Blog

With the growing popularity of using AI (Artificial Intelligence) in everyday life, we wanted to highlight some things to look for when selecting generated images to use in your financial advertisements!

When generating images, you may or may not notice some pretty peculiar results. AI often struggles with human details such as hands, and some facial features. Details surrounding your main subject can also generate very strangely. We would recommend doing an overall look at the images you are selecting, to ensure you are choosing quality images to best represent your institution.

Getting Good Results

To get the best results for what you need, it’s a good idea to be as descriptive as possible with your prompts. Describe your subject in detail as much as you can. Here are some example questions you can think about as you create or even revise your prompts:

• How is the subject shaped or sized?
• What color is it?
• What kind of textures or patterns does it have?
• Are the surroundings important, and if so, what does it look like?
• Should the image have a certain artistic style to it? (EX: a watercolor painting)
• Should there be a specific mood or emotion?

Alternatively, you could also leave your prompts open-ended to allow the AI to have more ideas to work with. This can give unexpected results, but occasionally can still produce some great images!

If you don’t care for the results you’re seeing, you can either regenerate the prompt for a new batch of results, or adjust your prompt to be a bit more specific. And if you see something you like in one batch of results, download it before you hit that generate button again! You’re unlikely to get that image back as you continue on.

Catching Subtle Mishaps

For my examples below, I’m creating my AI prompted images with Adobe Firefly. Copyright is a big looming controversy with AI-generated images at the moment. It’s essentially the wild west out there right now as creators discover their work being used as source without their expressed permission. Adobe ensures that their AI-generated images are sourced from content in their existing stock catalog and are commercially safe to use. So we advise using Adobe’s AI tools to hopefully avoid potential legal issues that may arise in the future.

Prompt #1

“mother and child adding money to a pink piggy bank”

At first glance, this image looks pretty great! But let’s take a look at some of the details:

• If you look closer at each of their hands, they both appear to have six fingers – including their thumbs which are obscured.
• The coins they’re dropping into the piggy bank look odd. It looks as though they’re trying to stuff three coins down the slot.
• The perspective of the child’s resting arm and hand looks a bit small.
• Though it’s a very small detail, the mother’s ear piercing is ill-placed.

Prompt #2

“couple moving a large red couch into their new home”

Now, this prompt turned out to be a tricky one. I kept getting results where the couple looked like they were moving a child-sized couch, even when I added the word “large”… so I had to run it several times to land on something sort of decent. This looks ok, but there are still some oddities here:

• It looks as though Firefly was trying to put the woman in high heels. But it didn’t quite get the style right!
• There’s a very odd doorway or closet generated into the corner of the room.
• The way the two people are holding onto the couch doesn’t look very normal. What exactly are they gripping onto?

Prompt #3

“red haired man in casual clothing driving a new white car; wide view of car”

Of the three prompts for my examples, I think this one was the most interesting. I had wide-ranging results each time I clicked generate. Some pretty funny even! But I had several that actually looked MOSTLY decent! Including this one here. But unfortunately, there are still some issues:

• The striped pattern on the man’s shirt is distorted in several areas. The hemming of the V-neck also looks incomplete.
• Though it’s subtle, I do see some issues with the shape of the man’s nose.
• His hand over the steering wheel doesn’t look quite right. Especially the thumb.
• There are various interior issues with the vehicle itself: the passenger seat, the dash and turn signal, and the headrests on both sides do not match.

Final Thoughts

As you can see, AI-generated images are still a bit off from being perfect. There will be flaws, whether they’re glaring or miniscule details. While it is a fun tool to play around with, it still might be a good idea to continue browsing actual stock photography while AI-generated images continue to be improved upon.