5 Tips for Optimizing AI-Powered Art Direction

Sep 1 •

5 Tips for Optimizing AI-Powered Art Direction

Article URL copied!

Written by a human.Edited by an AI.

Supply’s Executive Creative Director, Sam Jeibmann, was recently tasked with creating an art direction strategy for articles featured on our website. He saw this as the perfect opportunity to dive into the creative sandbox and have fun with a new tool our studio has been excited to experiment with —generative AI.

While defining an art direction strategy for Supply’s articles, Sam also uncovered a few best practices for establishing an art direction that leverages generative AI. We thought we’d share them with you.

1. Identify your metaphor.

Distilling a message down into a simple metaphor can be immensely helpful when creating content with generative AI because a metaphor provides a visual shorthand that will encapsulate your core message and theme. We like to use a simple framework we call SEL (Subject, Emotion, Landscape) to help us with this.

Subject
What is the central theme, concept, or message you are trying to convey? What takeaway would you like your audience to walk away with?

Emotion
What feelings and emotions are you hoping to elicit in your viewer? Do you want them to feel joy, excitement, nervousness, optimism, etc.?

Landscape
What does the contextual landscape of your narrative look like? Are there adjacent or supporting concepts that might add visual interest or help create tension in a composition?

An example metaphor that we created for another article.

2. Develop a mood board.

Mood boards are crucial in creating art direction because they help communicate and define a project’s visual and emotional direction. They are a source of inspiration and a guiding framework for a creative team. Setting out to create AI-generated content without a mood board can quickly lead to predictable and uninspiring results. While AI can generate art based on various algorithms and data, it lacks the intrinsic understanding of human emotions, intentions, and context that a mood board can provide.

Sample from the mood board Sam created as part of his art direction. (Source: 1. Andrei Cojocaru, 2-4. Unknown, 5. Frank Moth)

3. Create a color palette.

Colors have psychological associations and can evoke specific emotions. Defining a color palette allows you to set the emotional tone —this isn’t something you want to leave to chance when creating art with AI.

When you define a subject in a prompt, you are, by default, effectively locking in a color palette based on what the average version of that “thing” looks like on the internet. You can add prompt language to tune the color space from there. Keywords like “cool tones,” “warm tones,” “crushed black,” “sunset-inspired hues,” or film keywords like “Kodachrome” will change the overall tonal space of the image. You can also affect the subject by adding more obvious color modifiers such as “blue banana.” From there, it’s best to make final adjustments in posts using Photoshop or other editing tools to tweak your colors.

4. Plan your composition.

Too often, people focus all their attention on the subject and not on the environment around their subject when creating content with generative AI. A composition plan is like a roadmap for generative AI’s creative process. It gives the AI clear directions, purpose, and a sense of order. Without one, you risk producing content that misses the mark and is less effective. Take time to consider the space around your subject carefully. What elements are off to the sides? Above or below? Behind and in front?

Establishing a spatial composition through a well-crafted prompt is helpful when composing a photographic image with generative AI. When designing an illustration, you might be better served by splitting your final composition’s foreground, middle ground, and background into separate prompts and then stitching them together later in post to achieve greater control.

5. Define keywords and attributes.

Generative AI systems, particularly those used for content creation, often rely on the keywords and attributes within prompts to guide their creative process. These keywords and attributes help the AI understand the context, treatment, and content requirements for generating the desired output. When writing prompts, carefully considering the keywords and attributes you include will enhance the quality and relevance of the content being produced. Here are a few examples of Keywords and attributes we use.

“Photo” renders the image based on three main characteristics of digital or film cameras: the lens, the sensor or film size, and the film or digital file. This keyword can be augmented by including specific attributes. For example: “Photo, Nikon 7D, 85mm lens.”

“Studio” renders the image as if it were on a stage or production studio. This keyword can be augmented by including specific light attributes and background colors. For example: “studio, grey background, left and right softbox, rear key light.”

“Hyperrealism” or “Hyperrealistic” renders the image in an artistic style made famous by painters in the early 2000s. This keyword generates a near photo-real image but is stylized to create slightly more organic shapes and vivid colors. This keyword can be augmented by including other artistic media attributes such as “Oil,” “gouache,” or “colored pencil.” You can use any number of artistic styles and corresponding attributes to achieve your desired effect.

Cover art examples created by our design team in collaboration with generative AI for Supply’s articles.

Closing Thoughts

Generative AI is more powerful than ever, but it is essential to remember that it’s still a tool. A hammer alone can’t build a house. A carefully constructed art direction remains a critical part of the equation, and generic AI art directions will produce generic results. Until a time comes when we can quickly and easily build AI models that understand the nuances of individual brands and that have the social-emotional intelligence required to anticipate how an audience might react to various visual triggers, a solid art direction along with a plan for generating quality content is still needed. Hopefully, these best practices can help you create better results the next time you leverage generative AI in your creative practice.

