How to use AI image generators for marketing
By Creative team. Top image: Dall-E 3, upscaled with Adobe Neural Filters, edited and saved as a GIF by a human designer.
Want to create stunning visuals by utilizing AI? Here’s what you need to know
1. It’s not as easy as it looks
Despite the AI hype, coaxing an on-brand, on-brief picture out of an image generator is a question of iteration not “abracadabra!”
Our process below will immediately improve your results but the machines aren’t about to make humans obsolete yet.
Three different image generator results produced from an identical prompt: "A happy influencer's selfie as if taken from their own phone."
2. Choosing the right image generator
You’ll find explainer posts proclaiming Dall-E 3 to be “best for ease-of-use,” and Midjourney “best for quality” etc but it’s really not that simple.
For one thing, AI is advancing at such a rapid rate that what’s true today isn’t necessarily true tomorrow.
But the reality is that you never know which image generator will respond best to a particular request.
The tool which produced a brilliant result for one brief, can often get the next one tragically wrong.
Your best bet is to always use at least three market-leading image generators.
Think of them as robo-picture agencies that are responding to your brief. Sometimes one will be bang on, while another is close but no cigar, and the third… comically bad.
Currently we typically turn to:
Leonardo.Ai
Adept at complex scenes.
Copes with text.
Reasonably good at people though they’re a bit too perfect.
DALL-E 3 (included with ChatGPT Plus)
Great at visualizing stylised objects or more abstract styles e.g. videogame aesthetics.
Super-simple interface.
Prompt helper adds detail to your ideas.
Ideogram
Excellent quality.
Very good at people and text.
Interesting ideas assisted by the magic prompt.
Adobe Firefly
Good quality and good at people.
Slick editing tools.
Adobe say Firefly is “designed to be safe for commercial use.” That’s because it’s trained on licensed Adobe content and copyright free imagery. Check out Adobe’s FAQ and terms for more.
Depending on the task, we also use Night Cafe, Yodayo, Sea Art, Pixlr, and Midjourney. It’s worth taking some time to play around with the various options to see which works best for your brand’s campaigns.
"Please create an abstract painting showing a robot painting a picture of itself" - Leonardo AI.
3. The art of prompting
Think of your prompt as your brief to the AI. Write it as if you’re talking to a colleague or assistant.
AI aren’t mind readers so they typically need more information from us than we realize - followed by a short bout of iteration to hone the result.
The good news is that you don’t need to worry about complex ‘prompt engineering’. Natural language inputs are the way to go.
However, your image generator does need certain specifics to hit your brief.
Include the following information in your image creation prompt:
What - As in what type of image do you need? Is it a photo? A fantasy poster? An anime illustration? A cute cartoon image?
Quality - Should the image look like the work of a world-class pro photographer or like it’s been snapped on a smartphone?
Subject - What should the image contain? What characters do you want to see? What’s the scene? Does it include particular objects or text? Be as specific as possible, tell the AI exactly what you want.
Context - Add in any extras. For example, what’s the mood of your characters, the setting, and the light? Do you want to see specific colors? How about particular clothing, a certain demographic, precise positioning, a distinct vibe, or action dynamic?
Prompt action
Here’s an example prompt featuring the different elements:
Create an informal professional photo of a diverse group of friends in their 20s, having fun together in the park on a sunny day. Aspect ratio 16:9.
And here’s the prompt again broken down into its key parts:
Key:
WHAT - QUALITY - SUBJECT - CONTEXT
Create an informal professional photo of a diverse group of friends in their 20s, having fun together in the park on a sunny day. Aspect ratio 16:9.
Color-coded like this, you can see that each aspect of the prompt is playing an important role in building the brief that your AI will work from.
In fact, it’s a good idea to think of yourself as a creative director who’s telling a photographer or illustrator exactly what you want.
There’s no magic formula here. We simply use the structure to help us think through our ideas and clearly communicate what we’re looking for. It’s all too easy, especially under pressure, to forget to explain what’s in our heads.
Now feed your prompt into your chosen image generators - remember you want at least three pitching their ideas back to you.
"Create an informal professional photo of a diverse group of friends in their 20s, having fun together in the park on a sunny day. Aspect ratio 16:9" - Leonardo AI.
4. Hone your result
This is the most important stage of all. The AI probably won’t get it right first time. But you can iterate your way to success.
It’s time to play with your prompt.
Think of this as feeding back to your agency:
You may want to add in extra guidance. For example, an addition to our earlier prompt could be: “The friends are playing frisbee.”
Or you could ask the AI to expand on an unexpected idea it’s introduced. For example, “Show more of the boating lake.”
Perhaps you’ll want to take something away: “Remove other people from the background.”
Or change the direction: “Show the friends walking side by side,” as opposed to the staged group shot result you got initially.
In fact, that’s the advantage of using multiple image generators: you’ll get interesting takes on your brief that you didn’t think of initially and can work up.
This stage need only take a few tries and may well be more successful with the paid packages than the freebies.
While your initial prompt may only contain a couple of lines, your final prompt could easily be 100 words or so as you layer up the detail.
That approach tends to work well. Sometimes a very complex initial prompt can knock your AI for a loop.
Other times you don’t need to change the prompt but just “reroll” the AI dice to see if it does better on a second try.
"A kitten reaching out to touch Planet Earth" - Adobe Firefly.
5. Finishing touches
It can also be useful to feed your AI some examples of the kind of thing you’re looking for. Hunting around for inspiration is also a great way to develop your own ideas for your original prompt.
If you want to generate believable, UGC type pictures with an everyday vibe then try prompting with phrases like:
Disposable camera picture
iPhone selfie
Photo uploaded to Facebook
Sometimes it helps to tell certain AIs to ask you questions before working on your brief.
You can also instruct it to, “Explain your thinking.” That’s useful if you need an underlying rationale to explain your choice to other stakeholders.
6. Image rights
This is a very gray area which is why, on behalf of our clients, we use AI images for internal ideation but not for public distribution.
Some AI models have been trained on copyrighted works and that may entail risk.
Some AI services claim they are trained purely on licensed or open source content but even then please check your in-house rights policy before publicly posting any images. Also review your chosen AI image generator’s terms of service policy and apply the appropriate credit for use of material created by their tools.
Unfortunately, we cannot provide specific legal advice here on the use of AI-generated images - especially as the law is playing catch-up with this issue.
However, Dialect always secures appropriate rights for all images that we produce on your behalf.
7. Managing expectations
AI can only get you so far. As humans ourselves, we’re pleased to report that the squishy, organic element can’t be eliminated yet.
You still need to bring your own creativity, judgment, and experience to the party to achieve a knockout result.
If your needs are very precise then at some point you’ll run out of road with AI.
It will often get you close but not close enough. And they won’t be replacing your high-end photoshoots anytime soon.
Sometimes you will see amazing work out there that’s billed as “AI generated”.
In fact, the best results always feature a human who knows how to craft a result from several different tools. For example, they’ll generate a picture from a specialized AI app, apply specific textures from another program, outpaint in Midjourney, and upscale in Photoshop.
Have fun playing with the possibilities of AI yourself, and get in touch if you’d like to know how we make the most of it.
"A 1960s dancer doing the twist on the surface of Mars" - Leonardo AI.
If you would like to discover more about our integrated approach and work together on a project, get in touch.
