RenderCon 2026 Keynote: The Future Of Rendering

Added: 2026-05-25

[00:00:04]Welcome to Render Con number two. Uh what an incredible year uh it has been since last Render Con. Uh and uh it's again a honor and a privilege to serve uh so many creatives and talented folks uh and to have a day dedicated to the uh the great work that uh you know this industry is um all about. Uh this first talk today is uh my usual keynote about the future of rendering. And you know the tricky part about doing this is that the future keeps changing really quickly. We are definitely on an exponential curve. And if you think about where we were April 15th, a year ago when we had Rendercon, there was no VO3. There was no audio. We were in the silent movie era of generative video, right? I mean if you think about the world today, we now have models that are I think frankly surpassing VO3. I mean it keeps you know they keep leaprogging each other and the entire you know nature of AI is so deeply embedded in our lives not just as artists not just in media but everywhere. It's clear that it's going to have a massive impact on society. Um you know Ari will be joining me a little bit later in this talk and you know a lot of this is about how this affects jobs, how this affects everything around us. But you know I want to start with focusing on rendering itself right I mean if we're looking at things from you know the visual effects space the CG space 3D design space you know the trends in the last year some of the things we predicted I think we're were right on like I had this slide about uh MCP which is the you know API or the connective layer between applications right which anthropic um created for cloud and uh I showed blender uh you know being driven by uh cloud through MCP And you know, we do have Francesco, the CEO. He's now the CEO of the Blender Foundation here today. And it's it's interesting because this pretends a world where we don't have apps. We don't necessarily have dcc. We don't even have web services. It just feels like we're going to end up in a place where your agent or the fluid interactive layer that you're connecting to uh is basically creating the user interface uh for you. And of course there's going to be a lot of foundational tech like Blender for example could be creating a good chunk of the scene graph but a renderer for example wouldn't necessarily have to go through a API it could be handled by another agent. So this is the future that is um definitely going to be in the hands of artists I think this year. Uh the other prediction that I think and this is all of course AI based but we have real-time rendering right and this is uh in octane 2027 finally after promising 12 years we'd crack how to get path tracing with no noise uh it turns out that you know a bit of a tensure cores in a GPU uh cleans up the noise tracks it and you know that's how we have uh octane 2027 and it's something that uh without AI assisting rendering right we wouldn't be able to necessarily get to work. And you know, we shipped, you know, if you think about AI being part of the rendering equation, it's been shipping in Octane since 2017. It's almost 10 years since we shipped the AI denoiser. Uh, and so there's, you know, the whole conceit of whether or not AI is going to replace rendering, whether it's in odds with it. I mean, it's just always been part of the stack. It's a question of how much you're generating versus creating and using machine learning to assist you. Uh the other big uh step function which I I've been hoping for and it certainly connects to our vision of the holiday is I do feel like everybody now knows what a gazian splat is right everybody understands this is a format for point cloud with reflections everything can I you can drop this now into a web browser it's going as of yesterday I think my team just told me that it's in uh glTF and USD so it's a standard and it's great and the thing is one of the you know goals that I set for the uh the octane was I want to be able to render not into a picture, not into a movie, directly into this 3D volutric format because that honestly is is how rendering should work. We can direct the rays into doesn't have to be a plane that matches a film camera from you know designed in the 1830s. It can be uh something that is shaped like the actual object that you're you're creating in this point cloud. So I think having 40 gazian splats which are you know temporal movie based volutric um media is is a given and I think you know our render and others are going to be targeting that.

[00:04:30]So you might end up in a place where the you know the media that you store on your phone that you work with is going to be volutric and I think that's a good thing for all the different uh pieces of the ecosystem. So summing up these three things I would say yeah the uh future of rendering is definitely neural. Uh, I believe Jensen's prediction of every pixel will be neurally rendered is probably correct if nothing else from DLSS 5, which for those who don't know it, it renders over a 3D video game, right? Even old ones. And it adds AI lighting and texturing. And of course, people did not like that. They they reacted negatively. Um, but that's something that is Yeah, it's true. I mean, I think that there's a lot of of concerns about how AI is affecting uh ownership and IP and creativity and and whether it it you know, it's and all of this is these are fair things. On the other hand, if you think about what I was just saying that it's been 10 years since we've had machine learning as part of rendering and there's always going to be a spectrum, it's just a question of, you know, moderating uh the good parts with the parts that are, you know, definitely not great. Um but on the render network side I think certainly as as we have neural rendering which to me is like you know 3D assisted um you know machine learning rendering you know I think work's going to be increasing significantly um one of the uh partners that we're bringing here is LTX light they uh do an open source um generative video model and definitely looking at other models as well and their models not only can be run locally and fine-tuned but they can run on consumer GPUs which is a big deal. We've had beautiful work uh done in the last year.

