The Rise of A.I. Art
Unless you’ve been living under a rock, you’ve heard people praising or shitting on A.I. Art — artificial intelligence that creates images from textual descriptions. The first time you try it, it feels like magical sorcery. I type in ‘Simpsons in the style of Tim Burton’ and this is what happens:
It was amazing. My mind was blown. It opens so many creative possibilities — from concept art to matte paintings to character designing to fashion. It’s revolutionary. Months ago there were only a few options, but now there are dozens — with more popping up daily. Midjourney is my favorite, but DALL-E 2, Stable Diffusion, Deep Dream Generator, The Night Cafe, etc. Even Google has a AI image generator platform that looks really promising.
Within only a few months, the entire space quickly went from a nerdy hobby to the mainstream. There was an image that took the top prize at a state fair, the first comic book that was made of art that was generated by an AI, a horror AI glitch called Loab, and Paul Trillo used AI to do an FX-heavy fashion show.
I’ve heard from various artists about how troubling this new technology is — stating that it will put thousands of artists out of work. I personally don’t believe this will be the case. It will just be another tool in our tool chest as creatives and creators. I am obsessed with all the creative uses for this new technology — especially when we start using it to create moving images.
You can use AI to make your dreams real… like a new octopi species:
Or maybe you want to envision John Candy as Indiana Jones:
Or perhaps you want to visualize location imagery with the perfect moody sky:
The possibilities are truly endless. The more you play around with AI-generated art, the more addicted or the more frustrated you will become. It’s not always accurate and it does take time to get what is truly in your mind's eye.
Recently, I have been using Midjourney to conceptualize art for a deck I was creating for a feature film pitch. I would generate tons of art that would actually inspire the script we were working on. Locations, characters, make-up design — it was a creative tool to conceptualize things that were bouncing around in my head. The image above took a few dozen prompts to get something I was happy with.
But the greatest thing about AI-generated art is the happy accidents.
Believe it or not the prompt for this was nothing like the image it created — but I love it. “Accidents” are my favorite part of the process. Since prompts are just words describing an image, the AI can interpret it in a million different ways.
I tried to put in some text from various screenplays to see if it could generate anything close to the still from the movie. I chose Lost in Translation. In one of the opening scenes, the description was:
The back of a GIRL in pink underwear, she leans at a big
window, looking out over Tokyo.
This was the image it created — after a few redirections. Midjourney chose to add a painterly style to the image (something I didn’t ask for). So you have to keep giving it more information: nighttime, hotel room, blonde woman (not a girl), neon-soaked Tokyo.
Trying similar prompts to attempt something more photoreal and somehow the ‘pink shorts’ description tinted the entire art pink, once again not something I wanted. It’s a deep rabbit hole that keeps getting better or worse depending on how much direction you give it.
Who is the artist?
One of the more interesting debates happening right now is who the actual artist is in the creation of AI art. One side says that the prompter is the artist, while another side says the computer is the artist and the prompter is the art director.
It’s true that to get really beautiful, highly detailed, professional-looking art it does take quite a bit of work. Most recently, I found a few websites that let people sell prompts.
So I wondered, could anyone generate the same piece of artwork if they use the same exact prompt as someone else? I posted this to the group:
a woman with long white stringy hair, dark background, 50mm, beautifully detailed, dramatic, cinematic lighting, high contrast, 8k, photo realistic --ar 16:9 --testp --creative --upbeta --upbeta
Running the same prompt a few more times, I got a new woman, new lighting, new look every time. So did others. So I was educated about using the SEED command to try to replicate the image again. No luck. Here are 4 different tries, the bottom right being the original.
So if I can’t replicate the same image twice, who is the actual artist? There’s a huge amount of randomness that comes into play using AI to generate art. So much is put upon the machine to interpret what it thinks you mean. As one person pointed out on the FB group:
Diffusion works by starting with random noise and slowly bringing an image into view, kind of like a “memory” — hence even if we have the same exact prompt and the same seed used, it’s never the same.
So just like our memories, some things are accurate and some things are grossly embellished. For now, as good as some of my prompts have been, I have to give most of the credit to the AI in terms of being the artist. But just as I work with designers, storyboard artists, and cinematographers to create a film or a commercial, or a website, my role is simple the director. Creative director, art director, or simply “director” — we are asking a member of our team (the AI)to realize something by telling it what we want.
This is something I am very used to — as filmmaking or any large project is made by dozens if not hundreds of individuals. It is a very interesting dynamic when it’s just you and a non-sentient machine.
An interesting topic that is also coming up quite a bit is who owns the art you create. Technically, if you are a paid member of Midjourney, you own the artwork and you are free to do with it what you wish. I read an article recently about the 2.3 billion images that were used to train the AI. For the AI to understand the style of a specific artist, it has to be fed images (ie knowledge) to comprehend the command.
So, basically, it is emulating the specific styles of painters, the precise lighting of specific photographers, and the color palettes of famous imagery in order to illustrate your dreams. On one hand, it’s a very complex ethical conversation and on the other hand, it’s imitating and being influenced — just like other filmmakers, photographers, and designers do everything day. We are all inspired by other artists.
I just installed Stable Diffusion on my Mac. I have 64GB and can run it decently well. There is a good article on how to get it running on a Mac here. While Midjourney uses a Discord bot to take prompts, you can run SD locally, your images staying local and not in the cloud for ultimate privacy. I still prefer MJ, but it’s amazing that I will own images in SD or MJ (with a paid subscription).
If AI-generated images are the now, obviously video will come soon. What will AI-generated video look like?
Wide shot, outerspace, a massive space cruiser slowly pushes past the camera as a smaller ship is in pursuit.
Runway just teased a text prompt AI video generator and if it is anything like this, it will be a game-changer.
Generating photo-realistic videos at a variety of frame rates will be a massive drain on a processor — as just one frame takes a bit of time to render. Imagine 24 or 60 frames per second at least 2k or 4k?
I can imagine its usage for changing pre-existing video content. I searched for any examples of this and was blown away by this example I found on Reddit. Not sure if it’s real, but the concept is there. This would obviously be a massive game-changer for visual effects and open the possibilities for video editors to instantly change the content in videos they are editing in semi-real time. It’s huge.
Additionally, just as we saw the Bored Apes break out of the NFT space and become major IP assets, I predict we will see something similar come out of AI-generated images — and it won’t take years, it will take weeks or months.
It’s safe to say that AI-generated images are a genie that is not going to be put back in the bottle — it's here to stay. While some artists find this extremely scary, I have always loved to watch major technological advances transform industries — allowing for new things we haven’t dreamt up yet to happen.
Since this is a genie that we will not be putting back in the bottle, I’m excited about being a part of the community — playing, testing, learning, and experimenting.
I will leave you with an image of a unicorn riding a unicorn.
MORE IMAGES I’VE CREATED:
In full disclosure, I write and direct a lot of horror projects, so some images are a little disturbing. Enjoy.