This week, Teja sits down with Kent Keirsey, the CEO and Founder of Invoke AI, an open-source generative AI product designed to give creators more control over the images they develop. They talk about the importance of building with an OS community, having a solid foundation in product, and why Discord is still (and probably always will be) the place you can learn the most about your product.
(THE FRONTIER THEME PLAYS)
Today, we’ve got Kent Kiersey, founder and CEO of Invoke AI. You guys will love this conversation. Had a lot of fun doing this one. (THE FRONTIER THEME ENDS)
Invoke AI promotional video voice actor (00:22):
(PROMOTIONAL VIDEO AUDIO PLAYS) (SIMPLE MINOR PIANO MUSIC PLAYS AND FADES OUT) The Unified Canvas is one of the features that makes Invoke AI capable of supporting an end-to-end workflow for realizing your creative vision. (PROMOTIONAL VIDEO AUDIO FADES OUT)
So how much of your time is spent like, directly interfacing with like, the Invoke community versus doing like, executive management, like business stuff?
Yeah, I mean, so I make the joke to some people that it’s like working two jobs, because I, and maybe this is just like, on me, but I spend a lot of time in Discord. It’s kind of absurd, but it’s where the community happens. It’s where you feel like, the closest to the cutting edge of what’s happening in the space. I learn a lot from our users. I learn a lot from the developers that are working in the ecosystem. Frankly, like, all of the AI news that, you know, there’s some cool new paper that just got released and some new capability that has been uncovered and released in open source, all of that is, I’m discovering that, in our Discord, not on some like, AI newsletter or anywhere, it’s like, someone’s like, “Hey, I just saw this. Here’s the paper.”
There’s kind of a meme in the ML community of like, “hold onto your papers,” and it’s like, “Oh my god, this is some cool stuff, right?” And we have an entire channel dedicated to that, where people can just drop in whatever cool new paper or a new model gets released, and it’s a constant stream of insights. So I spend a lot of my time there and, you know, recently, we brought on a community manager/OSS product manager, kind of a hybrid role, able to interface with the devs who are contributing their time, making sure that we have a clear roadmap and where people wanna raise their hand and say, you know, “Hey, I want to take on that task or that enhancement.”
We have a really, really strong community that is forming around our open source project, and that is what we end up using in our commercial project. We’re a downstream consumer of our own work in the open source space, and a lot of times, we’re working, half of our roadmap is an open source and, you know, we’re making that publicly available with an Apache 2 license. So, you know, we spend a lot of time with the community, and so it’s incredibly important to be where the community is, and for us, it’s on Discord.
<Laugh>. That’s sick. Yeah, I have so many questions, like, but I’ll save them for later. Like, most of my questions are like, you know, “Why is Discord so popular?” And like…
Oh yeah. I’ve got theories. I, but you know, we can dig in wherever you want to dig in <laugh>.
Okay, sweet. No, I feel like the audience should hear like, how Invoke like, emerged as an idea and as a concept, you know? Like, walk us through the founding journey and your background.
I am a product person. Yeah, my background is mostly in product management. I did some technology consulting at Ernst and Young, out of school, but most of my career’s been in product management in various startups.
You were a management consultant?
Well, yeah, technology, IT, risk management…everything like, cyber security to SOC audits controls, like, IT control stuff. It’s why I know enterprise so well, is because I lived that very deeply. But I did that for three or four years, and then I worked at a software company when I was in college and left to go join a new MarTech startup in Atlanta, and that just started my career of bouncing around the Atlanta startup scene. I’ve done B2B, B2C, B2B2C. I did like, a two-sided marketplace. I mean, I feel like I’ve done every industry known to man. (Teja: <Laugh>.) So all that to say, like, I’m a generalist product person. I’m more technical, maybe, than the average product person, but I don’t have like, an engineering background. I just am nerdy. I played a lot of computer games when I was a kid. (Teja: Totally.) You know, DALL-E2, Midjourney, they were out 2022. Everyone’s getting into the image generation. Generative AI is like, on the cusp of being hot. It’s not quite there yet, but if you know, you know. You know, I’m playing around with the GPT3, API, and the Playground.
Chat GPT hasn’t come out yet, but there’s a lot of cool stuff going on. August 22nd, a model drops called Stable Diffusion, right? And this is like, the watershed moment of image generation AI. There’s a lot of really cool research being done. And I see a post on Hacker News talking about Stable Diffusion. I was like, “I got a computer. I wanna run this on my computer. I wanna find something,” right? So I start, you know, perusing the list of repos that’ll run it, and one of them is by a, it was created by a cancer researcher, works at the Ontario Institute of Cancer Research. He goes by the name Lincoln Stein, world renowned like, cancer researcher, does machine learning on like, the human genome. Like, he’s like, he’s a brilliant guy. He created this little tiny CLI that lets you use Stable Diffusion on your computer.