[00:06:02]Um you'll hear from many of these artists today and tomorrow and uh there's a lot of great artwork all around the uh the event today for everyone to check out. Um so it's it's again it's a great day to have um you know such uh you know a great focus on on all this great work. Uh I I want to talk about something that I touched on last year and it was sort of more of a predictive thing which is you know we're we're now used to these models that can create 3D assets and videos and images and of course LLMs and then there's world models right which makes sense.

[00:06:35]It's it's like well that's your AI game engine or video game. It it can basically instantiate all the laws of physics and rendering and everything else in real time. And the way that it works um today is is simple. I mean the ones that are out there and we'll show some examples are just the video models.

[00:06:51]They're just running fast enough that they can you know you can interact with them. So as you're generating a video on like runway for example or Genie which is basically VO uh running on an LPU with a 2C delay uh you can tell it you can steer it and that basically becomes you know as that latency gets closer to 40 milliseconds it becomes a video game.

[00:07:10]But here's the other thing and this is why this slide is relevant. um you know there's there's this thesis and I'm not sure whether it's correct or not that LLMs are going to hit a ceiling that there's only so much that you can scale this and if you have an AI that's trained on the real world even a virtual version of that um it can definitely possibly have a lot a greater degree of um understanding and modeling and imagination to to solve harder problems.

[00:07:36]So that is something that of course extends I think beyond our use case for world uh models which I think for us would be obviously if you have a 3D engine that's in real time and it's uh driven by all the um benefits of having neural rendering that could be amazing for being able to control some of the trickier parts of generative AI and also of course having um games and simulations that are not necessarily tied to polygons and geometry and volumes. So it is a um I think there's a lot of investment around world models.

[00:08:03]you're seeing companies um like the trends raising billions uh to pursue this and I think that it has a dual use right it's something that if if it does lead us to a um a deeper uh smarter AI that's that's one path but I think that on the way there we're going to get um really good simulations and rendering and we see that already the video models that we're seeing today are really powerful the first world model uh Gen3 and I mentioned that to my understanding this is basically Google's video model just running with a uh a real you know for 60 seconds and you can talk to it and you can send it arrow keys right and and move the camera through it and you know when I when you see this it does give you a sense of how quickly this technology is approaching we are going to get games that you can create and and we'll see how good those work but it is something that's um that's interesting and it's really a question of speed so uh we'll hear from Richard Carson from Nvidia in a little bit um and ver Rubin is their next GPU I think it's coming out next year and you run video models on at GPU they're basically running in real time. So this is all going to be something that I think is you know going to be in our uh near future.

[00:09:09]>> Each one of these interactive environment generated by Genie3 a new frontier for world models. With Genie3 you can use natural language to generate a variety of worlds and explore them interactively all with a single text prompt.

[00:09:32]Let's see what it's like to spend some time in a world.

[00:09:36]Genie 3 has real-time interactivity, meaning that the environment reacts to your movements and actions. You're not walking through a pre-built simulation.

[00:09:44]Everything you see here is being generated live as you explore it. And Genie 3 has world memory. That's why >> I'm going to pause it there because the world memory piece is is the tricky part. Um it I've tested Genie. I'm sure many of you guys now that it came out have tried it as well. Um and and persistence I mean you have LLMs that hallucinate, right? You have it's worse with world models. They will forget the world. So what's interesting is you can have a world model generate the world once like basically just follow through which is basically just creating a video in all these different paths and uh you know we partner with this company um world labs uh it's on our our our cloud service. you can just get back a gauian splat and that loads into Octane. And the gauian splat is really easy to generate out of you know an imagined uh scene. It can also be of course used for creating things out of photos. But um this is probably a more useful practical step. Uh although of course if you can get this in real time and dynamics and everything are are there great. But for right now world models are at least useful in these pipelines. And you know what's more interesting than just creating a whole cloth is being able to edit uh models or edit point clouds through machine learning. Uh but sort of seeing how well it's able to reconstruct something like typical photography use cases. Uh the Nomad in downtown LA were looking at it for um for a project and unfortunately they changed the whole thing. They sold it to a different owner. So this doesn't exist anymore.

[00:11:10]And you know it was interesting to feed just the top left image. There's a Genie 3 running in my browser and it has pretty good uh spatial consistency. So, this is this is quite promising. Uh it doesn't always work this well, but it definitely is, you know, is my it was my first sort of experience with something that felt, you know, pretty solid. Uh again, if you run the same thing through world labs and you get back the splot that's running locally, it's super fast.