Share this article

Article URL copied!

Supply's Executive Creative Director, Sam Jeibmann, was recently tasked with creating an art direction strategy for articles featured on our website. He saw this as the perfect opportunity to dive into the creative sandbox and have fun with a new tool our studio has been excited to experiment with —generative AI.While defining an art direction strategy for Supply's articles, Sam also uncovered a few best practices for establishing an art direction that leverages generative AI. We thought we'd share them with you. Distilling a message down into a simple metaphor can be immensely helpful when creating content with generative AI because a metaphor provides a visual shorthand that will encapsulate your core message and theme. We like to use a simple framework we call SEL (Subject, Emotion, Landscape) to help us with this. SubjectWhat is the central theme, concept, or message you are trying to convey? What takeaway would you like your audience to walk away with?EmotionWhat feelings and emotions are you hoping to elicit in your viewer? Do you want them to feel joy, excitement, nervousness, optimism, etc.?LandscapeWhat does the contextual landscape of your narrative look like? Are there adjacent or supporting concepts that might add visual interest or help create tension in a composition? An example metaphor that we created for another article. Mood boards are crucial in creating art direction because they help communicate and define a project's visual and emotional direction. They are a source of inspiration and a guiding framework for a creative team. Setting out to create AI-generated content without a mood board can quickly lead to predictable and uninspiring results. While AI can generate art based on various algorithms and data, it lacks the intrinsic understanding of human emotions, intentions, and context that a mood board can provide. Sample from the mood board Sam created as part of his art direction. (Source: 1. Andrei Cojocaru, 2-4. Unknown, 5. Frank Moth) Colors have psychological associations and can evoke specific emotions. Defining a color palette allows you to set the emotional tone —this isn't something you want to leave to chance when creating art with AI.When you define a subject in a prompt, you are, by default, effectively locking in a color palette based on what the average version of that "thing" looks like on the internet. You can add prompt language to tune the color space from there. Keywords like "cool tones," "warm tones," "crushed black," "sunset-inspired hues," or film keywords like "Kodachrome" will change the overall tonal space of the image. You can also affect the subject by adding more obvious color modifiers such as "blue banana." From there, it's best to make final adjustments in posts using Photoshop or other editing tools to tweak your colors. Too often, people focus all their attention on the subject and not on the environment around their subject when creating content with generative AI. A composition plan is like a roadmap for generative AI's creative process. It gives the AI clear directions, purpose, and a sense of order. Without one, you risk producing content that misses the mark and is less effective. Take time to consider the space around your subject carefully. What elements are off to the sides? Above or below? Behind and in front?Establishing a spatial composition through a well-crafted prompt is helpful when composing a photographic image with generative AI. When designing an illustration, you might be better served by splitting your final composition's foreground, middle ground, and background into separate prompts and then stitching them together later in post to achieve greater control. Generative AI systems, particularly those used for content creation, often rely on the keywords and attributes within prompts to guide their creative process. These keywords and attributes help the AI understand the context, treatment, and content requirements for generating the desired output. When writing prompts, carefully considering the keywords and attributes you include will enhance the quality and relevance of the content being produced. Here are a few examples of Keywords and attributes we use."Photo" renders the image based on three main characteristics of digital or film cameras: the lens, the sensor or film size, and the film or digital file. This keyword can be augmented by including specific attributes. For example: "Photo, Nikon 7D, 85mm lens.""Studio" renders the image as if it were on a stage or production studio. This keyword can be augmented by including specific light attributes and background colors. For example: "studio, grey background, left and right softbox, rear key light.""Hyperrealism" or "Hyperrealistic" renders the image in an artistic style made famous by painters in the early 2000s. This keyword generates a near photo-real image but is stylized to create slightly more organic shapes and vivid colors. This keyword can be augmented by including other artistic media attributes such as "Oil," "gouache," or "colored pencil." You can use any number of artistic styles and corresponding attributes to achieve your desired effect. Cover art examples created by our design team in collaboration with generative AI for Supply's articles. Generative AI is more powerful than ever, but it is essential to remember that it's still a tool. A hammer alone can't build a house. A carefully constructed art direction remains a critical part of the equation, and generic AI art directions will produce generic results. Until a time comes when we can quickly and easily build AI models that understand the nuances of individual brands and that have the social-emotional intelligence required to anticipate how an audience might react to various visual triggers, a solid art direction along with a plan for generating quality content is still needed. Hopefully, these best practices can help you create better results the next time you leverage generative AI in your creative practice.

Jun 9 •

Help Us Get Good Stuff In The Box

Our work creating digital products and services always starts with a discovery phase.

