I Used AI To Build An Image Library. It Delighted And Frightened Me.
Last week I was planning the creation of an image library for a client who needs a ton of visuals every month.
One of my principles is: never build a brand on stock if you can avoid it.
But as I started compiling our wishlists of photos and illustrations, my enthusiasm waned. The time it would take to create this was overwhelming. Time, and of course, money. Two things that don’t come in abundance in the social impact space.
At that exact moment, an email arrived saying I had been approved to try out DALL•E, one of the first AI image generators out there.
It seems robots can already read my mind.
I had been on the waitlist for a while. So I thought: why not test if I could build the image library with AI? I wasn’t expecting it to be a viable option for years to come.
What I was able to generate through AI astonished me. It was immediately clear.
This changes everything.
It’s astonishing. It’s frightening. And we should talk about both.
Can we already use it?
What would make it better?
Should we use it?
What are the pros and cons?
How might we make AI-generated imagery fair?
At first, it seemed too big to write about. But we must discuss the pros and cons and the bigger implications together.
So let’s go.
What are AI image generators?
DALL·E is an Artificial Intelligence system that can create images from a written description.
DALL•E is created by a company called OpenAI. It’s one of several AI-based image generators that are currently live, like Stable Diffusion or Mid Journey.
You enter a short description of the image you want to create. An armchair in the shape of an avocado. A purple furry monster in a dark room. Within seconds, there it is, no visual skills required.
DALL•E users were posting their results and they were impressive. This particular AI also appeared the most ethical of the options out there (more on this later), so I used DALL•E for my experiment.
My experiment
I wanted to run a test to see if I could create a series of images that were both high quality (as if made by a talented human) and on-brand (a particular look and feel that we are going for).
First, I used keywords to describe the scenes we would need for our content and campaigns.
For instance: “African female entrepreneur working on her laptop”.
Then, I added a style description: photo, oil painting, pixel style, in the style of Miro/Matisse/Vermeer, 3d rendering, or digital art.
As the images popped onto my screen, I felt two big emotions rushing at me at once. It was amazing. It was frightening.
I showed two colleagues. Jaws dropped. As it all sank in, one colleague said: “I don’t know if I want to live in a world where this is possible.”
Some style filters and keywords worked well, others not so much. Still.
I had expected AI to be years away from being a viable option. But at least half of what was generated was usable.
And so the machines have come for us. Not in twenty or thirty years. But today.
Is AI imagery usable?
The photo-realistic images, though impressive, differed in quality per query. The faces of the male entrepreneurs were blurry and twisted. But the virtual women generated by AI were indistinguishable from the real thing.
The 3D renderings were downright awful. With this query, the men came out better than the women.
The illustrations were much more promising. I did not like the illustration styles per se —most were totally off-brand for us. However, I was very impressed with how well DALL•E was able to produce different images in a particular style consistently.
With the technology as it stands today, I could build a small set of images, but it doesn’t yet outperform custom imagery, let alone stock.
But it feels like we’re 90% there. Which surprised me.
What would make it more valuable?
I see AI complementing the creative team only when we can create imagery unique to us.
I would want to train an AI system to create imagery based on our own brand’s unique style parameters.
Current AI generators allow you to create imagery in someone else’s style. That’s no good for a brand. I want to create imagery unique to us. So I would want the ability to work with AI developers to set our own visual style parameters.
My instruction for AI would be, for example: “A woman working on her laptop in [our brand’s] style”. Our ability to generate visuals for content and campaigns that are unique to us would explode.
And within one set of brand parameters, we could define unique sub-styles per campaign.
Should we use it?
Now that we’ve established that, yes, we CAN use it — the question is: SHOULD we? Here are some of my first gut feel pros and cons.
The potential positives
When I saw the images pop up on my screen, the benefits appear to be clear immediately.
For the downsides, hang on a second —those will follow right after.
More people will be able to create imagery.
AI opens up high-quality visual output for those who are not visually gifted. In the same way that Adobe, and later Canva, democratized creative industry tools, AI image generators will do the same, but at a far more fundamental level. At first, it will generate gimmicky stuff. Yes, it’s cool that DALL•E can create a 3D rendering of an armchair shaped like an avocado, but who really needs that? Quickly after, people will find ways to use it more purposefully, for better or for worse.
It’s a game-changer in terms of efficiency.
The logistic side of creation is considerate. Selecting, contracting, shoots, location booking, model release forms, post-production…
No more endless searches for the right creatives or models available at the right time and place. AI models and locations don’t require booking, contracts, or travel. With AI-generated imagery, you skip over logistics and go straight to creation.
With greater efficiency, prices will come down.
Creating and training an AI to generate visuals in our own brand style will most likely be a big upfront expense, but after that initial cost, the cost per image would lower than it is right now.