[00:11:36]I can completely control it, but you lose the dynamism. But having both is important and ultimately if the engine itself is running locally let's say we have a Vera Rubin GPU in our uh you know at least our desktops and these things can all be running in real time that becomes our 3D render maybe we don't ray trace we don't rasterize anymore uh and also again this is a splat even though you know we we did some work to get it to run inside of Octane you can ray trace into splats you can sh you know shade them and light them so they're no different than any other type of 3D model um especially when you consider point clouds and and SDFs and other types of things. But uh it it does allow us to capture and and I think this has been a problem especially when you're capturing 3D characters uh like in a basketball game. Splats are a great format. All the reflections, all the uh things that polygons don't map too well are are covered by this. This is another example. Uh this is Genie 3. Uh and it's again I mean when you're going forward in the scene, it handles it pretty Well, I mean, I think any generative model if you give it a first frame will be able to track that to the next one. Uh, but I' I've basically found it I mean, I don't think Genius3 is useful in this in this current state. However, being able to save the state of that world and use it as a 3D model is promising. And there's other uh world models. Now, there's open source ones. Uh, this one is from Tencent. Uh, this is me trying it out. I mean, it's impressive. I was I found it even better than Genie 3. And we're seeing uh the open source video models like WAN and now LTX uh being adapted to become faster and faster so they can yeah they can work in real time. Uh and I think that there's there's a lot of of of unknowns. You know I talked to a lot of the folks in this space and the feedback I get is that in 90 days they don't know what you know and this these are people working on top frontier models. They don't know in 90 days whether you what's going to be going on with them or with their competitors. Um, so I always go back to John Carmarmac who who would have been here I think he unfortunately wasn't able to do it because he had a cruise but I did I did send him uh you know the feedback from last year and I got this long email back from him and too long didn't read is that uh he doesn't know you know I mean he told me like 10 years before you can totally replace normal rendering with um neural rendering but it is moving faster and there's certain things that are driving that you have um you know not just GPUs but LPUs right which are inference specific hardware that Nvidia did not necessarily have. Of course, now with this arrangement to invest in Gro, which is was making the specialized hardware now that is part of Nvidia stack, but there's always um accelerance and you never know when the models themselves can just leaprog into a whole another level. I mean outside of our space um with Claude Opus 46, you start to see code that's being written by uh by AI that is quite good. I mean, it can definitely, you know, keep up with, you know, pretty high-end programming team.

[00:14:29]And I think that's something that's useful even for us to consider for artists. If you can talk to your application and it can build interfaces that are specifically for you um with this same level of skill that a coder or plug-in developer for that app might have, that's pretty powerful. So, we are definitely looking into that. Uh, and this is John's full, you know, description of it, but it's it's as I described it. I mean it's definitely um you know we are moving on a very fast path with a lot of variables but ultimately I think we're going to end up seeing in this year a lot of pretty decent attempts at getting us real-time neural rendering and certainly if it's running over a 3D engine to begin with it'll it'll do quite well. uh you know I I think as far as you know use cases for this stuff I mean the stuff when it's running live is powerful because you can use it as a brush you can use it as an artist right you don't have to worry about whether the the slot machine is working well or not you can basically use your existing 3D pipeline or run video and edit it and this is uh this was done by an intern I think at Adobe that was uh you know they fine-tuned WAN to be able to deliver this in real time but it's great I mean if you actually can you know see what you're doing in real time a lot of the guesswork and inconsistencies with AI I might be improved and ultimately if you're driving a 3D scene graph underneath which is where I think this is going then even better. Uh so we're we're nearing that point with uh with both the software and the hardware this year. Uh as far as in production so I was mentioning that we have um a team from LTX light uh that has uh has an open model and David you actually posted this uh video. It's basically taking your 3D um root shaded viewport and rendering over it and it looks quite good and you know there you have full control over your scene. It's just the rendering instead of being handled by rate tracing in Octane is being handled by this uh this neural rendering model and that's legitimately a way of rendering your 3D scene. Um and ultimately when you consider that yes we are rate tracing and we have neural radiance cache and all these things. I mean we are using machine learning ourselves. Um, our goal with at least within Otoy and connected to the render network to productize these things is to offer everything and to find a way to link these pieces together in a way that's smart and helpful and useful. Uh, right now it's on a it's on otoy.studio. Um, and if you're, you know, if you've ever used or or of course render um it's something that we built on top of our existing cloud services. It includes things that I think are sorely missing like you have a 4D uh you know splat timeline so you can do you know holographic videos built into the browser. I think those are kind of essential. You'll see some of the workflows in fact um don't have to use it on the web. We also are working on support for Blender. Um and these are some examples of things that were done.

