Moonvalley CEO on building AI filmmaking tools trained exclusively on licensed data

Have a good one. Cheers, Richard. Thanks for joining. Good chatting. Uh next up we have the founder of Moon Valley coming on the show. Um uh talking about uh generative imagery. We're going to pull up our website. Very very cool stuff. So uh welcome to the stream.

Hopefully we can pull up this website because we have fully passed the uncanny valley. Don't you agree? I mean this Are you Are you on the website? Can we pull up the website really quickly and show uh the the the video? So, so everything on your website, this is AI generated. Is that correct? Uh, that's right.

There's some um like after effects stuff on top of it, but yeah, it's basically all AI Mary generated. It's remarkable and I feel like it's under discussed. Anyway, uh please introduce yourself and the company because this is Yeah, absolutely. Uh thanks for having me. Great to great to meet you guys.

Um great to meet you. So, I'm Naim. I'm one of the founders of Moon Valley.

uh you know at a very high level we're a team of kind of a unique structure you know a good chunk of our team are worldrenowned researchers um in visual intelligence primarily but you know our our focus uh in a specific level is is we're building the biggest and the most capable production grade generative video models um and on the flip side is we also have a large chunk of our company are filmmakers so we have um we have a movie studio in LA it's it's one of the oldest most wellpreserved soundstages in the world.

Um, some of the first Charlie Chaplan movies were shot there. Uh, and so it's like kind of a ground zero and and we have, you know, folks that have won Emmys that have been nominated for Oscars on the team.

Uh and and so we've just kind of brought both worlds together to figure out how do you take this tech from being you know kind of interesting research uh and and um you know cool things that you can share on on X and actually become things that sort of push the boundaries of you know what we sort of we think of it as like movies and stuff but these are sort of limited abstractions more of just like visual media broadly um and and kind of the artistry around it.

So, how do you think about the trade-off between training new models, being a foundational model company, hiring researchers, huge training runs, and then the application layer, the distribution, actually working with filmmakers that feels like most companies have kind of split between one or the other?

Are you doing both right now? How would you describe the shape of the business? Yeah, you know, it feels a little bit maybe it's like a bit of a post chat GBT phenomenon, but I I do think that increasingly foundational companies quote unquote are having to think a lot more about the application layer.

Um, and and I think that that's only going to continue to be the case. Like I think a foundational research, which you know, and you kind of alluded to it in in your chat with with Richard, but like there's an element of commoditization that's happening, right?

there's an element of of um you'll get to there's a point uh as and this applies to visual media as well where the the output of the model um you kind of hit this point of diminishing returns from a consumer perspective from like a user perspective.

So there might be really interesting kind of research thing that's happening um and you you'll continue to invest in that but to drive the same amount of like business value as like you know a GPT4 style leap that requires kind of thinking about other other areas.

So um and and the other piece for us is I think it's it's different with things like LLMs but with with visual and video in particular I think one of the issues is that like uh research labs and and technology companies that have been in the space they have been largely divorced from like the end practitioner uh in in in a lot of ways and for LLMs that are so such general technology I think that makes sense but in our case you know we're building tooling that filmmakers ers will ultimately use that like creators will ultimately use.

So we have to understand that inside and out and that needs to help guide the research rather than the research happening somewhat in a vacuum and then you know kind of trickling down to to the target user. Yeah.

A lot of the crazy technology vision of the future is kind of just like you're going to with one prompt just be like give make me a new Top Gun movie and it'll just one-shot it. Uh clearly going to be a while till we get there.

uh where are you actually seeing value or demand from Hollywood from filmmakers because you know AI is so broad it could mean just like pull a green screen key better do some rotoscoping do some camera stabilization uh there's been AI tools in film making for a long time they're obviously ramping up set extensions there's so much that you could do in the 3D pipeline and the VFX pipeline and 2D pipeline um where are you seeing actual adoption where you excited for there to be adoption in the next year or the year after?

Yeah, for sure.

I I think you know it's a good point where especially in video I think more than in other kind of fields there has been this like uh you know when when it first started there was this sense of like well you know kind of like the hollow deck right like that's the world that we're we're going to move towards and to an extent it is.

I do think that there's kind of like a misunderstanding though of where the value of the end content comes from. Um, and it's sort of like we're now at the place where you could relatively credibly write a book with an LLM. Like you could have Chad GPT, you know, publish literature.

Problem is nobody's going to read that literature, right? And like that's the that's kind of the missing piece. You know, people have done it. People have. And if you go on the Kindle store, apparently it's like swamped with AI%. Yeah.

