The world is getting smarter – and we like smart technology, after all, it makes life easier. We control devices with our voices, our fridges can tell us how many calories we’re consuming and some cars can drive themselves (almost). With such a shift in everyday life we hardly notice it when Google predicts what we might type next.
Marketers are already using AI to identify and segment audiences, test or improve advertising performance, and optimise spend – all automatically, in realtime, and at scale.
Now AI offers the promise of being able to create web content, bespoke logos and imagery from a series of simple input prompts. So, could AI be set to take over the marketing world?
And is it going to strike fear into the hearts of copywriters and creatives, leaving them in fear for their livelihoods?
Is it game set and match to AI?
Mads Berg is a renowned Danish Illustrator who was hired by creative agency Space, to create the visual style that defined the Wimbledon Championships 2022.
Mads’ work also inspired advertising agency 10 Days to see if they could create an image capable of competing with the original, but rather than briefing an illustrator, their imagery would be created using AI image generation software.
The agency trialled Midjourney, typing the word ‘imagine’ followed by six genre-based word prompts like ‘sci-fi’, ‘noir’ or ‘cinematic’. In less than 60 seconds four unique images were created. 10 Days reimagined and refined the Wimbledon themed results a total of 24 times, and these are the results:
It’s the speed that is quite remarkable – taking 60 seconds to create a set of four unique images – that got us really excited.Jolyon White - Creative Director, 10 Days
Incredibly, the entire process took just five minutes from start to finish, and every image is one of a kind, existing nowhere else online.
Ok, so the images aren’t poster-ready yet, but the implications of AI for the industry can’t be underestimated. Some of us have been in the industry long enough to remember the move from paste-ups to the Apple Mac – this feels just as seismic a shift.
We were excited, and decided we had to give Midjourney a go here at Michon. Could it really produce in seconds the image our designers had in their heads, and which would take illustrators and retouchers hours to achieve?
Sticking to the tennis theme we decided to try and create an image using the prompts: ‘Serena Williams, tennis, mix Wonderwoman, poster.’ Less than 30 seconds later we were faced with these images:
No wonder the creative industries and software developers are sitting up and taking notice. Adobe – who created Photoshop, the gold standard photo manipulation software used by agencies across the world – is already trialling a beta version of Photoshop with a built-in backdrop creation neural filter, which is essentially an AI image generation filter.
However, before we all get too excited, AI image generation isn’t perfect.
We also put in the prompts; ‘weasel mammal, riding a green woodpecker’ in a bid to recreate an original photo taken by Martin Le-May (see below left). It was an unlikely image, and AI produced some unpredictable results:
Ok, so not quite what we had in mind, but the process took around 10 seconds and Midjourney does allow you to generate a new set of images based on your preferred image from those previously created.
The technology doesn’t always work, but it’s early days, and the more people that use these platforms, the more AI will learn how to create better images. Unsettling, yes, but the very crux of AI is to learn and improve. For instance, Alexa and Siri remember our likes and learn from them, improving the quality of their responses based on our instructions.
Why we aren’t scared for our jobs – yet
Perhaps creative designers, retouchers and art directors should see AI and sites like Midjourney and Dall-E2, not as a threat, but as a new and exciting tool to speed up the ideation process. The creative possibilities are endless and these new tools will enable the fast production of images based on abstract themes and moodboards.
Want to see what an iPhone 20 might look like? AI might just have the answer. Want to see a pair of ‘back to the future style trainers, flying in 80s street’, then you can – and we did – in less than 40 seconds. Our creatives’ minds were blown. We also asked for non-dairy, fun, bright, milk images. Again, not quite what we expected, but isn’t that the point. The results could be a jumping off point in the creative process in a totally different direction.
The results were not always reliable however. When asked to create a kaleidoscope image with ladies in swimwear and flowers on a white background, we may not have got the image we wanted. (We were aiming for something more like the bodyshop poster on the left) but AI’s literal interpretations of the brief are stunning, and tell us that there may be a learning curve to get the best from AI.
So, what’s the down side?
AI is only as good as the data it finds on the web. That means it’s subject to our web user’s unconscious and conscious bias. If most people looking for a handsome man image select a young, white bodybuilder, then that is what AI will believe to be the definition of handsome.
The way AI learns has led to a frightening discovery
Steph Swanson, a Twitter user known as @Supercomposite, was experimenting with AI image generation using negative prompt weights. The technique should produce the theoretical opposite of whatever you ask for. Her request was for the opposite of Marlon Brando (“Brando::-1”).
The result was a logo of a transylvanian castle, but when Steph requested the opposite of that logo, what she got wasn’t Marlon Brando. She got four images of the same woman, who appeared with the letters “LOAB”, and so that became her name.
When Steph tried to mix the image of the woman with other images the macabre results were so graphic she won’t release many of them.
When the images were ‘reimagined’ until the woman disappeared, she would later reappear, always in the same location: a house with brownish-green walls and filled with junk. This is odd because if we put in the same prompts for Serena Williams again, we don’t get the same images (we tried), AI generated images appear to be unique each time – except in the case of LOAB. (Source: abc.net.au/news)
Some say Loab is a warning of the threat of AI, some say it’s a glitch, but this old, poor, beaten up old woman is worrying, not just because she tells us a lot about modern beauty standards but because AI learns from us, and we’re not always kind – which leads to some very valid concerns about the misuse of AI.
AI misuse and copyright infringement
AI could be used to generate violent, adult or political content. Both Dall-E2 and Midjourney state that they’re taking preventative measures to stop images being generated that can spark hatred or abuse, which also means training their AI to ‘not’ produce photorealistic images of real individual’s faces.
These are scary possibilities for the misuse of AI and ones we hadn’t considered, but it seems that even the most benign artwork could be in muddy waters when it comes to legal ownership.
Most computer-generated images are protected by copyright in the UK. Unlike other countries, the UK does give copyright to computer-generated works which do not have a human creator. Which sounds great, but before we all go off to create a new image for our next campaign we need to consider legal ownership, or copyright.
Many of these image generation platforms are still in the Beta, or testing phase, which means that by default the images created are shared with a community, who can also use and remix them. Paid subscriptions will allow for more control in the future – but who owns the rights to your creation?
These platforms take images found online to train the AI and to create new images, so it’s theoretically possible that an image, or part of a generated image could be recognisably the same as a pre-existing image. If so, it would infringe that artwork’s copyright. It’s something that’s being hotly debated and it’s difficult to predict the effect of AI on copyright law.
So is AI set to take over the world of marketing?
No. Not just yet and until things change, we’d not recommend using AI to create or use in any client work, but perhaps it could serve a purpose as a springboard for that initial spark of inspiration.