[00:17:09]um you know it hasn't been up for very long and some of the models that are used for these things have only been out for like 6 weeks but I did bring some examples and we have one artist um who I plucked out of the ether uh to come here today because I thought his work was really compelling. He's one man shop and he did this Star Wars piece. It's a Star Wars clip that uh fills in the blanks between Empire Strikes Back. It's a fan fan-made thing but you can see some of the images here and he did it on Otoy Studio. He's using Clling which you know will be on the panel with the Cling folks later. um his background is in um visual effects. His name is Darren and I'm excited to you know to sort of have him talk about how he approached these things and also it's been good to get a bit of uh feedback on the service. So on the site now I don't know if you know for those that have been checking it out it now has much more 3D stuff. You know it was initially just a bunch of you know generative image and text models but a lot of the workflows that we're focusing on understandably are are 3D models right 3D rendering. So again, you want to generate a splat or a mesh, that's what this this thing's for. Also, rettopizing, editing, assembling, all these things. I mean, ultimately, this exists because I do foresee us moving off the desktop into these sort of blended services, machine learning or not. Um, and I guess, you know, when when you look at how these pieces are put together, it's worth seeing. I mean, he did 13 minutes for this piece. I'm only show a couple of minutes here, but uh you can see sort of how he's assembling characters. Bubba Fett, there's a character that was never done in live action. the assembled app I think from the lunchbox. And let's uh let's see if we can play this clip. Um actually I think this is the um this is where he makes a little video showing I think the uh uh you know sort of the node graph for the for the scenes he rendered. But consider that this was done in uh three or four weeks. um a lot of shots in there and and to be fair, I mean the the feedback that I experienced and I think he did and many others is that when you're generating something uh and it's not consistently bu built on a real thing, you're going to run into a lot of issues. You you know that's really the trick here is how do we get to back to a 3D rendering the way we know it with the benefits you get. I mean it's certainly cheaper to light with neural rendering, right? I mean it just it it figures out the light transport without having to shoot all the extra rays we have to with a rate tracing card. So, you know, this is um this is definitely again I mean the models that he was using uh the the references you have when you can send a reference character into a generative model that can be from a 3D uh render for example with multiple uh angles, it works quite well and it starts to feel like we're getting closer and closer to being able to treat these things like a 3D scene graph. So, I'm going to share um a couple of clips from >> and we can check it out.

[00:19:45]>> Certainly doesn't seem like taking the place by forces.

[00:19:49]We all want Captain Solo back, but we just don't have the resources for a mission of this magnitude.

[00:19:58]>> Prince Shehor is very powerful, very dangerous.

[00:20:03]>> We cannot afford to divert what little strength we have left to hunt a single man. We can't just leave him behind. We have all made sacrifices, General.

[00:20:14]I assure you.

[00:20:17]So, it's 13 minutes, but this is a couple minutes of it. And it's I mean, it it's quite remarkable, I think, to see. You can go to the O toy studio website. Anybody with the wherewithal, I mean, I guess could do this, although I think it requires, you know, I mean, Darren's background, as he'll explain, is cinematography and things like that.

[00:20:34]But it's it's really interesting to see this. And this was done, I think, with Cling 03, which didn't exist in January.

[00:20:40]So, um, you know, again, that model, and I that's why I brought the cling guys here. like you can create a 3D render, you can create references. That changes everything. It's no longer prompting.

[00:20:50]It's no longer image based. It can be much closer to 3D uh rendering as you'll see some of these other models that we have as well. You know, you don't lose the 3D stuff either. I mean, you can basically take any piece in in any of these pieces and you can send in a 3D model, you can pull out a 3D model, you can create a splat, you can, you know, get out a splat. All those things are part of my vision for this. uh because otherwise it's you're never going to get to the uh the kind of consistency and and reliability you need. And you can see here, you know, inside of this web paging, you can move things around. You can get splats. You can turn every single frame you're generating or every video you shot into a reliable 4080 splat. You can load that in Blender, download it, rerun it. So, I'll be showing that as well as the clips that were that were built for this piece.

[00:21:33]This is another one. And I'll show the >> force be with you, >> the relighting stuff after.

[00:21:46]again.

[00:22:06]>> I'm telling you, I haven't seen him.

[00:22:08]Keep talking.

[00:22:10]>> I thought you and Boba go way back.

[00:22:22]Little boss. Are you crazy doing that in here? D shetti.

[00:22:44]>> He said before he gets here, he had to stop in the last place in the galaxy he should ever be.

[00:22:50]That's all I know.

[00:22:52]Salar Mitchka twom.

[00:23:07]Keep your filthy credits.

[00:23:11]I may not make what you call an honest living, but I'll never dance for a twobit gangster like Jabba the Hut.

[00:23:22]to my day.

[00:23:31]>> Carver Russell.

[00:23:33]If you read the comics, you know whose character So again, there's I have one more clip I'm going to show. We're going to talk about this later today. But you know, some of the things that go wrong when you're relying on a neural render or these video models to to finish the pieces, if something is missing, how do you edit it in 3D? You just rerender it.

[00:24:08]So we're trying to solve that. And first thing is every single frame whether it's coming from a gendered model or a video what you can pull that back out is not just a gauian splot but a relightable gauian splot. So this is I have the entire 13 minutes pulling back into Octane. I can relight it. It's I can separate the things. It's a 3D model effectively um as as a spot would be. Uh and and of course in Octane it's got you got path tracing now it's real time thank you to you know the latest GPUs.

