Looking to spin forward its 2021 Heinz campaign asking consumers to "Draw Ketchup," Canadian agency Rethink decided to try artificial intelligence. Within seconds, AI text-to-image tool DALL-E 2 generated dozens of images responding to prompts such as "ketchup bottle on table” or “impressionist style painting of a bottle of ketchup.”
How ad agencies are using AI image generators—and how they could be used in the future
The images, which bore a strong resemblance to the iconic Heinz brand, were then used in ads this July that featured the tagline, “This is what Ketchup Looks Like to A.I.” The brand even asked consumers on social media for suggestions on what prompts to ask DALL-E 2 to draw next, which led to more images that were featured in a metaverse art gallery.
Rethink is just one of the many agencies using DALL-E and other AI tools such as Midjourney and Stable Diffusion, all of which have the potential to change how ads are made. Beyond creating compelling content, agencies are using these tools to save time and money as well as brainstorm ideas. The tools, which all launched within the past year, potentially open up new horizons for agencies in the future, such as custom ads created for individuals, a new way to create special effects, or even making e-commerce advertising more efficient.
"We are at the very beginning of a revolution for our creative industry,” David Raichman, executive creative director, social and digital at Ogilvy Paris. “AI represents an incredible potential that impacts the way we conceive, design, produce and do justice to realize the idea's fullest potential."
However, while these image generators have plenty of upsides, there are still issues to consider such as questions around copyrights, the potential for bias and concerns over what reliance on AI might mean for creatives in the future.
Finding the 'unpredictable'
Dentsu Creative’s Portugal team recently launched a campaign using Midjourney to create abstract images promoting the European electronic music festival Jardim Sonoro.
Gil Correia, a creative director at Dentsu Creative Portugal, said that using the image tools was simple. The team spent a few weeks learning how to use Midjourney before spending about three days on the project.
The agency was looking for something “unpredictable” that could match the festival, said Correia, who explained it was important to match images with the various artists being promoted.
“We adjusted—word by word—until we managed to make the images link with each other, even if they were completely different,” Correia said. “But most interesting of all is that if we put the same sentences again, the result would always be different. And that’s a good thing.”
The campaign ran through print, digital and out-of-home.
TBWA/Melbourne this month created promotional art for the Melbourne Writers Festival by adding passages from classic literature, including "1984" and "Moby Dick," into MidJourney. The output became the inspiration for out-of-home and digital ads promoting the festival. There are also plans to release illustrated ebooks of the classic novels, featuring further AI interpretations, in the coming months.
In August, OpenAI launched a new DALL-E feature called “Outpainting” that allows users to create an extended version of a preexisting image—for example, expanding the background of the "Mona Lisa."
Ogilvy Paris launched an ad for Nestle brand La Laitière that altered its logo, which is based on “The MilkMaid" painting by Dutch artist Johannes Vermeer. The final result was an ad that expands the painting, illustrating a new image that shows the rest of the room.
While there are multiple examples of work that used the AI image tools, almost every agency executive that Ad Age spoke with agreed that these image generators can also be used to supplement the creative process.
Brainstorming and saving money
Executives from agencies such as R/GA, Dentsu Creative, TBWA, Huge, Accenture Song and Wunderman Thompson emphasized the benefits of using these programs as part of brainstorming sessions. Natalie Comins, a group creative director at Huge, said it's useful for creatives to view these tools as a “creative sparring partner” to access new ideas.
“After a briefing, some of our creatives are spending an hour or two throwing concepts at DALL-E—be it an early line of thought, an oddball question or an impractical art direction notion we could never afford—just to see how it responds," said John Doyle, executive VP, brand experience strategy at Colle McVoy. “The result is a sort of serendipitous Rorschach test that almost always surprises the team and sometimes sparks new thinking that inspires actual concepts.”
Using these tools can also save time when creating mood boards or presentations for clients, according to Comins, who says that can often take two to four hours searching through Google or Getty for images.
Wunderman Thompson recently used DALL-E to create a photorealistic image of dogs playing poker for a client presentation, said Jason Carmel, the agency’s global lead of creative data, who added that it saved his team hours of work.
“I’d be lying if I said that we haven’t, on many an occasion, used a DALL-E or Stable Diffusion to create a rendering for a presentation slide where stock photography or image search failed us,” Carmel said.
Not only do agency executives see the tool as a time-saver, but it can also be cost-effective because these tools can generate multiple variations of one prompt within minutes.
Paul Caiozzo, co-founder of Supernatural, an agency that uses AI to inform its strategy, says it recently used image generation to “animate and colorize” a film. “The motivation was financial, but the time saved was certainly a bonus," Caiozzo said. “We didn’t have the budget for stock or to illustrate and animate or shoot live action.”
Buying vs. investing
The difference between buying a stock image versus investing in a platform like DALL-E is significant, said Hikari Senju, founder and CEO of Omneky, an advertising software company that uses AI technology to create personalized digital ads.
“If you think about DALL-E, for example, it is $15 for 115 credits and each credit gives you four generated images. So that means each generated image is like a fraction of a cent or something like that versus the cost of a stock image that’s maybe $10 or $20,” Senju said. “If you have a subscription, then maybe it's slightly cheaper, but still probably more expensive and limited.”
Omneky, which has around 100 clients and launched in 2018, uses a combination of AI software that generates both content and copy to make ads that at times feature AI-generated humans such as the ones below.
Future use cases
Although many agencies are still in the test-and-learn phase with these tools, that hasn’t stopped creative directors from imagining the possibilities they bring. But the tools are not all the same.
