Synthesia CEO Victor Riparbelli on AI and Video Avatars

foreign speaks we're talking about artificial intelligence in the video sector we'll take a look at AI driven content creation what's happening now and what we can expect in the future to discuss that I'm joined by Victor Ripper belly Chief Executive Officer of Synthesia Victor very good to have you with us today thanks for vitamin James excited to be here so AI video content creation it feels like we're sort of in this Brave New World video avatars video Avatar is talking with us uh let's step back I know that Synthesia is an AI video creation platform can you give us an overview of the technology and how it works yeah sure so so this is the world's largest platform for AI video generation today and what that means is essentially that we we help our customers make awesome video content without the need for any physical recording process so that means no cameras no actors no studios and none of all the things that that generally make video production um a lot more difficult than making a PowerPoint slide for example the way it works is that you you go to a website you select one of our stock avatars which is real actors who are basically paid for every video you generate with them and then you simply just type out your script so kind of like your director of a film or something like that you just type it out and we have uh I think of it as like a sort of a canva PowerPoint install editor around it so if you want to add in some screen recordings logos text images all those things you have in in in normal videos you can also add those then you hit generate you wait a few minutes and then we serve you back a video of your chosen actor reading out your script screen that's like the simplest way of explaining it and today it's very focused on what I call kind of business content so think of it a lot as training learning sales marketing content various books like talking head style of video um and I think a good analogy here is sort of like things you would make a PowerPoint about are usually great types of content for working with on this DC platform or the 510 year Horizon our goal as a company is to enable anyone to be able to make a Hollywood film on their laptop without the need for anything else than than just their imagination and a Synthesia login and while it does to some extent seem far away I think most of the Technologies are there the building blocks are there it's a matter of connecting them and of course improving upon them um and and I'm very excited for us to sort of move into a world where everyone can make video content without being a professional and uh I think they'll they'll help all of us communicate much better today this relation obviously no doubt there's a market for that um is there also scenarios in which someone would interact with a video Avatar in real time or is it only reading text and it's it's it's set text and that's that's what they can do yeah absolutely I think that we're very early in AI generated content you know we've been working on this for five years and I think the last year or so has really been the Breakthrough year we've seen AI generated text AI generated images and generated video really kind of moving from being you know cool stuff that people did in a research lab to something that businesses use every single day which is very exciting but right now I think we're at the stage where normal video and AI video sort of looks similar you know it's a video you play from A to B and it's it's still the sort of broadcasting methodology but what's going to happen in the coming years is that AI video AI generated content is gonna become its own media format that's going to be different from the images and films that we know of of today right and I'll give you a little analogy to maybe help you picture like exactly what I how I think about this so the first websites they look like a newspaper on the screen right that was kind of what people could imagine right the first films where people using a camera to record theater plays because what else would you record right obviously from there on we went to websites which have links and video and audio and interactive it's very different than reading a newspaper and the same thing with films right a theater play and a movie space to very different things and that's that's because they enabled wire technology that you could do new things I think we've got to see the same here we're going to see much more personalized content because the interactive content where let's say you're taking a a training a learning course online there'll be some embedded intelligence in that video you interact with it maybe it changes depending on how well you understand the material and it can maybe skip some sections it can explain some things even better than before it's always hard to read about exactly how this is going to look but I'm very very convinced that AI video and AI content will become its own thing and I think the key headline for this is that we're going to move away from broadcasting which is that sort of you know you create a video which everyone that watches it's a much more personalized and contextualized uh experiences and I think that's gonna that's gonna change how the entire internet works and how we interact with media in a very exciting way hmm what about the idea and obviously I know you must be aware of chat gbt I mean it's it's been a big milestone it came out recently and everyone's been you know using it to create content people I mean the interesting thing about chat GPT in my view just the fact of it is that it's it's just it's generative right so you can ask it to write a poem it will write a poem for you where something like Alexa can actually retrieve information chat TP will actually you know create something so it's it's really a milestone in that sense what about the idea of of combining chat GPT with since Asia oh we've been using GPT for the last two years um so if you go to our Tick Tock channels you'll see lots of GPT generated like right now we have ai Santa Claus who on our Tick Tock livestream is reading out a gbt generated uh Christmas Stars that's a bit more of a fun thing but we also use it internally to create video content that we're also build an integration into the product so that you um if you want to create a video about fly fishing you can come into Synthesia you just type out uh five part calls about fly fishing and we'll then generate sort of the entire video script and all the sort of video slides that matches that and then you as the user can just go in and you'll probably still want to edit whatever you know we sort of generate for you but it'll be so much faster than you sitting from scratch writing the script putting in all the visuals right and I think it's going to be a big theme uh with awesome with other creation tools that we're going to move from sort of in a world where we as a human sort of come up with everything from our blank screen sort of to being much more about editing content right so you ask GPT and Synthesia to like hey build me a video about this particular thing and then instead of you know you sitting and doing it from scratch you'll just edit the last 20 to make sure it sort of fits The Narrative of the story that you're telling and so I think this embedded intelligence is going to be something we'll see definitely Synthesia it's coming within a couple of months but also in in many