[00:24:35]This is in Blender. So, uh, if you want to go use, you know, and I think Blender is amazing because there's such flexibility in that tool. You can bring the full 13 minutes in and, you know, again, all of that's there. Uh, and of course, you can render it with Octin from from within Blender, but it it it is a 3D asset. So, everything that you have in in some form, at least with the pipeline that we're envisioning here, is still in 3D uh with relighting, with dynamics, and other things. And we're just at the beginning of this, but if you have that, then editing things, cleaning things up, um, a lot easier.

[00:25:07]And, uh, let me play one last >> the odds.

[00:25:32]I want to tell you what happened.

[00:25:40]>> Me too, Chewy.

[00:25:53]Where are we going?

[00:25:55]The Galactic Center. Sounds crowded.

[00:26:31]So he did a lot of work on the film grain, wanted to make it look just like the 70s. Um, but again, you know, when it came to like, you know, as he was, um, I think coming out today or putting on YouTube today, there were a couple of shots that he was like, "Oh, I want to re, you know, redo this." But again, if you have the ability to pull in the actual frames, edit them, and rerender them as that you would in 3D. That's a huge improvement. So, that's something that, um, I think we want to, you know, keep building on. And, and I think for virtual production, this is a big deal as well. We did we did this one of our Star Trek pieces. We're doing these shorts. Uh four years ago, we did this shot. It was from the comic book that we wanted to to render and it didn't work.

[00:27:08]It was like the tracking was difficult.

[00:27:09]It was just it just you know we have the CG model uh we had the scanner the actress and it just but at the time it didn't make any sense. Again, we have a new model, a switch light that that's now on the service. And what's fascinating is I don't know if these guys knew this when they did this, but if you if you have the 3D model and you feed that as reference, uh you end up with something that works quite well, better than just green screen compositing, right? So, this is an actual green screen shot. We have a static mesh which we captured on our light stage. We have this scene that was done by Aaron Westwood and Octane for Cinema 4D, right? But combining all those things was a no-go and it was expensive because you're you're losing one or the other. You're not going to mix these things together easily. So feed it into the one node and 30 seconds later this comes back. Green screen is completely cleaned up. The tracking is completely done. And the CGC scene which I know this scene well because it is a CG scene is actually referenced correctly. And you you notice right the weather, the ripples, the reflections, it's all there. From a purely compositing workflow perspective, this is a big deal. And it's something that I realized as I was just testing the shot out. I was like, why not? uh how far we've come with uh with some of these things just for normal production work.

[00:28:20]So that's something that's pretty exciting. Um I want to close out before we bring in Ari uh to talk a little bit about the content and of course you know we've we've been working on the Gene Rodenberry archive for for many many years and Rod Rodenberry is here. We're going to talk about that and Gene Rodenberry's vision for the future. Um but I will give an update. You know, I keep I in my previous presentations, I talk about how we're building the one to one life-size enterprise, you know, every deck and we are actually making progress. I think we're about a third of the way through. Um, and it's interesting because of course, you know, I'm thinking about, okay, you got world models, you have Genie, what if I just give it the um, you know, this is our step. This is our 3D render, right? And, uh, you know, let's see if you if you if you all it can accept is one frame. So, I was like, okay, Genie, let's see what you And it's pretty good. At first I was like, "Wow, this is like, okay, it understands space and you know, we're I'm looking through this and I'm like, well, wait a minute, there's oh, what what is going on there? Oops. You know, and it's like and this is the the thing is like the generative model is not it it doesn't have memory that that keeps track of things." Like granted, if the you know, if you run out of memory in a in a longer tail, if if that's addressed, it still is not the same as having an actual 3D model, right? I mean, it just drips. The further it goes, the more of a fever dream it becomes. So, uh, world models have issues. Uh, you if you want to rely on these things to reliably give you a world, um, you might get this. This looks like it's the Ubisoft game. So, you know, for a whole different era of Star Trek mixed with Kirk's Enterprise.

[00:29:48]So, I feel like, yeah, there's there's there's issues that need to be solved and therefore we go back to 3D rendering and 3D modeling and maybe mixing those together is is the scaffolding that's needed. So, um, you know, with our work, the team is the RI archive team is doing incredible work. I mean, we are literally it is a world of on that ship and it's going to be it's going to probably a couple years, but we're going to be finished with it and deck by deck you're going to have that that vessel there. And this is a beautiful video the team made um which I want to close out by sharing with you guys. But look at that. It's it's gorgeous. And this is all um in real time. It'll be on the website. Um and uh it's it's definitely I mean the idea of course having simulated worlds and having all of this at our hands is is a dream that I think you know Gene had with the holiday right going back to um you know even TAS. So this is great work and I think that you know that I I think the lesson for me is that there's a lot of assistance you can get from AI. for sure going to be changing the ability of any artist um with any existing workflow to do things faster and possibly also to give artists that don't have um the tools to do the things they've dreamed of more power.