Caiozzo says he prefers Stable Diffusion because its images are more “nuanced and natural" and because it is available as open-source, meaning the code underlying the AI and the model it is trained on are publicly available.
This is different from OpenAI, the organization that owns DALL-E 2, which doesn’t allow access to its algorithm and also has stricter filters. Until this week, OpenAI didn’t allow the generation of images showing real people. OpenAI, though, has a policy against users uploading images of others without their consent.
Stable Diffusion and Midjourney have the option of adding “not safe for work filters,” said Rick Barber, chief technology officer of Addition, an AI company that works with brands and agencies. Addition recently helped Droga5 create a campaign for the New York Times that turns headlines from each subscriber's reading history into a visual description of that person.
Addition’s Founder and CEO Paul Aaron said what excites him about the future use of the technology is creating custom “generative” advertising that is different for each person. This is a concept most of the people Ad Age spoke to for this article viewed as a possibility in the future.
“It's conceivable that we go from browsing to landing on a product detail page for a product and the assets and imagery are completely unique to you,” Derek Fridman, a design partner at Work&Co said. “Let's say a person comes in from Google, they love the color blue, and then when you get to a home site, you see blue cabinets. It’s the idea of generating an experience that’s super bespoke per person.”
Fridman also added that the tools can be particularly helpful for making e-commerce projects more efficient.
Simplified special effects
Another example of a future use case is using these tools to simplify the process of making special effects.
Paul Trillo, a director who has directed commercials for brands including T-Mobile, Nokia, Carvana and Microsoft, and also directs music videos for artists such as Bas and The Shins, has started experimenting with Dall-E 2 since he gained access to the platform in June. He has posted multiple videos on Instagram, showing how to create special effects with these tools that have garnered thousands of views. One video includes a woman’s outfit changing 100 times.
Trillo said he has directed similar scenes where 3D clothes were added onto the person digitally. The scene could also be created where the person would have to continuously change their clothes and a shot be taken of them each time. Both cases are time-consuming, according to Trillo.
"You would shoot the same shot over and over again…you would do an overlay of the previous take…[and] there are still inconsistencies and you’re limited to what's available to purchase in the real world or what you can make as a costume,” Trillo said. “If you took the whole CG approach, you would need to design every single outfit, you would need to approve the designs for each outfit. You could even end up drawing up concepts that don't get used and then you would have to model, texture, light it, and composite it together,” Trillo added.
Since posting his videos Trillo says he has received multiple requests about integrating AI into advertising videos.
Limitations and legal hurdles
And given that the AI tools are informed by preexisting data and images, the potential for bias to appear is likely. “There's a general risk of being unbalanced, not representative, not relevant, and potentially even offensive,” said Alex Naressi, global head of R&D at Accenture Song.
“If you go to these platforms and you're like, 'Show me a man walking into a pet store,' it's usually a white guy,” said Dave Meeker, head of design and innovation at Dentsu Creative. “Why? Because it saw lots of images when it was trained and there were more white guys in those images. We have to overcome that; we have to make it neutral. Before we can use these tools in our day-to-day for deliverable commercial work, we have to overcome those barriers.”
Earlier this month, a piece of AI-generated work won first place prize at the Colorado State Fair’s annual art competition, which emphasized the big question that comes with these tools: Who owns the work?
“DALL-E and others are pulling from Google Images, Dribble, Behance and other designer foundries to generate 'original' compositions,” said Micky Ogando president and chief creative officer of Austin-based agency Bakery. “The problem is when the base of the composition is made up in its majority by a very real, very human artist's work. An artist who doesn't get named, credited, paid or acknowledged in this process."
Because the AI-generated images are based on preexisting images, ownership will have to be decided legally, according to Bradford Newman, who leads the North America machine learning and AI practice at global law firm Baker McKenzie.
“There's the argument that if I type in ‘Paris’ and it gives me images of baguettes and wine and cute little cobblestone streets and the Eiffel Tower, that's one thing,” Newman said. “But if there’s a world-renowned artist with his own style and I train the algorithm on that artwork so that someone like me or a member of the public using the algorithm can type in that person’s name and get a near copy of their artwork but they haven't consented to any of this, we're talking about legal issues that are going to be decided.”
Newman says the complication increases when the tool is being used for a commercial purpose.
“If I am an ad agency retained by the world's most famous sneaker company and I use one of these AI generators and I type in ‘competitor shoe that we're launching against,’ and it gives me some variation of that competitor's logo and brand and shoe look, there are going to be legitimate questions over whether this is fair use. The issue is like the early days of [music] streaming and MP3s…should the original artist be entitled to some cut?”
This question of ownership will make brands take “extra precaution” moving forward, Newman predicts.
Some content platforms, such as Getty Images, have banned AI content due to potential copyright issues.
AI as an additive
Despite the advancements in the technology, all executives that Ad Age spoke to for this article agreed that AI tools should be used to assist creatives, not replace them.
“There will always be a need for and value in creative direction, human curation, human refinement of an idea, and decision-making in terms of what is right for the brand we’re working with,” said Ben Williams, chief creative experience officer of TBWA/Worldwide. “Because of that, it’s the combination of creatives leveraging these technologies where the real opportunity lies.”
“It’s not AI that will take your job, it’s the other creative who knows how to use AI that will take your job,” added Stephan Pretorius, global chief technology officer at WPP.
Pretorius said not using these image generators now might not have an impact in the “short term,” but are still important for creatives to learn about.
“Today it’s largely still a novelty; by 2025 most creative teams will use them as standard practice," Pretorius said. "But if you haven’t developed a fluency with these tools by 2030 you will probably be at a significant disadvantage.”