And I was just like, “Hell yeah. Let’s get this installed.” Install it, start playing around with it, and that was really my first exposure to contributing to open source. I’d always, you know, played around at the open source ecosystem, was very aware of it, but that was like, the first time I really got hands on. And, you know, I wanted to play around with adding features and all that kind of stuff. I very quickly realized where I could add more value to the project was just the product side, the UI/UX, building the community, making sure that people knew about us. And so September came around, we had started getting popular. You know, Lincoln says, “Lincoln Stein’s repo is probably not the best name for a popular Stable Diffusion repo. What should we name it?” I was like, “How about Invoke?”
And so we released that, and renamed, and then launched the Discord server and all that kind of stuff, and it just kind of takes off, because we had a very UI/UX focus on the experience. And, you know, we were approaching things a little bit differently in a much more, maybe, sane engineering architecture approach, and in a lot of ways, I think we cared more about the artists and the creatives who were gonna end up using this tool. We saw this as like, you know, AI’s coming, there should be open tools out there that make it accessible to use. And so for all those reasons, you know, we kind of continued to grow. And then, it was in January, we started getting inbound from various companies that were interested in using the tool like, in a professional sense.
The big companies started reaching out. So, you know, we had chip makers, we had the AI companies, all of them were kind of interested in, “Hey, are you guys interested in partnering?” Some of them tried to hire us, and that was the moment for me where I said, “I think we’re on to something. I think this is like, there’s a way to make this a business.” (Teja: Right.) Open source businesses are hard, (Teja: Right.) but I know the playbook enough. There’s been a lot of open source businesses in the past, and most of them are not application layer, just to be clear, but there was a way to make this successful and do good in, kind of, conjunction with that. So, founded the company, raised a pre-seed, started hiring, sprinting, raised our seed round, and then, you know, we’re piloting with some of the biggest companies in the world right now, using our tool to power their team’s generative AI, kind of creative approach.
That’s sick. How big is the team right now?
We are at nine right now. So we have grown from just me to nine. That’s not including, obviously, our open source community. And I think it’s important to call out how critical the open source ecosystem is to what we do. Lincoln, who I mentioned created the repo originally, is an advisor to the business and still an active maintainer. We’ve got another maintainer who’s in the games industry and actively contributing, and we have a ton of people who help make the product better every day, whether it’s through feedback or translations. We’ve had, I think, over 18 languages added to our translations, and that’s all coming from the community. We have a very engaged community of open source contributors, and they’re critical to making what we do successful.
That’s amazing. So, if you were to measure like, the, let’s say, contribution to the product of like, the internal team versus the like, open source community, is it about like, 50/50 like, balanced right now?
It’s hard to compare. (Teja: Yeah.) I would say, we are doing, the way that we think about it and then extracting it, there are multiple layers that can be contributed to. (Teja: Right.) We like to think of ourselves as being very much responsible for the core engine that runs everything. A lot of the like, architectural decisions for how things are gonna work are made by us. (Teja: Right.) We opened up, we did a re-architecture, which made our backend very modular. We have kind of like a graph-based workflow builder for diffusion pipelines, and what that means is we gave contributors the opportunity to add their own nodes, which are essentially just one like, foundational building block in that graph.
And so we have an entire community of people contributing just one function in that backend, but we’ve made it very modular, very easy for people to plug those in, and that is its own kind of like, wild ecosystem where people are throwing in all kinds of things. Like, there’s somebody who created a node that uses a local large language model to generate nightmare prompts. It’s just like, nonsense prompts that create weird images, but there’s like, an entire group of people who enjoy using that node, and so people are like, downloading that, and installing it, and having fun with it. So that type of community contribution, it’s hard to measure, because it’s just like, there’s so much of it. To the core engine, to the core UI/UX, I would say we’re probably doing the majority of it.
We get some help here and there, but where it’s interesting is, the community contributes in ways that are hard to measure from like, a commit or pull request perspective, because you have all of these really, really smart machine learning researchers, whether they’re like, in an official capacity, a machine learning researcher, or they’re just kind of like, this hacker machine learning mad scientist who’s playing around with this stuff. You have all these people throwing ideas into the Discord channel and saying, “Well, why are you doing it that way? There might be a security vulnerability if you do it that way, and maybe we should do the queue this way. Here’s this research approach that could be better than the way we’re doing it now.” And what’s magic about that, is that’s an extremely valuable perspective to have. (Teja: Right.) It just like, it gives you more of that diversity in approaching problems, and it also just shows how the community sees this as something that they use the word “we”, right?
