midjourney – in their own words, are “an independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species”. In my words, it’s a chatbot that lets anyone become an Art Director.
I scrabbled for an invite to the Beta and was soon on my way.
Midjourney works through an AI bot on Discord servers (for the over 30s, Discord is a modern, more focussed version of a chatroom). All it needs from you is to type an ‘imagine’ command (for example ‘/imagine prompt: ‘London City’). About a minute later – the Midjourney Bot has created 4 unique low-res images.
From these, you can ‘Re-roll’ for 4 more variants, ‘Version’ to iterate upon one variant, or ‘Upscale’ a variant to high-res detail.
Version control
At this point in the blog, you’ve seen how it all works – so in the spirit of Midjourney, here’s four options for you:
V1 – You’ve made it this far, why not finish the post? You’re probably getting paid whilst reading this, and there’s some very interesting links at the bottom.
V2 – If you’re too excited by the prospect of making cool stuff, and don’t want to read anymore – a trial version of Midjourney is now open to everyone! Sign yourself up here and let me know what you create.
V3 – Love the idea, but don’t fancy doing it yourself? Check out the Midjourney gallery of what people are making.
Then come back here and finish the blog.
V4 – Just stop reading and forget about the whole thing. It’s not for everyone. Thank you for your time. Here’s a fun video, made using AI frame generation.
A journey further into Midjourney
Thanks very much for taking the red pill. From here, we’ll take a very brief look at experimenting with imagine prompt inputs, then I’ll wrap up and provide some jumping off points for learning more and getting involved with creative AI.
At time of writing, we’ve just lived through the hottest days ever in the UK.
Full of inspiration and gratitude, I want to see how 15 minutes in Midjourney can be used to create an artistic monument to the humble item that kept me alive for the last week, the oft-ignored Bag of Ice.
Statues seem like a decent place to start. No-one does monuments like the ancient Greeks, and they loved a statue.
Kind of cool, but not really doing it for me. The AI is really struggling with where the hands should go, and there’s lots of weird shapes being made – this could be down to most reference of statues being of humans, or my prompt is too vague and there is confusion about what the hands or supposed to be holding. At this stage I could keep iterating or try to refine my prompt – but for now we’ll move on.
Still in the mood for a classical Ode to Ice, I turn to oil paintings.
This rules! I love how grand the whole thing feels. But the rays and orange are feeling a bit…urinal?
The holy vibes (and the Mr Blobby Dall•E Mini output) have given me a new idea though – it’s tapestry time.
Yeah, this is feeling way closer to a work of art. A few extra variants, a click of the upscale button and…
It’s beautiful. We did it! A tribute to ice – ready to hang in a museum, done in about 20 minutes.
So, time to reskill?
In short, no – you’re probably alright, for now.
Midjourney is an incredible tool for idea generation, and all signs point to Dall•E 2 and Imagen being the same. I never could have depicted, or even thought up some of these results on my own, plus the happy accidents and sparks that came from each iteration often lead me to another idea, and another imagine prompt. Without the restrictions of figuring out how on earth I’d create the image, it was a pleasure to just generate new things from pure creative thought.
But, for me at least – Midjourney creates starters. Images that are brilliant for mood boards and reference, providing the sort of thing you could use to brief a human 3D artist, photographer, or illustrator when the stock libraries can’t capture the feel of what you need. The chat with the AI is ultimately one-sided (every Art Directors dream when dealing with a designer), with no real rounds of amends. Whilst you can keep iterating and hope for the best – there’s no “can this bit be moved here”, or “make that section green”. To Midjourney, I’m just another client who ‘will know it when I see it’.
In my other experiments, I found that there was no personal satisfaction in sharing just what the AI had made for me. It wasn’t until images had been cut out, cleaned up and designed into something new, that I could lay claim to any sort of artistic involvement. And even then, I was just making a collage of Rat Statuettes…
There’s also the issue with a lot of the output looking very Midjourney-ish. The mix of uncanny valley and upscaled estimations of objects often skew towards images that either feel very unsettling (see the above rat abominations) or have that Midjourney fantasy-art house style.
Thankfully – the house style is beautiful, and we all want to keep our jobs, so the positives far outweigh the negatives here. And it really can’t be overstated how very, very early days it is. All these Artificial Intelligence systems are still being trained and are only getting better by the day, with more render options and more accurate outcomes coming with every update – new ’stylize’ and ‘quality’ commands were added to Midjourney just yesterday, at time of writing.
Go away and make something
Midjourney is now free to use on a trial version, for everyone – you just need to sign up. I heartily encourage you to try it out for yourself.
Get in quick though, the server is now more popular than both the official Fortnite and Minecraft Discords and is nearing 1 million users!
As you’ve seen in my experiments above, the difference in how you word your prompts can make huge changes to what the AI creates. I’ll let you discover what works best, but some great additions to mix up your prompt are ‘Octane Render’, ‘Pixar’, ‘Vaporwave’ and ‘Realist’. For more inspiration check out the General Image Gen channels on the discord to see what people are adding to provoke certain responses in imagery.
By Alan Clarkson
The Great AI Linkdump
Somehow, even at the end of this very long post – what we’ve looked at today is just the tip of the iceberg (from the perspective of an idiot pretending to understand it).
If you’ve found it at all interesting, here’s some further viewing/reading on AI in the world of creativity.
How AI Image generation works
Vox – The AI that creates any picture you want, explained
Next Level AI creation – Follow Barney McCann on Instagram
Barney.Mcann – London at Night
AI putting designers out of work (but not really)
Cosmopolitan – The World’s Smartest Artificial Intelligence Just Made Its First Magazine Cover
AI putting retouchers out of work (but not really)
Adobe – Neural Filters (Adobe Sensei) in Photoshop
AI putting texture libraries out of work (really)
Curtis Holt – The INSANE Potential of AI Art! (MidJourney & Blender)
Creating animation using AI generated frames
Joel Haver – How I Animated This Video
Inherent plagiarism in AI generated art
EnGadget – Is DALL-E’s art borrowed or stolen?
The ethics of winning art prizes with AI imagery
Washington Post – He used AI to win a fine-arts competition. Was it cheating?
(Appreciate your time on this one. If you make anything cool in Midjourney, or have any thoughts about all this stuff then pop me an email, or if you’re in the office – come over for a natter (bring chocolate)).
