Another year!
Last year was an incredible year for artificial intelligence (AI). If the rise of the internet was an inflection point, the rise of artificial intelligence (in my opinion) is an even more massive one.
I think what I’m most excited about is that sometime in the near future, everyone will have the ability and the access to build anything digital. It could be websites, movies/films, songs, mobile apps, comic books. You just need a device (phone / laptop / tablet), internet, and creativity.
In the 90s, with the rise of the internet, this started to become possible, but you still had to run to Barnes and Noble to buy books on how to code. The most motivated quick learners reaped the benefits of the internet in the first inning. You also often needed servers that filled closets and needed to handle the maintenance of all this yourself. It was not a super accessible period of time — not anyone could do this — but it laid the foundation for where we are today.
Fast forward to today, you don’t need to know how to code to build digital things. There are so many no code tools available to take the heavy lifting off of building, and you can learn about all of this with simple video tutorials on YouTube. There is so much that has already been made accessible to the billions of people on this planet. I learn a ton from watching YouTube videos — there is no better educator than YouTube (in my opinion).
Yet, despite these advances, there are some things that historically have still been hard in the last decade or so. For those who have no talent — like me — design, for example, is still hard. I can read and watch tutorials all day, and I still will not be able to draw well to save my life. If I practice for years, I could probably get better, but I won’t be able to come up with cool images or videos in the next couple of months.
Until last year.
Enter Midjourney and so many others. Last year, I thought I could finally fulfill my childhood dream of becoming a cartoonist to draw comics about my favorite stuffed animals I grew up with — all with zero talent.
But, is it hype or is it real?
Meet Warren Hippo that I created with Midjourney
My childhood stuffed animal Warren Hippo should be at least 30+ years old by now, but he is somehow forever 5 years old and will always be 5 years old.
Midjourney was able to make him look pretty real. And more importantly, pretty close to the original Warren Hippo too:
Meet the original Warren Hippo
Pretty close-ish right? I thought I was on my way to becoming a professional artist.
Unfortunately, image consistency quickly became a huge problem, and it’s one of the top issues that Midjourney users want solved (according to a poll in their Discord community).
Midjourney renderings of Warren in various poses. He seems to have…evolved into a different hippo.
Modifying Warren into different poses was an entirely different exercise. Pretty challenging, and he turned into a different hippo. While it’s still really impressive that you can generate these new images of different poses in about 30 seconds, the character consistency will have to get better in order for the technology to truly enable the next generation of comic book artists, brand marketers, and designers, etc.
And this issue isn’t just limited to Midjourney. The generation AI tools I’ve played with are all really good at generating single-use images, but it’s much much harder to create themes, characters, and brands that you can use over and over again. This is getting better, as you can now seed designs with other images and you can preserve past images as well. But, as of writing this blog post, this is still a major blocking point.
In addition, these tools are all great with common images — like dogs. If you want to create a cartoon dog, it’s easy.
But what about something that doesn’t really exist? Something that doesn’t really exist even in people’s imaginations? Like a hippocorn?
I wanted to create a cartoon of our stuffed hippocorn at Hustle Fund. The original stuffed animal looks like this:
Dunky has wings and a rainbow horn and two large teeth
Unfortunately, because the internet doesn’t really know what a hippocorn is supposed to look like, it’s hard to describe it using generation AI tools. You get weird stuff like this:
Early version of a generated hippocorn image
In fact, we did many iterations on our prompts to try to create a good hippocorn design for our announcement of our plushie. Alas, we were unsuccessful.
Generative AI tools are, by definition, horrible at creating images that don’t really exist, because it has no data to work with.
Over the holidays, I was determined to create a great hippocorn cartoon using Midjourney. So I spent a couple of days working on this. (I know, this is a ridiculous hobby.) I realized that if I were going to be successful, I needed to feed it the data to train on. I ended up seeding Midjourney with a ton of photos of our stuffed hippocorn, and the results turned out a lot better.
After a few days of work, this looks a lot closer to our plushie hippocorn Dunky
But even in seeding it, Midjourney struggled to identify its wings (this is why you see a white shawl around the hippocorn). It took a long time to interpret the two white squares as teeth as well.
In addition, it struggled to render different poses of Dunky as well. You can see the teeth were lost.
Renderings of Dunky the hippocorn after having a tooth extraction at the dentist
But, maybe it’s good enough for some use cases.
Ultimately, we were only able to create hippocorn cartoons because we already had designed a hippocorn. In other words, we needed a designer to create our stuffed hippocorn in order to feed the AI models with our design. It would have been impossible to create a hippocorn from scratch with no designer.
This is where AI still falls short. If you are creating something completely new, you will still need a designer to design what you are developing. That being said, AI can probably speed up some bits that are too tedious to do manually, saving your designer some time.
Why bother with your stupid hippocorns?
As we assess products in AI, which is an incredibly competitive space, these nuances matter…A LOT! I think it’s easy as a VC to watch a demo of a product and say “Wow, that image generator can do ABC things.” But really testing the limits and edge cases is important to understand the state of AI and stay on top of who has the lead.
It’s also important not to write off any of these companies because of these limits. I’ve seen so much change in all of these generative AI design products in just the past few months. They are getting better so quickly, and I suspect in a year from now, some of these problems that I’m writing about here will be solved.
What the future looks like?
Eventually, we’ll look back and say, “Wow, the 2020s was an amazing era for technology.” You’ll be able to build movie studios from your computer and distribute your films on the internet, disrupting the traditional movie industry. You’ll be able to write or draw books, including graphic novels, on the internet and distributed on the internet, disrupting the publishing industry. You’ll be able to create websites and mobile apps more easily — as just one person.
So, if I were going to start a new company today, I would probably not build an AI company — there are so many of them. I would probably build a company that thrives with the assumption there will be a lot of AI companies to help us design and build faster and so much more. For example, at Hustle Fund, we hire no-code developers, because you no longer need to code everything from scratch to get things built. That’s an occupation that didn’t exist even five years ago. So, what are the new roles that will emerge in the next five years? I’m sure there will be prompt engineers in the coming few years. Or maybe QA testers for AI. Maybe you’ll need tooling or legal for the creations you develop through all of these AI tools. Think about what that world looks like — I would build a company for that world, because it is arriving so quickly.
Momentum is moving creativity forward quickly. We’ll see new brands arising from individuals and influencers. The next few years will be remarkable.
If you liked this post, sign up for my newsletter:
Processing…
Success! You're on the list.
Whoops! There was an error and we couldn't process your subscription. Please reload the page and try again.