“We” are building this, and I think that’s like, incredibly important to foster, is a respect for the fact that we are building this together, and this is a project that we are building, and it’s a hard balance to strike with an open source project, because you really have to make sure that from the very outset, you draw the lines. When you have a commercial side of the business, you draw the lines and say, “This is how we’re gonna approach this,” (Teja: Right?) ’cause otherwise, you know, you set expectations wrong and then you violate those expectations, and the community shatters. We don’t want that. We want this community to thrive, and we wanna honor that commitment to what we are building together, and so we have a very, you know, structured way of looking at what is open source versus what are we trying to do on the commercial side? Who are we trying to serve, and what is the commercial business aiming to do? (Teja: Yeah.) But the community I think has responded really well to that.
That’s so interesting. I mean, you know, this question might be sensitive, so just feel free to tell me if that’s the case, but how do you think about the commercialization opportunity like, just in general, and how do you balance that with trying to, you know, provide wins for the community? Like, are you trying to commercialize large companies (Kent: Yeah.) mainly, and then sort of, there’s a free tier, and everybody can use it to the extent they want.
Yeah, so the way that we think about it, part of it is strategy, right? It’s how do we strategically, how are we strategically advantaged by being a main contributor to this repo and being kind of the shepherd or steward of it, (Teja: Right.) and how do we build a commercial product that takes advantage of that position? We are focused largely on the multi-user ecosystem, right? So the way that we see our responsibility is making sure that the open source studio, that’s kind of what we call the community edition of our product, is like, the “studio piece,” (Teja: Right.) central generative AI, that is always kept up to date with the cutting edge of generative like, research, right? Because that research is happening in open source. If it’s getting contributed by open source contributors, it’s because they want to see that functionality in their tool when they use it on their computer. From our perspective, that’s the purpose of the open source repo, is to be everything that an artist needs to compete in this world where AI is taking over everything. We see ourselves as very responsible for creating high quality, good software for society, for the people, right?
And that relationship there is symbiotic, right? We benefit from them using the product. We benefit from their feedback, we benefit from their help in building it and making it a better tool, but where we as a business are focused, are really taking that experience and making it work well at an enterprise scale. So we take that tool, when you talk to an enterprise, they don’t want to install a local like, install on 5,000 computers, and manage that, and try to figure out how to user management. They don’t wanna do all that, right? They need SSO, they need, like, they need sane kind of controls in place around this stuff, and at an enterprise scale, they’re also looking at compliance, right? And so there’s a lot of compliance things that have to be taken into consideration.
We can build all of that. Our OSS users don’t want those features. They don’t care about those features. It doesn’t matter to them. So the fact that we’re putting that in an enterprise version is kind of like, irrelevant, but it’s extremely valuable and helps us deliver that enterprise experience. And that’s kind of how we view our position, is really making enterprises successful with this technology as a technology partner. And, you know, I think broad, this is like, a commentary broadly on how AI’s developing, but I’m very much in the camp of open source AI (Teja: Right.) being where most businesses go. I don’t think where AI’s going, which is a core enabler of most businesses, it’s not something that’s just like, a side project or some like, you know, a Skunk Works thing that somebody’s gonna hack on and doesn’t matter.
It’s like, this is core if you’re investing in AI, and it is on every executive’s agenda to like, figure out AI, you are asking very fundamental questions. “What does our business look like in five to 10 years, and how does technology enable that, and how is AI a key part of that strategy?” Almost every single executive, if you ask them, “Would you rather be investing very heavily into a model ecosystem that you do not own or do not have control over, or would you rather be investing in something that you actually have the ability to control? You have, you know, a vendor optionality, you have a way to take this and make it specialized to your business. You know, which of those worlds do you prefer?” The latter for by like, 95%, right, and that is enabled by open source AI. That is enabled by the ability to take a model and fine tune it on your business.
I think Llama 2 is a great example of like, cutting edge research that is now able to really drive a lot of business value, but there are a lot of businesses to be built in helping businesses take advantage of that ecosystem. And that’s kind of where we see ourselves. We’re not tied to any one model. We are that creative suite that any business who needs this type of high production value tooling, we can be a player for that.
I have a lot of respect for trying to like, I mean, you guys aren’t threading the needle between like, having a product that has sufficient, I don’t know if like, I’m even thinking about this right, but sufficient like, defensibility to be able to be commercialized, like, sufficient levels of being unique, that sort of thing, but then also like, having a vibrant community and leveraging the power of having awesome and dedicated supporters. That’s difficult to balance, (Kent: Yeah.) and, you know, and yeah, from what I see, it looks like it’s going like, super well.
It is <laugh> so far. So far, so good. I mean, we’re only a year into this revolution. We got a lot more to go.