For creatives, it’s just another playground.
In the hands of great creatives, the gimmicky aspect of AI can be overcome, and everything still depends on a good concept and good execution. AI is just one means to an end.
The potential negatives
The negatives raised by my experiment were equally obvious. And also deeply frightening.
For every positive I mentioned above, there is a tremendous, unpredictable, and potentially devastating negative. Technologist and writer Andy Baio was spot on when he called it a Pandora’s Box.
AI image generators represent the perfect storm of ethical dilemmas.
Human jobs and livelihoods, intellectual property and copyright, social justice, privacy, and more, are at stake.
I am not good at predicting the future. If my opinions on online shoe shopping and selfies in the 2000s were recorded and played back today, I’d be a laughing stock (I said it was all a fad).
But the past 20+ years have taught me something about the progression of tech.
Like most tech, AI comes with tradeoffs, and those are as unpredictable as they are massive.
What’s at stake are jobs, money, and our humanity.
Let’s start with the last.
The previously uniquely human endeavor of visual creativity is at stake.
AI is often described as ‘highly autonomous systems that outperform humans at most economically valuable work’.
It doesn’t outperform us purely on its own, though.
AI uses the whole body of human visual creation available on the internet to train itself. Technologist and writer Andy Baio calls this “laundering human creativity”. He asks four fundamental questions:
- Is it ethical to train an AI on a huge corpus of copyrighted creative work, without permission or attribution?
- Is it ethical to allow people to generate new work in the styles of photographers, illustrators, and designers without compensating them?
- Is it ethical to charge money for that service, built on the work of others?
- Is it even legal?
For this article, I want to focus specifically on the impact of AI on the brand space, but that doesn’t mean I think it’s wise to ignore the greater societal impact.
If you want to dive deeper into the dangers of AI to society, Baio unpacks pandora’s box of AI-generated imagery clearly and urgently: deep fakes, copyright infringement, privacy issues, and non-consensual intimate media.
And of course, jobs will disappear.
We know what happens when robots move in. More and more jobs get sucked into the automation machine, and many won’t be replaced by the new markets and services that will emerge.
The creative space can already be a tough route to a decent living. But it was always considered safe from automation.
My experience with DALL•E showed me that the mid and top tier of the creative sector is no longer safe from automation.
Of course, the ‘democratization’ and disruption of the creative industries are not new. The rise of platforms like Squarespace, Canva, Noun Project, and stock photo sites did just that. And they created efficiencies without killing the creative industries.
Those efficiencies made certain roles at the bottom of the creative industry superfluous. But they also lower prices, making more high-quality creative assets available to more people. It’s hard to imagine the blogger, creator or influencer space would have existed without these tools.
But AI doesn’t nibble: it swallows whole.
The disruption that AI will bring to the creative industries is unfathomable. This is democratization and disruption at hyperspeed. It will affect designers photographers, illustrators first and foremost, and all the creative jobs connected to those.
I suspect the first to be struck by the rise of AI-generated imagery are not individual creatives, but the free and low-cost stock photo and illustration sites.
When AI can generate unique imagery that equals or passes the minimum quality baseline of generic stock, why bother with it?
The creatives currently feeding those stock sites will be hit hard by extension — if these platforms were ever a sustainable source of income for them.
But surely a counter-movement for things ‘human-crafted’ will emerge?
When everyone zigs, a percentage of people will always zag. Custom bespoke work will be commissioned by those who consciously want something human-crafted. It should appear human crafted because that’s what people will want to show off.
But let’s be clear.
This tiny niche won’t fill the gaping hole left by AI.
Big Tech will get even bigger and more powerful.
When creative industry jobs disappear due to AI, money will flow to tech companies instead.
The first AI image generators run on a freemium or subscription model. They are the stock sites, illustrators, photographers, and designers of the future. And in a way, that isn’t a bad thing. AI running on a Facebook-like business model that is free to use but makes its money through data extraction is too frightening to comprehend right now.
But…
the value of the AI business model simply cannot only accrue to tech companies alone — it needs to be shared with the creatives it depends on.
I want to argue for the survival of human creativity purely for its own sake, without needing to resort to arguments of its economic value. But I know this argument never works with tech evangelists. It gets a predictable retort:
Should we have kept piano tuners and cotton weavers in a job forever?
“You can’t stop progress” is an end-of-the-road argument to halt all discussions of the ethics of tech and its impact on human society.
And no, you can’t keep people in jobs that no one needs. But maybe it shouldn’t get that far.
Current inequalities could be exacerbated.
My experiment left me with the question if AI will exacerbate racism, gender issues, and global equality.
