Four years ago we were tinkering around with GANs (Generative Adversarial Networks) which are a form of AI that could generate never-before-seen images from a batch of sample images. We fed our GAN some old Halloween masks and asked it to generate some new ones. This is what it made us...
...pretty creepy right? Especially if you're afraid of pixels. And while those are definitely nightmare fuel they are nowhere near the level of what today's state-of-the-art diffusion models can create.
Fresh on the scene diffusion models like DALL-E 2 and StableDiffusion are trained on billions (yes BILLIONS Dr Evil) of images which gives them the magical ability to generate "beautiful" images from a short and sweet text prompt like "Highly detailed wide angle portrait of a creepy demon clown, photo, surrealistic, creepy, dark, epic, cinematic, style of Atget, detailed."
Yes, that whole thing was generated by an AI. You know, those tech things they keep telling us we shouldn't be afraid of. Yet, here we are!
With Great Power
So, now that you obviously want to make your own haunted AI images you can head over to Stability AI, or if you're on an M1 Mac download DiffusionBee and prepare to never sleep again. If you're looking for some prompt inspo just head over to Lexica to browse prompts and their outputs to use for your own creations.
Brainstorming Killer Klowns from Outerspace not quite your thing? Well, with AI image generators you can brainstorm basically anything. ANYTHING. Let's say you're a cookie company looking for Scary cookie ideas. Just change the text prompt above from "demon clown" to "demon cookie" and you'll get a whole new set of nightmares.
"Highly detailed wide angle portrait of a creepy demon cookie, photo, surrealistic, creepy, dark, epic, cinematic, style of Atget, detailed"
Let's also say you are a shoe company and you want to generate 10,000 fresh new shoes designs but just don't have the time? This is what AI is great at, doing lots of things fast! It works out great too because humans are great at responding to visual prompts. It's one thing to stare at a blank screen and design a new shoe but it's a whole other thing to be presented with endless variations of things and then drilling in on the ones that appeal to you as a said human.
So Matt, does AI just generate old, dark, and scary pictures to give humans nightmares? Nope, it can generate happy joyful things just as easily. Swap out those dark and scary words in the text prompt for bright, cheerful, and Pixar and you get an entirely different set of options. I especially like the double-stuffed-high-high-heel-tops ones myself.
So now what?
These AIs and their kin will be built-in to everything eventually, Photoshop, TikTok, YouTube, Product Design, 3D Modeling, Game Dev, Filmmaking, Advertising, Coffee Mugs, everything. And while Text-to-Image AI can currently generate the fun still images you see above, Text-to-Video is being worked on by Google, Meta, and others. AI breakthroughs in the Creative industry are pretty much happening on a daily basis now.
Prompt engineering will become an art of its own and imaginations will exponentially run wild. AI is so perfect for brainstorming that we've actually built our own in-house creative team member we call "Createatron", it dreams up never-before-seen ideas and is a fun time, promise. If you're in Nashville swing by our office and hang with Createatron over some GooGoo clusters and Hot Chicken why don't ya? Disclaimer: just don't travel within 24 hours of ingesting Hot Chicken 🔥.