[00:30:59]But I think that there's also limitations and understanding those is really pretty critical. So we're uh you know we put out a a piece two years ago.

[00:31:06]It it it's you know been a really a welcome thing on top of doing the interactive experiences. We got 38 million views. had a nice shout out from the new owners um of the of the studio and um you know we're we're excited to just expand uh what we're doing sort of with archiving more uh work from other creators besides Gene Ronberry for about the same day that we announced the Gene Rodenberry archive publicly. I also announced the Alex Ross archive. Alex has been he's a great comic book artist probably if you're a comic book fan you know his work um and we've been working for many years on that as well. So in our talk later couple of hours we'll be discussing the Alex Ross archive and uh again just beautiful work. Um I'm going to close this uh this presentation out with a article from 2004 uh October 2004 and it's interesting this was the first time that anybody had ever like heard of Otoy. I was living out of my mom's house and somehow the reporter found me. But it's interesting that very same week um a man came to my house and I didn't know how I still don't know the story exactly how we found out what I was doing but it was before this article came out but it was that week in in uh in 2004 and his name was Ari Emanuel and he's been with me every step of this journey on the way here. So it is my great pleasure to welcome Ari Emanuel in person my good friend to the stage.

[00:32:26]Thank you, Dorian.

[00:32:34]>> God forbid that somebody does the math on 2004, but um yeah, 2004.

[00:32:40]>> I mean, it's been >> It was a It was It was shag carpeting.

[00:32:43]>> It's still there.

[00:32:44]>> Computers all the way to the ceiling.

[00:32:47]Him um having uh the Star Trek universe run through.

[00:32:54]>> I don't even remember.

[00:32:55]>> It broke through the web browser. That was the magical thing. I took a out of the frame and it gets lit by the >> I'd never seen anything like it. I said, "What the [ __ ] are you doing?"

[00:33:04]>> So, >> yeah.

[00:33:05]>> But now we're here.

[00:33:06]>> Now we're here and and >> and Jules was actually talking about what the world would kind of look like.

[00:33:12]Um I don't think you you were talking about GPUs way back then.

[00:33:16]>> Yeah.

[00:33:16]>> Yeah. I I was, you know, I don't know if you remember, but in around 2008 or 11, I was like, I need to either focus on GPU graphics or GPU AI, but the AI thing is too much. Like, it's going to kill me. Graphics are more tractable. But I always figured, okay, you have thought vectors, you could build something. This is six years before that paper.

[00:33:33]>> I don't know about the rest of you. I still only understand about 5% of what he says to me. But >> you know, you you know more about GPUs now than I think a lot of >> probably. Yeah.

[00:33:42]>> But but it's also I mean it's now GPUs are everywhere. It's like it's Jensen's the, you know, world's most valuable company and it's like gold and it's it's crazy.

[00:33:51]>> So, what do you think about GPUs to CPUs?

[00:33:54]>> I still think GPUs win. I mean, it it just made me leaprogged by something more specific for AI because that's really the prize, you know, and that's what these LPUs are. But then Jensen was like, I'm just going to buy the, you know, or or pull in the LPU company. And now that's >> $20 billion later he's got that in the in the stack, >> right? But it's I mean if it weren't for AI I don't think that that Nvidia would be you know four to five trillion dollar company. So >> I agree with that.

[00:34:19]>> But I want to talk a bit about you know obviously from the creative perspective I mean you're seeing all this stuff right up close with movies with what is what is happening with the world with all this.

[00:34:29]>> I think what you're going to have now is kind of a revolution in content based on what we've been talking about what I'm seeing. um you'll have the the marginal cost of production by implementing a lot of the stuff that you're working on and the and the stack that you're building inside render and otoy >> the cost of production will come down >> um >> um I think IP is going to be very valuable new IP is going to be very valuable >> but then because the cost is coming down I think there's going to be more >> um production um And I think that's just good. I've, you know, since we met and I introduced you to George Gilder and >> Life After Television and, you know, infinite distribution.

[00:35:18]>> Um um uh I think because costs are coming down, instead of a movie costing$125 million, it's going to come down significantly just because of the production and all that. And then you're going to have a lot more content. And I think a lot of the services are going to start now competing with YouTube because if they can do a short >> Yeah.

[00:35:43]>> for those costs, um there's going to be a lot of creators out there putting stuff up. They're gonna the Netflix, the Parammont, uh Warner, the uh um Amazon, Disney are going to build their services for vertical, which is on phone plus um you know uh um traditional television and movies. You're already seeing um coming out of China now here, you know, the one minute dramas.