Our global digital image body is built by those with access to digital media. That creates a geographic and economic divide. The global north decides what the global south looks like online.
Brand managers like me are already responsible for the underrepresentation and misrepresentation of people across income levels, gender, ethnicity, religion or sexual orientation.
The white gaze (or in non-European/NorthAmerican countries, the upper-class gaze) in brand image libraries is deeply problematic. Brands depict our hopes and aspirations. Who we cast as the heroes of the stories we tell, and what we require them to look like, is deeply biased.
Because AI trains on existing body of images, the human biases already existing could bias the future output.
The DALL•E team seems deeply aware of these ethical issues, but other AI companies seem less concerned.
How might we make AI-generated imagery fair?
We need policy makers and legislators to step in quickly
Currently, AI trains on the global body of human imagery online without asking the creators for permission, let alone compensation.
In a world where intellectual property rights and copyright protects creators, this seems highly dubious.
I’d ask:
Shouldn’t human creatives be asked to opt-IN to their images being used to train AI?
And if so, shouldn’t they be rewarded for their role in automation?
Recent history tells us however that you can’t expect these companies to do the right thing on their own without pressure from citizens and action from lawmakers.
The perfect storm of legal issues and the complexity of a global permission and reward system will delay the public and government response.
A delay that tech companies will use to roll out their product further until we all depend on it, and alternatives are no longer available.
The sooner we put pressure on tech companies and legislators to do this ethically — the better.
There will be new AI-related creative roles.
Is there a win-win for creatives in the brand space? AI-generated imagery is here to stay, and this is only the beginning.
We can’t just sit and wait for how this is going to shape the creative industries. We have to start thinking of ways that human creatives can harness this technology.
As a brand director, I imagine it will be my role to oversee the training of an AI.
It will be like creating brand guidelines on speed. You can’t (yet) get AI to develop a unique set of brand principles, so this will remain the domain of humans.
I imagine working with a creative team that is amazing at its craft, as well as skilled at the translation to an AI learning system.
In the boutique design space, there are already fashion designers designing purely digital fashion, venues, public spaces and products, but that’s in a way a digital version of the same craftsmanship.
AI takes that up a level.
We will need people who are highly skilled at painting a picture with words.
It’s not our ability to craft the scenes we envision, but our ability to describe the best scenes.
Visual arts will become verbal arts.
The winner of the Colorado State Fair’s annual art competition that caused such a ruckus last week, Jason M. Allen, is an example of this shift of skills.
Creatives will need new pricing strategies.
I envision there would be the need for an economic model on two levels: both for the passive use and active use of human creativity.
Passive use is the AI learning from the entire body of human creative work.
For active use of human creativity, creatives will need new ways of pricing their work. Some of these might be entirely new, and some could build on existing models like royalties.
Imagine the style of a particular creative (or team) is used to create an AI system or an AI filter. Remember the ‘oil painting of an entrepreneur in the style of Matisse’ from my experiment? Let’s say I’ve trained an AI system to generate work in my style: ‘Line drawing of an entrepreneur in the style of Anne Miltenburg’.
Instead of traditional pricing per illustration or per hour, I should now charge for the value of the AI system’s ability to endlessly generate images in my style.
This could be an upfront fee, or some sort of loyalty or licensing agreement. For instance, a royalty per item created by their system, or by the period of use.
There is a caveat though. Business models like Spotify’s have proven that only the highest performing global artists would be able to generate significant returns on this.
Agents and associations of creatives should be hard at work experimenting with pricing models now.
The only thing that’s certain is that we will be conflicted.
DALL•E clearly offers an entirely new playground for brands.
As a brand director, it’s a big part of my role to create amazing creative work that’s on brand throughout the year. AI could be a huge asset for me.
But creative collaborations with humans are our life’s blood and a source of joy. And there is no denying that for all the positives of AI, huge negatives will come in to play — and humans, not AI, will suffer.
I already said I’m not a good predictor of the future. But I am 100% sure I will spend the next decade being deeply morally conflicted about the options we face.
Are you a creative professional in the brand space?
Tell me what you think and feel about AI-generated imagery. What excites you? What frightens you? How do you think it will impact you? Leave a comment below and it’s possible I will get in touch to ask you some follow up questions for a next article.
More on building brands for change
Brand the Change, the guidebook — your guide to building brands for change, with 13 case studies, 22 tools, the anatomy of a strong brand explained, and 5 guest essays by brand experts with tips and tricks on everything from trademarking to PR.
Brand Thinking Canvas — our go-to-tool to create strong brand frameworks
Brand the Change Community — Join our support network and let’s help each other to build the brands the world so desperately needs.
The newsletter — feed your brand-thinking brain with regular updates and bite-sized knowledge.
My TED Talk: How branding can help accelerate social change.
