About Stability AI
- Founders: Emad Mostaque
- Founded in: 2019
- Employees: 110
- Money raised: US$101 million
- Ultimate goal: To make all static datasets intelligent and make open source AI tools available to everyone
In the past, you needed to have the ability to draw in order to draw a good picture. But nowadays, you simply type in what you want to see, and an AI software turns it into a nice picture. One such so-called tool is the open source package Stable Diffusion. The software was developed by the research community of start-up Stability AI, which creates AI tools that anyone can use.
The young British company was subjected to heavy criticism after the launch of this text-to-image tool. For a start, the software was then used to generate deepfakes and pornography. People also questioned copyright infringement if new images were based on existing photos. In spite of that, the company did manage to raise US$101 million in investments. In this instalment of start-up of the day, CTO Tom Mason talks about developments within the platform and the ethical side of text-to-image tools.
What is Stability AI?
“Our company has reached out to and funded several AI research communities over the past couple of years. By doing that, we have created an ecosystem that is comprised of online communities spanning a thousand to sometimes fifty thousand people. They are creating public, generative AI models for image, audio, video, 3D objects and biology. Lots of them are 3D game developers, graphic engineers or PhD students. All from different fields. Some of them do it in their spare time for fun. They work mostly autonomously, but to do that they need two things: financial support and computing power. Our CEO, Emad Mostaque, got in touch with these clusters and provided funding and computing power for two or three principal researchers in each community. That’s how Stability AI was born.”
What is your revenue model, since you can’t really earn anything from open source software?
“Besides creating open source tools, we also have a commercial strategy, which is all about datasets. Many companies make money from the content that people use, but don’t do anything else with the datasets that are generated with it. We create a model for them using one of our AI tools, but one that is tailored to their own dataset. We then either get a certain percentage of the revenue or we host it on our own server. We’re also planning to develop pro versions of our free open-source tools. Then you can subscribe to those.”
Is Stability AI responsible for any unethical content created with Stable Diffusion?
“We absolutely bear some responsibility in that respect. We created the ability for this to happen in the first place. Although we shouldn’t be the company that gets to decide what is right or wrong with generative AI. Any tool can be used to create harmful content. People have to understand that this technology is extremely powerful. There needs to be a civic discussion about how this technology can be used properly.
We need those reactions from the wider public and our online communities. It’s only been two months since we released Stable Diffusion and already there are groups getting together to evaluate it all. Thankfully, we are having that conversation out in the open. Just imagine a company doing this without that kind of transparency. You would have no control and no expertise to be able to have this discussion at all. That’s a much more frightening scenario.”
Was it difficult to get funding?
“There was great interest after we launched the open source model of Stable Diffusion. It was incredible. I think people looked at us and wondered: who is going to do it better? A company with 500 people that operates in a closed setting and keeps everything within its own walls, or a company made up of a hundred people who are part of an open community of a hundred thousand people? It just seems like this technology should be free and open. If we can encourage that and support the community, it gives us the opportunity to compete on a whole different scale.”
What makes Stability AI so unique?
“I think it’s down to the fast pace that we are developing at. We have different kinds of models. We have large groups of people working on these. We have a lot of money to develop some amazing products and build an ecosystem. We also have the capability to train customized models at a lower cost than anyone else thanks to our relationship with major cloud computing providers. And you can’t get computing power at a lower cost anywhere else than from us. People can make new models with us cheaper than at any other place.”
What is your ultimate goal?
“We want to turn all the static datasets in the world into generative models. A pretty big goal, in other words. For me, it’s also just tremendously exciting to be using this technology. It really is a kind of revolution. Recently, we carried out a survey among our own users. It revealed that one of the greatest advantages of this platform is that you have a feeling that you can be pretty creative. We want to introduce AI tools to the whole world so that everyone can also have fun with them.”