[00:36:11]>> Yeah. um that will kind of I think a lot of AI will come into that because the cost of those things are $100,000 for 16 to 20 episodes. Um they're starting to use AI in China on animation for those.

[00:36:25]>> So I just think there's going to be so much more production um because costs are coming down. Now will that disintermediate some people that are in the in in the system now? Yes. Um but there's going to be a lot more production. Yeah. Now it feels like you know even within you know the the circle of folks that we work with I mean okay well maybe you don't do rotoscoping anymore but you learn how to generate worlds and things like that. Yeah, I mean if you're having a creative event, you can adapt. But I think in art space in particular, there's all it feels like if if there's an infinite amount of content that's that's in demand and there's much cheaper ways of building it or the tools are better or smaller teams can do it, right? You know, and I think that's something that in at least in this sector my my plan >> I think also um depending on the um the movie the type of movie you're making.

[00:37:15]So if it's a sci-fi movie, action movie, >> you not only will be making the movie television show, you'll simultaneously because of the assets >> Yeah.

[00:37:25]>> be making games >> like and as opposed to >> the lag time lag that you currently have that will the release schedule will be different and you can time that out. Um, and um, I don't yet know, and since I'm an owner of sports, I don't know the effect on sports. We're starting to implement a couple of um, uh, features inside our sports, which is the UFC and the WWE and bull writing, and I know all of you are fans of bull writing. Um, but, um, how uh, that starts and then improves the interactivity. Hopefully, it does doesn't only go to prediction markets and gambling.

[00:38:09]>> Yeah.

[00:38:09]>> Uh but it makes also the experience better. Um we're working on that. I think the NFL is working on that. Uh the NBA is working on that whether it be AR, VR, and a bunch of different but that will come too. There will be some kind of leaps uh in that direction.

[00:38:26]>> Yeah. I mean, the tech that I was just showing which is like kind of only a week old where you can just we can just take things and just fully turn them into 3D models.

[00:38:33]>> Yeah. relight them, basically almost turn them back into production assets. I mean, it's great if you're generating something that's from a 3D asset, but also games, NFL, all that stuff. I mean, it just feels like when VR, remember when when Facebook first bought uh, you know, the Oculus guys and they were going after NFL and all these things. I was like, all you really need is not cameras. You just want to understand what the players are doing, have a, you know, sense of their their bodies and rebuild that in 3D. And I think we're we're going to get there. The question is are they going to >> that you're going to have >> as opposed to um one is I think the most important thing is great ideas.

[00:39:07]>> Yeah. So you see with this gentleman that that that made the the Star Wars thing I mean that's his take on it that can affect what a traditional guy's takes on it and that conversation just creates great but the other thing that it does it keeps a piece of content if you integrate short form mobile games in the zeitgeist >> and if you open that and you let the world kind of play with it. You don't know where the story line's going to go, which is kind of exciting. Um, and I think the on the studio side that owns a lot of the IP, if they start building that into their process, that short form can come up, keep it the story lines going, keep the um audience engaged on a full-time basis. I think it just makes it better.

[00:40:03]>> Yeah. Well, I mean, you know, we we've experimented just with no toy doing shorts, which I think is, you know, fine. And also, we we can, you know, we also were able to experiment more with the technology, but it it does feel like there's there's just a whole different market than there was 10 15 years ago for these kinds of short form pieces of content. But as far as playing with them, I mean, it's like, you know, the world of Star Trek and Star Wars, I mean, it like this stuff, uh, people love that world. I mean, there's like 10,000 3D artists building Star Trek ships for fun. Um, and I think the same thing with, you know, writing stories that connect to your favorite thing of lore. So, there's definitely as these things become more >> interactive as you're making the film or the story. I mean, that's the promise of a role model is you make the, you know, you render it. It also could be interactive. It can it can be smart even. I mean, there's all these possib >> I mean, I'm assuming people here know, but if they don't, you know, in the television business, you have a whole writing staff when you're doing a a series.

[00:40:55]>> On the movie side, you know, you got a writer, you got a director, sometimes to write a director. This kind of gives you a a creative staff if you want to start early where >> breaking the story, you can you can enable the audience or the the kind of fans to kind of come in and help gener.

[00:41:15]I think that eventually is going to happen.

[00:41:17]>> Um, and as these models become I mean I think the >> the one you send me the Star Wars one probably only costs a couple hundred thousand dollars.

[00:41:28]>> That's an incredible development pipeline.

[00:41:30]>> Oh, that's it was like hundreds of dollars. I mean it was, >> you know, that's an incredible, you know, kind of pipeline in January.

[00:41:39]>> Yeah. For story development. Um um and I think that's going to start I hope get incorporated because I think it's just going to add to the creativity and the story lines and everything else.

[00:41:50]>> Yeah. I mean I think and and also you know the the if you want to have I mean I I see AI productions where they're still using actors or there's voice over and it becomes a lot like just an animated film. Uh so I feel like as long as there's a story that's compelling that you're telling and and all of that, it just becomes another tool or medium.

