AI Outtakes, Vol. 6

Hi and welcome to Volume 6 of AI Outtakes! In Volume 5 we looked at the curious case of Waldo, the blue-suited bearded man who shows up in AI-generated stock photos even when he's not wanted. In this episode we're focusing on the exact opposite - how to make sure you do get the same character showing up across different photos.

Building Your Brand Mascot

I'm willing to bet most readers can immediately name the brand that these actors are known for:

Pretty sure you were 3 for 3, maybe 2 for 3 if you're under age 35.

So how can your brand use generative AI to get a mascot?

Well if you want Waldo as a mascot, you're sort of in luck! But even in that case, it's not really the same character. There are variations; they just all look remarkably similar. What if you actually wanted to have a consistent character used throughout your marketing material?

Generative AI can do this! There's a few tricks though. Let's take a walk through the fascinating-but-sometimes-creepy world of AI brand mascots.

Consistent faces with generative AI

Finding a mascot you can consistently generate with AI starts with defining the characteristics of a person who would be a good mascot. If your first inclination is to start with a celebrity, you're in luck! Generative AI is packed full of celebrity images, and it's the one category of face it can reliably produce.

Setting aside the issue of just how the AI dataset acquired all these celebrity photos and whether there are copyright issues at play, there are a few hurdles to overcome. You can't just use a specific celebrity. Even if the images are AI-generated, implying they are affiliated with your brand is likely to get you sued. But there's a trick! Generative AI does a really good job of combining faces. So if you can describe your mascot as a combination of 2 or 3 celebrities, you can make an entirely new person and generate images of them in scenarios to fit whatever your marketing material needs!

Here's how it works

Start with a prompt that includes the two people you want and the scene you want to put the combined character into. For Orchestra, given our brand has deep ties to music, what about Yo-Yo Ma and Freddie Mercury?

yo yo ma, freddie mercury, photo of a middle-aged man, standing on stage in front of an orchestra, facing camera, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores

The key to making this work is to start your prompt with the two (or three!) names, followed by photo of a [description of the combined person]. You can then add a location, followed by any modifiers to control things like camera, depth of focus, etc.

The other thing you'll want to record is the seed. Generative AI uses seeds to add randomness to the output. Typically, each generated image uses a different random number to add variation to the output. But a specific seed can be provided to the AI tool, which guarantees it will use the same variations that it used in the original generated image.

Let's take a look at a few more combinations!

Maybe you would like to take advantage of two of the biggest blockbuster music tours of the past year:

taylor swift, lizzo, photo of a young woman, in front of a christmas tree, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores

One thing to beware of, however. The more images of a celebrity that are in the AI tool's dataset, the more it will 'weight' the image towards that person. So if you wanted, say, a photo of Taylor Swift and Jon Hamm dressed as a cowboy, this is what you're likely to get:

To balance this out a bit, most generative AI tools (like SDXL, which we're using here) have the ability to 'weight' different keywords in the prompt. Given Taylor Swift's fame, dropping her weight down a bit will make it a better blend.

(taylor swift:0.5), jon hamm, photo of a young person, dressed as a cowboy, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores

Once you've found your ideal celebrity mash-up, you can start inserting them into your stock photos!

(ed sheeran:0:5), daniel radcliffe, photo of a middle aged man, in an office meeting with coworkers, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores

I hope this was as informative as it was fun! If your company is looking for a brand mascot, Orchestra can help you find one as part of our digital marketing service! Contact us to learn how.

Until next time:

steve martin, martin short, photo of an older man, wearing a suit and waving goodbye, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores