AI Artists are CTO's (Timing your AI Art Projects)
My most famous AI Art Piece/DALL-E Prompt to date is probably A Photo of a Confused Grizzly Bear in Calculus Class:
This piece went viral, has been recreated by others, referenced in a few industry research papers, and so much more! I’m so proud and delighted to see it grow.
I made this in 2022 just a few weeks after DALL-E 2 came out. But, how successful would it have been if I had made it in DALL-E 1 (which never got released)? At the same time, back then, which prompts did I enter other than this one that would have been better off with a future version, say, DALL-E 3? How much of my own success do I actually owe to the superior capabilities and overall timing of the DALL-E 2 model release?
For comparison, I just ran the exact same prompt in an earlier 2021 model (roughly around DALL-E 1) and here’s what it came up with after 200 iterations:
… if I had used a primitive model, would I have dismissed this prompt or the potential of the concept altogether? Would I have shared it with the world? Even if I did, do you honestly think the world would have taken the confused grizzly bear in calculus class seriously?
The Lesson
I think there’s a lesson here for everybody. The truth is, there’s probably some balance between trying really hard and learning a new AI art tool/model when it comes out and simply waiting until the AI models get better year over year (or lately it feels like month over month). It may not be fully worth exerting all of your energy (and likely picking up bad/legacy prompt design practices) into a primitive model early on. Obviously, it’s still worth getting some practice though using AI models and writing prompts. However, you really do have to see it as a longer term game and spread yourself out over a few generations and developments of the technology.
At times this idea seems obvious, but it’s easy not to follow it in practice because all of the creative AI art tools can just be so flashy, fun, and there’s a lot of FOMO. It takes discipline, pragmatism, patience, and a sense of vision to keep some great ideas in your backpocket until the AI model technology catches up and the timing is just right. You have to be willing to let go and believe in the overall exponential progress in AI like we’ve already seen up to this point.
Put Some Ideas on Hold … For Now
Do you have AI art ideas you are holding out on until the technology gets better? I certainly do. A lot of my best ideas I’m aching to make are simply not even possible with today’s technology. I basically have postponed my GPT-3 and Identity series because I want to reboot it as a very thoughtful and detailed animation series. Since I cannot afford an animation department of my own, I have to make it with a future AI video model once it is released.
It’s not to say it would be impossible to make the series with today’s technology - I suspect you could get a lot done especially if you’re willing to put in a fair amount of human elbow grease to compensate. This means late nights and maybe combining hand techniques and photoshop editing/3D modeling to get the high quality results you want. However, in my case, I’m holding out because I actually want the series to be great. I want to make things the world hasn’t seen before using tools that give me much greater command over my vision than the AI models of today. I want to make things on my own at the scale that would require hundreds of people today to make.
I also wonder, what details and suggestions would future AI models include that today’s best models would miss? How much more successful could future tools make my project?
Artists are Strategic Technologists
In my view, Artists may need to become more like CTO’s (Chief Technology Officers). This means, just like other considerations like budget, time, the merits of the project, tools, typography, team, or concept, AI artists should bring in projected technological developments in mind too as an additional design factor consideration.
How should you think about this? Should you put your project on hiatus until future models are released? I’ve simplified a lot about how I might think about this problem in this 2 by 2 matrix:
For many art and creative projects, I suspect you may not have any resources or budget at all. It also depends on your attachment and conviction to your idea and its potential. You may be struggling and prefer to be more lean, iterative, and use AI to put your ideas out quickly instead. You may even,“find a way” with clever prompts to realize your vision with today’s existing AI models. I want you to know that all of this is ok.
But you may have some ideas which you can feel today’s AI creative tools just aren’t ready for. If this is the case, in the meantime, it probably makes sense to chart out anticipated AI model capabilities, keep track of updates to tools, practice, understand existing limitations, and do background research/concept development on an idea you might already have and are saving for the future. Then, once a model capable enough is released, to strike with conviction while the iron is hot and immediately work on the project if the technology is ready.
Company CTO’s have to find a sweet spot between where the technology is today, the work they are going to put in in the meantime, and what will be possible optimistically speaking as technology exponentially improves a few years from now. It’s a strategic and a very intuitive thing, but is a necessary part of the fast paced world we live in.
Along with planning their creative projects and areas of focus in advance, I’m saying that I believe it actually makes sense for artists to make strategic technical bets and execute them on a longer, exponential time horizon, just like a CTO would. Moore’s law and AI scaling laws apply to artists too. I think artists and creatives should save some of their best ideas and should not settle for any less.
The industry may tell you to start immediately, other artists may pressure you to jump in or miss out entirely, but I’m here to tell you not to waste your best ideas. For some creative projects, it may make pragmatic sense to hold off until the timing is right. Newer AI models and tools will allow you to give your great idea the level of craft and attention it deserves. They may dramatically increase the odds of your success and give you control and capabilities you could not have dreamed of even a year ago. The most impactful thing you can do nowadays as an AI artist is learn to time your best ideas accordingly.