[00:42:07]But my prediction again is that it's just going to end up being very much like CG in the end with some of the same, you know, with maybe greater power and and faster speed and fewer costs, but still um it's going to be, you know, different skills and certainly different creative minds. I think you're going to have different outcomes.

[00:42:22]>> One of the things I wanted to ask you kind of because I know you've been working on for a while, the kind of um the 3D models for Star Trek.

[00:42:30]>> Yeah.

[00:42:33]>> How long will that take in the future?

[00:42:35]Because I think then you have to come into every single studios and build the models for some of their big IPs.

[00:42:40]>> Yeah.

[00:42:41]>> Um how long does that take?

[00:42:43]>> I mean to build the quality we're doing it's for the ship like these size the Empire State Building it's like five years because we have one or two artists on it.

[00:42:50]>> If you want the AI to do it for free you you will get missing chairs and you'll get the wrong you know I mean things missing. But so there's there's a there's an immediate return of a 3D model or 3D scene that you can now try on on on some of these services. the quality of the actual thing itself I think is is still requires you know a human to check it but you can save that you can build it you know piece by piece and I think the tools to build that are going to be faster like a lot of the busy work and rotoscoping films or even 3D modeling with texturing all that stuff I mean all of that I think could be accelerated so if you're a solo 3D artist you can just get a lot more work done um but there might come a day where >> didn't you didn't you do that with the batcave >> well yeah we did it so with the team I've been working for 30 years we like two or three bat caves like that are perfect like same quality of the Star Trek thing and Warner Brothers thanks to you know invested in us as did Disney and I think that you know I guess now they're going to you know we'll see what happens with the uh the acquisition but you know it would be I think there's building the backlot of every valuable IP certainly the ones that we love and our team has this this affection for I think is great and I think you know Starship I mean four million people visit those Starships we built just online >> so those are going to be like museums >> well yeah and that's the thing is that the real thing is in the Museum of the Smithsonian. But I think the the more interesting part for me is well now you can see a museum of the actual show.

[00:44:08]Like this is what Kirk is doing in this year. You can skip ahead 300 years. The whole story like George R. Martin Game of Thrones or the DC Marvel stuff. You just see the world that's that's really rich laid out in a way that is not necessarily you know easy to experience just through one you know lens of an episode. So all of that I think becomes possible. That's always been you know the dream that I pitched for for Otoy from the the start. I think that's going to happen. And then then it's easier for them to you can plug and play. Oh, we're going to create the game or oh, we're going to create the game in the movie or oh, we're going to create the game in the and and the series and they're going to play off each other. Yeah.

[00:44:44]>> Um and you know, once you have that, the cost uh to go into production is a lot cheaper.

[00:44:53]>> Yeah, for sure. So >> I also think in some weird ways the IP issues I mean I wouldn't say it's sorted itself out but you look at the thing with seed dance right I mean I think a year ago I was saying you know in China and Singapore there are no real protections against IP you have these Chinese companies that are doing this stuff but in the end um you know it's not like you can stop the AI being trained on things it's very hard but you can stop it from outputting this thing you can hold somebody accountable for that and that clearly is what's happened with Cance because it was it freaked a lot of people out and I think that is probably the way that that you know in some way whether you're trying to find deep takes around. I mean, it feels like if you're responsible for putting out something that you're not supposed to for commercial purposes, you know, that's something that at least um it worked all the way to, you know, over in China with SEAN. So, maybe that's something that helps that part of the uh the business out. I don't know if you're you have any other thoughts on that.

[00:45:40]>> Well, I mean, we we started working on light stage >> y >> um uh you know, rendering actors so they get to control their digital >> um images. Yeah.

[00:45:53]>> Um and um one they can capture it at a certain point in time and two then they can license it out and know kind of where it goes. That might be a solution.

[00:46:05]>> Yeah.

[00:46:05]>> But the quality we get from lights is very different than any other method.

[00:46:08]But so that is you know ground truth.

[00:46:10]But yeah and and I certainly think that unlike I think when we were first going down the path and like >> those all those issues are going to cuz I don't think anybody solved this problem like who owns if you did a movie who owns that IP of your image >> we found out with the Roth remember started because because of you know five movies in you he was like I think Disney owns Jungle Cruise I think you know Universal owns this one you know and it's like well he should own his own face and licensed it to them. So >> yeah.

[00:46:39]>> Yeah, >> I that that that also will be sorted out in the next couple years. I don't think it's solved yet.

[00:46:46]>> So >> yeah, it's heavy times. So >> So >> all right.

[00:46:50]>> Well, thank you.

[00:46:51]>> Well, thank you, Ari. It's so great to have you here in person. Thank you.

[00:46:54]Thank you, Ari.

[00:46:56]Be good. Thank you so much.