No, dude, that’s sick. That’s amazing progress. Last question: Like, how’s it feel to be building this company in Atlanta? Do you feel supported by your community, because this is sort of like, this is like, frontier tech, you know, and Atlanta’s built on, you know, sort of MarTech, email tech, you know…
…FinTech, you know, yeah. Not really like, not really known for this type of (Teja: Yeah.) stuff. Yeah, you know, it’s funny. When we were raising money, there’s actually one VC who we were gonna get intro to, and they asked the question, “Where’s the founder located?” (Teja: <Laugh>.) and the answer was, “Atlanta.” And they’re like, “Nah,” <laugh>. (Teja: <Laugh>.) “We’re good,” right? <Laugh>. And you know, I think it’s a stupid response, honestly. Like, I think there’s a lot of, there’s a ton of value being in the Bay Area. Like, I’m not gonna like, dog on that, (Teja: Yeah.) but I don’t think what’s happening right now in AI is happening in the Bay Area. It is happening everywhere. It is happening globally, and open source research is not coming outta Silicon Valley. It is coming from around the world, and you’ve got these people who are creative hackers that are just like, you know, in random places. Like, one of the guys that was one of the first contributors to our app in the open source ecosystem built really like, the core front end.
He built a full React-type script app. He’s in Australia, and when we like, formally started the commercial endeavor, he was like, the first to raise his hand and say, “Hey, look. I will leave my job yesterday. I wanna work on this full-time,” and, you know, you’re making a decision as like, a founder based in Atlanta to say, “Okay, this guy in Australia, with time zones that are insane different, is that the company we build?” And the answer for me was, “Absolutely,” because I’ve seen what this guy can do, and there’s this myth, or at least people think it’s a myth, of the 10x engineer, right? That’s like, “Oh, it’s a myth. They don’t actually exist.” But I saw the passion that he had and how engaged he was, and that is what matters, right? Is you care about a product so much that you wanna put yourself in it, right?
You care to invest yourself in building that product, and that was what he had in spades, right, and so it was like, “I don’t care where you are; I want you to be on this team,” right, and that’s the type of, I think it’s the type of approach you need to take when you’re hiring as a remote founder. You need to be focused on finding the best people, regardless of where they are. I know that there’s this whole like, “return to office” thing going on where everyone’s trying to get back. There’s something that’s like, very important about being in-person, and I am a whiteboard addict, and so if I were in an office, I’d be at a whiteboard talking to people all day. That’s just like my happy place. (Teja: Yeah.) So being remote is very difficult for me, but location is not, it’s not as important as it once was.
And in the context of this type of product in an OSS ecosystem, location is completely irrelevant. You asked earlier, like, Discord, like, “How important is that to you, and where do you spend your time?” That is my office, in some sense, for the OSS community. That’s where all of the stuff is happening. So really, location is completely irrelevant for that, because we’re all just like, avatars in a chat box, right? (Teja: <Laugh>. Yeah.) And I think it’s magical being able to work with as many people from around the world as we get to.
Yeah. That’s amazing. No, I totally agree with the remote, sort of, first, I mean, our company’s remote. There is like, so actually we threw an event on Wednesday, and so we like, had a bunch of people come, and there was just a conference in Nashville, so we had a chance to reconnect with some folks, and honestly, grabbing drinks with people that you work with and people that, you know, that’s super fun. And so I’m like, damn, I miss doing this like, more often, but it is nice to like, you know, it’s nice to, like you said, like, I mean our head of like, the whole whole technology org is in Brazil, like, product and engineering, so, you know. It’s fine. Like, you see ’em every now and then, and it’s good, you know? So that’s cool, man. I mean, what an awesome product. What an awesome journey. I know you guys are just getting started, so you know, you should come back anytime you guys have a release or a launch you wanna celebrate. I’m sure that network would love to hear from you. Where can people find you and Invoke?
People can find us on GitHub and our website. Our website’s invoke.ai. If you google “invoke ai”, we’re like, either GitHub or our website are the first like, thousand results, so you can find us pretty easily. (THE FRONTIER THEME FADES IN) Our Discord is pretty easily discoverable as well and linked to from our GitHub, and yeah. I mean, we’re a passionate group of people who like playing with the latest and greatest in AI image generation. If you are interested in that, feel free to join us. We have a lot of fun. You are welcome to come and contribute, or just use the tool, and explore the wild world of AI with us.
Yeah, go check out their Discord, y’all. Go sign up for their product. Thanks Kent. (Kent: Yeah.) Appreciate you.
Yeah, thank you.
Faith, via previous recording (24:54):
Thanks for listening to the Frontier Podcast, powered by Gun.io. Be sure to subscribe on your platform of choice, and come hang out with us again next week, and bring all your internet friends. If you have questions or recommendations, just shoot us a Twitter DM @TheFrontierPod, and we’ll see you next week.
(THE FRONTIER THEME ENDS)