Messing around with AI and content concepts

I’ve recently gained access to MidJourney, which uses AI to generate images based on text-based prompts.

You will have seen a lot of these on your Twitter feed, as the tools become increasingly sophisticated and available to more users.

Like many in our industry, I’ve been thinking about how to potentially create content with them. Especially as it enters it’s novelty phase, before it becomes saturated.

Inspired with the testing done by others in the industry, these are some of the hypothetical ‘campaigns’ I’ve imagined with the tool*.

*Caveat: I’m not an expert ‘prompt engineer’, I’m using the MidJourney guide and the Teapot guide. Join their discord to see some experts in action, and don’t take my word for how to use the below!

Literal Pasta Shapes

I have had a half-idea in my locker for a while that was realising pasta according to it’s literal name. I.E Farfelle (Butterflies), Fusilli (Rifles), and Linguine (Tongues).

Having tested this a bit with the AI, I found that it just looks like…the original pasta. Maybe an expert pastamaker or actual designer could make this premise work, but this was chalked up to a bad idea in the end. (Probably something I didn’t need the AI to tell me.)

Alien Vitruvian Men

The AI responds well to recognisable styles. Da Vinci’s Vitruvian Man fits that bill, and has been pastiched throughout history as a result.

Which is why this concept feels a lot more concrete: What if iconic aliens from pop culture were Vitruvian Men?

It could definitely be refined, but I like it as a fun premise. The subverting of a recognisable form with an unusual subsitute is probably something we could build future campaigns on. It can be translated to styles and topics, like ‘Mount Rushmore with LeBron James, Michael Jordan, Bill Russell and Kareem Abdul-Jabbar’. A safe area that wouldn’t cause much debate.

Modern Artists with Vintage Albums

A similar setup here. Recognisable person, recognisable era, unusual setting. In this instance Olivia Rodrigo, who I chose because her album cover for Sour is just her infront of a block colour background, in the 60s, 70s, 80s and 90s.

I think this works well enough, but again, because of struggles replicating exact versions of people’s faces, works better as a test of concept.

The way to enhance this is to add details with a real designer. Make the Parental Advisory badge in the bottom right of the 1990s. Put the 1980s onto a cassette tape. Add wear and tear and a sleeve to the 1960s, or unique details like a record label logo.

Similar to what was done to enhance the AI drawings on the travel posters by Stasher.

AI Idioms

I posted this concept to Twitter, where I asked the AI to create images of idioms.

This isn’t a new concept, it has been done by campaigns in the past: Idioms of the World, The Eyes Have It, and Expedia’s Language of Love are examples from Verve Search alone. This is just a different form of that campaign type, perhaps with added weight because it’s an AI realising something very human.

This is my favourite of all the suggestions, and the only one I think needed less human interference to make outreachable. In this case, the story is enhanced because it uses an AI, as it’s a computer's literal interpretation of human language. It’s not just using AI for the sake of it.

Final Thoughts

The more I’ve read about AI and how it learns, the more I feel uneasy with the process. Whilst using Leonardo Da Vinci or styles of design eras feels like fair game, its liberal use of modern artists work and styles without attribution treads into dicey territory.

For example, I tried some prompts where I asked it to create new Pokemon by combining hybrids of the first three generations starters (I.E Charmander + Cyndaquil + Torchic).

I thought the end result was cool, but the style was so clearly influenced by some of the Fakemon and ‘realistic Pokemon’ art that exists on Deviantart, evidenced by the fact in some cases it tried to create watermarks. Were this to be used for clients or campaigns, it would make me uncomfortable.

Secondly, there is always a temptation in our industry to think that the ‘new thing’ is a substitute for an actual good idea. Which several of these are definitely guilty of.

Right now where there is an element of novelty, it could still work, but that will end quite quickly. The production of the piece has to enhance or build on the premise. If it can’t stand on its own two feet without the AI hook or feels like it’s done for the sake of it, it will show.

Still, for now it’s definitely fun to play with, and can help create some good starting points, even if it’s to then help a human designer realise them properly.