Our work creating digital products and services always starts with a discovery phase. But our clients often want to truncate discovery and get right to the making part. I kinda get it. Especially when we tell them we want to conduct our own interviews of team stakeholders, and we want to do our own user research. It takes time and costs money, and for clients who have already conducted research, it can feel like they’re buying the same horse twice. Of course we’ll take whatever they have—“Bury us!” we say—but we still insist on identifying for ourselves the inputs that will inform whatever recommendations we’ll eventually make. I call it “getting stuff in the box.” It’s a reference to that scene from the movie Apollo 13, where they realize the C02 filters are failing, and they need to rebuild the extra square ones on the command module so they fit in a round receptacle. “I suggest you gentlemen invent a way to put a square peg in a round hole.” (So good.) Anyhow, in the next scene there’s a bunch of NASA guys dumping a huge box of random stuff on the table with a mandate to “make this, fit into the hole for this, with nothing but that.” Discovery is how we get the stuff in the box that we’re going to need to solve your problem. Don’t let “nothing but that” characterize the stuff our team has to work with when it comes time to tackle your biggest challenges. Because here’s the thing: those NASA guys would no doubt have done their level best with a nearly empty box of sad bric-a-brac. But how great is it that they had duct tape, and a spool of wire, and a space suit, and some pliers, and a hose? Likewise, we’ll devise a solution with whatever information and insights are available to us. But as long as we’re in the phase where we’re putting stuff in the box, let’s assemble a rich profusion of insights and ideas, quotes and context, so when it’s time to get to work, we have inspiring material to work with. Credit to Steven Johnson, who first pointed me to this inspiring scene in his excellent book "Where Do Ideas Come From: The Natural History of Innovation". I read it 12 years ago and think about it once a week.

Jun 8 •

Turtle Excluder Device

There used to be a whole industry around tourists helping newly hatched loggerhead turtles get to the ocean.

I was 37 when I started working in design strategy, old enough to know not to shout out “Turtle Excluder Device!” as the meeting wrapped up. Most people avoided my gaze and walked away. I managed to catch the sleeve of one colleague gathering up his things. “No, really, you have to hear this.” (It was like that moment in The Rhyme of the Ancient Mariner, which begins with a bedraggled uninvited “grey haired loon” accosting guests at a wedding so he could tell a longish ghost story about why you shouldn’t slay an albatross. In my tale, you’re the wedding guest, I’m the loon.) In college Bio we learned that there used to be a whole industry around tourists helping newly hatched loggerhead turtles get to the ocean. So many hatchlings were getting run over while crossing roads on their way to the sea! So many were being eaten by gulls! If we could only save more baby turtles, the thinking went, their populations would rebound. But then a young graduate student named Deborah Crouse conducted a population modeling exercise for her PhD thesis, and discovered that no matter how many hatchlings were saved, the number of turtles laying eggs wasn’t increasing. And breeding pairs are what you need to make more turtles (I’ll cover turtle sex in a separate post.) Her work uncovered the real problem: that turtles weren’t surviving adolescence because when they got big enough, they’d get trapped in fishermen’s nets. This insight eventually led to the invention of metal hatches called Turtle Excluder Devices (TEDs) that were built into fishing nets. When a sufficiently heavy turtle gets caught in a net, it springs the hatch and the turtle swims free. Today TEDs are often mandatory, and the turtle populations are far healthier as a result. What lessons in product strategy can companies learn from the Turtle Excluder Device? Fast forward almost 20 years to me blurting out, “Turtle Excluder Device!” We’d helped our client design V1 of their telematics platform, after which people came out of the woodwork with bright ideas about what features they'd like to see next. Sales, Service, Maintenance, Product Management—everyone had thoughts about what they’d like to see in V2. Our challenge was to find a way to shepherd these ideas far enough along that they could be honestly evaluated, and potentially cross the gauntlet of doubt about their feasibility, likelihood of adoption and cost. So we created a workshop that incubated numerous ideas, deeply considering how they’d come to life and be experienced over time. And we had a large swath of critical stakeholders participate in shaping each idea, which helped because, well, everyone likes to have a say. The output we created was a marketing story for each—a website—that made it look like the idea was complete, built and launched. We shared these mock marketing websites with customers, other users, and senior leaders who eventually aligned on which ideas to fund and which to pass on. We created a Turtle Excluder Device—for ideas! Since then, I’ve realized that many companies have problems that could be solved by some version of a metaphorical TED. Maybe their strategy isn’t surviving the design phase because it was thrown over the wall from some long gone consultant. Or their design intent is getting lost in development, because engineers never had a chance to weigh in on its feasibility. Or the finished product isn’t surviving launch because Marketing isn’t clear on its true benefits. In each case, what’s needed is a process that can shepherd great ideas to fruition, and keep them from perishing in corporate environments that are weirdly effective at killing them off. That’s what we do at Supply. We help our clients succeed not just through expert digital design and development, but by helping them efficiently and effectively set up teams and processes to improve the parts of the product development lifecycle that need it most. Sometimes, that means building a Turtle Excluder Device. (Big thank you to my bio teacher Dr. Paulette Bierzychudek at Lewis and Clark College, who told us a good story about great science that I’ll never forget.)