Emad Mostaque - Stability AI @ TOKEN2049 London 2022

I'm going to talk about creativity and artificial intelligence um some of you may have heard of stable diffusion a AI model that we released you put in prompts like latent Spaceman The quillin Ridge and the Isle of Skye DJ hamster and a second later this is created this is a model that we released about two months ago and designed it to basically enhance creativity but it's just one of a series of models that we have released and what a model is is just a small file that can do just about anything because the pace of AI change has gone exponential and this model and this type of thing has been available for the first time and we've been kind of pioneering the open source side of it so a lot of you will have seen air nfts and things like that all those models were designed by our team at stability because stability we kind of combine open source Ai and Community to have communities with 20 000 developers from a Luther that did all the language models that drive things like alert Theory AI to carpet that does things like open source code models that we really soon opened biology biology models that open by ML and then image models like you've seen there the nature of AI has changed in that the AI that you were used to in web 2 was extrapolation AI for kind of targeting you with various ads and things like this this is a new type of AI which I call a big model AI where you take large amounts of data and you learn principles from it to be able to do things like what is a hamster what does a space hamster and things like that that requires a lot of data but not as much as previously in fact we used a hundred thousand gigabytes of image data to create stable diffusion but you need something more we actually have the 10th fastest supercomputer in the world to do this we also run on decentralized gpus and so we'll be making some announcements about that soon so there's Us in red about a quarter of the total compute of meta about 10 times the compute of NASA or the UK's fastest supercomputer you then need talent from the communities and those data sets like the ones we created the hundred thousand gigabyte lion data set which is two billion images that we managed to compress down to a two gigabyte file that you can run on your MacBook to create any of those images that you've seen right now and this is what we mean by distributed creativity you don't need to have a server you don't need internet you can run on your MacBook generate those images in 30 seconds and it's not just images we've got audio that we've released we've got video coming up we're basically democratizing AI for everyone and we're releasing an open source to the world and I think it's going to be very interesting going forward because it's been very difficult to communicate like the easiest way to communicate has been speech like we're doing right now I'm talking to you then text is more comp located and you've seen gpt3 and language models make that easier and we release GPT Neo and j and x which have been downloaded 25 million times by developers now but visual was the hardest be it a PowerPoint or an image like you've seen with stable diffusion now it's easy for anyone to create an image you can go on dream Studio or mid Journey or any of these other things and type a Word and second you generate any image but it doesn't stop there we're building a PowerPoint killer where you can describe anything and it'll automatically create PowerPoint slides that are beautiful every time visual communication is now possible for everyone and that flattens the curve be it if you're five or if you're 50 or 100. you won't have to struggle anymore because I hate PowerPoint and you know I have advantageous I can't visualize anything in my head now I can create anything and I think that really opens up things we've had about two million users using dream studio in the last two months which is our core AI Suite but about 20 million people using the technology in total generating about a billion images the stable diffusion model itself was downloaded 200 000 Times by developers so 200 000 developers have been using it making over a thousand projects and we'll show you some examples of that in a little bit um I think that this technology is about to take off because over the next year we're going to take it down from a two gigabyte file to 100 megabyte file and I think we'll be able to take inference time the time it takes to make an image from one second to 24 frames a second so you can generate videos basically live within the next couple of years of anything you can describe and you'll be able to do that eventually on a mobile phone but even on a mobile phone being able to generate an image of anything is a bit insane we've already got it working on an iPhone and so we think this technology goes to a billion people I think that kind of changes the internet as well because we're all here talking about web3 and you know I think we're moving towards an intelligent internet because a lot of web3 didn't have ai in it and so part of this was to give AI be it text search image video everything to web3 and fill that Gap there because right now we've got all these identity protocols and rails but we don't have the intelligence in that we don't have this cut to get intelligence the only entities that can build this intelligence and this AI are ourselves who are decentralized and have these communities and open source angle Google meta Nvidia and Microsoft slash open AI so we're really trying to kind of fill that Gap there um and I think eventually every person company country and culture will have this technology to be able to create and communicate anything they want to say again this will be extended so we have everything from kind of logic based to 3D to video so in the next three months video models 3D models we've released our Audio models you'll be able to create anything you can imagine and again have that customization this opens up opportunities on the nft side token Side Community side everything because every single company country and culture will have to have their own models and again this is being made accessible to anyone in fact you'll be I'll be showing a video in a minute you can customize your own model with your own face within about five dollars of compute now with the base that we've done because a supercomputer takes all the knowledge and compresses it down to a file that can then be adjustable so it's like a foundation for everything which is why also they're called Foundation models and I think this changes the nation of a metaverse as well a lot of us have talked about it I don't think there's this meta type thing whether you had no genitals and legs until recently you know I think it's everyone has their own AI models and they all bring their own content in here and they generate it live similarly you don't need Google image search when you can generate images within a second again this has only been possible within the last couple of months and I think everyone in this space needs to understand the implication of this from image to text to video to audio code and elsewhere so I think it's more Multiverse so Multiverse where everyone can basically participate and again I think that crypto and tokens are a real part to play in that I think it should be open again these rails can't be controlled by any one entity let's face it most of the value in web 2 was AI companies targeting your attention and so again I think what we're building is a vital part of all of that across all these modalities uh that's why I think it's AI for the People by the people and that is our GitHub so in two months we've got half the stars of Bitcoin and we're heading towards ethereum that shows you the developer pickup here and I think it's going to continue and then as everyone gets together and builds cool things it's going to be a bit insane how this thing continues especially because our next model that we're about to release in a few weeks let's just say it's got a lot of surprises there so I know it's been a long day I'm going to finish up there I've created a nice video to show off some of the things that people have created from the Community First releasing this open source and I look forward to you using this technology to create send me your favorite Creations right can you play the video now everyone foreign [Music] foreign [Music] it's an explosion of creativity collaboration and open Improvement it has absolutely blown up recently I'll just explain real quick hey you got this big model you can type in anything and you get a picture you're like okay I'm going to type my name in and like nothing comes up I trained the thing for a week and a half on it looking like me but guess what on my drive down this guy figured out how to do it in 15 minutes [Music] foreign [Music] options