other applications that that all of us use every day what if I wanted to talk with one of your video avatars in real time and just have a conversation about technology or whatever um so the the the video Avatar we will in other words could I interview a robot so to speak with with the chat GPT feed going into synesthesia Yeah we actually built that we had a kind of a hackathon a few weekends ago uh we built a full real-time Synthesia version connected to GPT free we haven't released the video yet but we have a quite funny one of our our CTO John interviewing one of our AI avatars as a candidate for for a research engineer position and it works amazingly well the latency is less than 1500 milliseconds um so it does feel like a real conversation and I definitely think this would be a thing of the future it's not something that we are planning to release as a product in the near-term future but uh but 100 this will be part of of the future of of AI video and AI content um interesting are there are there specific examples of how companies are using AI video now you've talked about them a little bit but it'd be nice to see you know real world examples how are companies really using this now so I think an overarching theme really is that we're moving from uh you know we're moving into a world where people don't really want to read they want to watch and listen to their content right we're also recording this as a zoom webinar because people want to watch and listen right they don't want to read a Long blog post about something and so that's great but for big corporations which is a lot of what we work with it poses a huge problem right because producing video and audio at scale is incredibly difficult and Incredibly time consuming right so what we actually help these companies do is instead of producing a lot of text and slide content they can now produce video content and they see it as they use Synthesia to replace the text and slides that they have so let's say you're a big fast food company for example right right and you're onboarding you know millions of people every single year you have to teach them how to operate a deep fryer how to handle an angry customer all these different things a year ago they would send you 28 pages of PDF documents that you would have to sit down and read and maybe that'd be like a five minute video uh to just like set the stage before right and then I'd hope that you kind of ingest that and understand it which of course especially in a Frontline worker scenario you're not going to remember that much of it now what they can do is they can create video content and the big unlock right is that the same person who used to create those PDF documents and slides they can just use Synthesia they don't have to go to a video production department and get a budget and involve five other people in making that video content they can just do almost with the same workflow they had before they can now produce video content this goes you know in this scenario it goes for you know SAS companies tech companies we work with want to build better onboarding programs for their customers where they can make that into video content which is always on brand you can always go back and edit it you're not relying on someone from your customer success team who creates 35 great videos and then leave the company a year after and all of a sudden whatever reason 35 videos If an employee who doesn't work here anymore we can't create you know there's all these problems of working with with like the normal way of producing video and audio of that matter that we solve with this so I'd say it's a lot of um it's a lot of like you know less um exciting use cases to some extent right but I think it matches very well what the technology is today right I think you have to be honest about not just our Tech but a lot of these AI tools there's a lot of way to go in our case you know adding emotions for example to the avatars making them more expressive that's going to be a big part in you know enabling people to create like real entertainment content that'll make you laugh for example right that's not really where we are yet but it's great and it's current state as being you know for instructional content for example so I think what we'll see over the next years is that as the technology gets better more and more use cases gets gets unlocked which is which is really exciting it seems like one of the real challenges would be adding emotion to the avatars we need that ultimately if we're going to be completely convinced by a speaker we need that speaker to have some emotions but it seems like that will be quite challenging you get into some nuances of facial expressions uh let's go into that a little bit if you talk about the challenges of AI video rather whether in prediction production or in the audience response what what's a challenge for AI video and what is still holding it back yeah well I think that where we are right now is that it works really well for a specific set of use cases right so I kind of talk about a little bit about that before but right instructional educational content um it's not yet you know at a point where it can make you feel something right in an emotional way like an advertisement would have to for example the challenges here are that obviously this technology is is um is is very new and extremely hard to do you're basically trying to teach a computer how to replicate real video of someone right we make an amazing progress and and you know next year we'll see some huge improvements in what the avatars can do we've just added in a micro gestures to the avatars so now avatars can nod their heads they can blink they can frown they can do these like different sort of actions and next year it's going to be um it's going to be you know even more functionality added to make them come alive even more the next like real big sort of Step change will be moving to avatars that can you know like dance around have have a conversation together be in a room whereas today it's this like talking head style of video and I mean just to maybe you know put into context how big these challenges are like you know we've just spent like upwards of five million dollars on building a capture stage in London you know 100 cameras in a green screen room we're gonna capture thousands of people next year but it's going to provide us with a data set of how humans move how humans look when they do different things how does our emotions change when we speak not just from a single camera but from 100 cameras right that's not because we want to capture avatars with 100 cameras but we need that data to be able to infer from just a video of you and create an avatar view that is very expressive so I think you know this what what we definitely see in AI is that it's very much driven by these step changes if not like month by month it just gets like better and better it's generally like a big step change happens and well now it's like three 5x better right another step change happens and that's what we've seen continuously you know through the five years of building Synthesia that there's these sort of technological step changes that that really impacts what you can do um with it so I think that's on the technical side on the audience side uh it's really fascinating to see this play out right because you know we had people three years ago telling us like but people would not want to watch videos created by Ai and you know all things and all right that might be right in the future like I think we will definitely look differently at a movie that's generated completely with AI versus one recorded with whoever your favorite Hollywood actors are and I think that's fine I think they'll live side by side but actually what we found so far is surprisingly that not a lot of people have any problem watching AI videos and it all comes back to what are you comparing it with what is the alternative right if the alternative is watching a super well-produced Netflix quality um training video that corporate has spent you know like 300 000 in making they'll definitely pick that and they don't want to watch the AI video right sure but if the alternative is to read 28 pages of PDF documents but you then have to remember and take a quiz everyone will pick the AI video even if it's not 100 as good as a real video right and I think that's what a lot of people miss when they think about how this stuff will be perceived is that it's it's a most of the successful use cases today I would argue Target inherently a new market it's not necessarily you know like I for you know image generation a lot of people got it oh but that'll kill shot of stock right and maybe it will one day you know it will definitely impact them to some extent but actually what I think these things do is they they open up new markets right people who couldn't afford strata stock I didn't use Shutterstock before all of a sudden now they can do something with it and it's a bit the same we're seeing here so I think it's not always as easy as to say you know will will people want to watch this content because often the alternative is not doing anything at all or it's for in our case it's like reading long pages of text right which nobody wants it sure well yeah no doubt no doubt about it it is a comparison Market I mean I I think certainly many workers face with a 28-page PDF versus watching the video I think it'd be no contest uh so the market there in terms of companies and training no doubt um we look ahead say you know a few years several years out the future of AI video you've talked about a little bit the idea of someone making a complete movie on their laptop I mean it's pretty wild what are you foreseeing in the years ahead not with current technology but imagine how technology you might imagine what might move ahead what do you see is the future of AI video I think that there is some useful sort of anchors in terms of like imagining what the future will look like right so I'll give you one example is uh and I produce music in my free time so that's that's something I think a lot about today I can open my MacBook and I can sit on a plane on my way to New York and I can synthesize and control a 65-piece Orchestra right you can make any sound I can make any vocal models that I can think of just by typing on my laptop right that would be completely inconceivably like 20 years ago right right it'd be an absolutely crazy idea right like you need a piano you know a guitar you need to have a real recording studio you need all of these different things right today you don't need that because we've managed to compress the entire production process into something that fits inside of a computer and that's because we can synthesize sounds and effects right um and that has changed the landscape of Music completely right 20 30 years ago you needed to be in the right City you needed to know the right people and then you would have a shot at maybe you know releasing a record at some point what has happened the last 20 years is that everyone bedroom producers they can just sit at home they can record whatever they do they can generate everything on their computer right they can share it on YouTube and all of a sudden it's more of a competition of who makes the best music as opposed to who do you know and all the traditional Gatekeepers that exist but in Film Production that's not really the case yet right because it's so expensive to produce film you still have all the traditional gatekeeping there's like the the dollars you would need for example there's where in the world are you there is knowing the right people right there's being on Trend there's all these different things you kind of have to go through and I think that we will move into a world where just that you can produce music on a laptop you can produce a Hollywood films right and it sounds crazy but I mean Hollywood already does this right most of the things you watch in the cinema is produced in a green screen stage in a warehouse somewhere even if it doesn't look like it it's just a matter of taking those Technologies and compressing them down into something that fits into a computer just like the first synthesizers they filled an entire room right so I think it'll have a huge impact on the business of film and entertainment in general and from a technical perspective it'll be something you could do on a computer but what I what I find even more exciting than that is that it'll be something that everyone can do you know it'll not it'll not be something where like okay if your visual effects artists from Hollywood you can now be more productive the systems we're going to be building is going to be type out two people standing in a wooden cabin there's a red couch in the background character a says to character B whatever character B responds this a dog walks into the scene and that's what you as a user is going to type then the software is going to come up with a scene that sort of looks like that and then you can go and say change the dog from a labrador to a golden retriever or you know change the appearance of something like that so it'll be this way of creation where it's not about your Tactical ability you are more kind of directing right like like a director would do on a set that would say James can do that line again but make it a bit more funny or upbeat for example I know this sounds like a bit like maybe a bit uh you know out there but if you look at how stable diffusion Del e some of these image generation tools that out there today work if you try and interact with chat GPT like the building blocks are there right it's not as crazy as it sounds and I think that in the next five years we'll see significant process progress towards making this a reality in the beginning it probably won't be able to you know do uh two people are hanging off a helicopter while flying through the jungle and shooting the bad guys uh it'll probably be contained more to like two people sitting in a in a studio or you know like it'll be a bit more simple in the beginning but it will eventually sort of get there and I think that will be a huge democratization um uh lever right just like blogging made everyone into a journalist I think these technologies will increasingly make everyone into into a Content producer fascinated it's really the democratization of content creation which has been going on for a few decades as you as you say I mean the music has been really democratized with midi Technologies and synthesizers and certainly you know blogging uh the internet itself so this is going to be a super interesting sector to watch um Victor fascinated talking with you thanks for sharing your insights today I've learned a ton I hope you come back and talk with us again sometime thank you definitely Thanks James inviting me

This transcript was generated automatically from the video's captions and may contain errors.