And but like and every once in a while these things break out, but it's more of like a novelty like, oh wow, like somebody actually did this thing. Like let me leaf through it. Wow. Yeah. that's, you know, they hit the periods and there's tons of m dashes.

I think it might just end up reflecting human creation where it's ultimately creativity. Creative products are a hits business where there's a lot of AI songs right now, but the only AI artists that I can think of is that the Velvet Sundown or whatever that's that's gotten popular in the last month.

And so it's yeah, I just think it's like this ultra power law potentially. A lot of media properties also and art generally is also very story driven. Like the story behind it. Like part of the reason why I like to go Tom Cruz movie is because I've heard the story that he's doing the stunts. Yeah.

We know who he is and we know the story behind the story and and that's what drives a lot of value in art is like, oh this this painter really spent years doing this thing and he Van got cut off his ear. And so that this painting has a crazy story behind it. So it's valuable even if like my kid could do it.

It's like my LLM can do it, but yeah, your LLM didn't. Yeah. But I I think it's like in AI music, I think you're kind of starting to grapple with that a little bit where it's, you know, you can you can use a lot of these tools. I I think that there's like kind of lowest common denominator content like what Yeah.

You know, what you don't see as much anymore is like blog spam that was just like, you know, crazy in the 2010s, right? Nowadays, that's just replaced with AI. Yeah. Exactly.

Um, I think that with AI music, it's like there's certain strands of like top 40 kind of radio, you know, where it's like it's already a very commoditized, you know, sort of form. That's what it does well. But I I just don't see a world where a transformer model replaces, you know, Kendrick Lamar.

Like that's that's it's so, you know, there's such a big gap there. Um, so so we think about it like internally it's the same way. We we, you know, like John Carmarmac calls it like power tools and and and that's that's largely the model we use.

I think in film as well it's very acute because to your point like unlike other spaces it's actually it's a continuation like there a lot of the things that AI video enables it's not novel like you know you'll hear studios that we talk to they'll say you know a research company came to us and they said hey you can do all these new things with with these video models and they have to remind them that no we can do all of these things right like what what we're talking about here is potentially doing them in a easier more flexible more powerful way But with VFX, like there's nothing you can do today with in in terms of like the output that you're creating that you couldn't do with VFX.

It's it's a workflow thing, right? Like we're just making that process easier, that process more affordable and and that kind of thing. So yeah, we look at it. We're selling we're selling SAS. We're selling SAS. That's what the market needs.

Uh what about um what you know it's been interesting to see uh X is its own kind of universe in terms of AI content what what gets picked up a lot of it ends up being stuff that that's getting made shared on Tik Tok or other platforms and there's been this idea of like an AI content creator which is like a new personality that is just being generated by a human that's creating you know using a a video or an image model to generate uh generate this person, you know, going about their life.

Um, and then there's the sort of like the the what's considered the slop, like the Italian brain rod, all all that stuff.

Um, what about like do you expect to see like an entirely new class of filmmakers in like you know like ba basically like net new YouTube channels things like that of people that are just a you know could be a teenager or just somebody sitting in a room by themselves focused on that storytelling focused on basically creating um you know the internet was beautiful because it lowered the cost of uh distribution to zero and you know mobilevices devices lowered the cost of content creation to zero or effectively zero and now um the the cost of of producing films is not going to go to zero but but may as well on a long enough time horizon and so I'm I'm curious when you think that moment will be where we start to see kind of the the the velvet sundown equivalent of of of film film making.

Yeah, I I would say that like we're a lot closer than people think. Um, so you know, we uh like Moon Valley's models like because we have uh you know, clean models or models that have been trained on licensed data, they've been the first models that studio, you know, legal teams have really like allowed them to use.

And so um that's been part of why, you know, when you heard about Sora like a year and a half ago, since then there hasn't actually been that much AI adoption in the industry. Now it's starting to pick up a little bit.

Um, I would say that like the the the point where the technology is capable of doing that kind of thing, we've already reached and there's places it's there's pockets of the world where especially like smaller film communities in different parts of the world where you're starting to see like real world produ productions.

often people aren't necessarily aware of it but very large parts of not just the pre and post-production but even the production process itself are being driven by these models.

Now I think that there's like beyond especially in this industry beyond the technology there's a lot of other barriers to adoption here right like you have to it's it's a whole new you know it's a whole new set of tools that you have to figure out it's a whole you know the adoption curve is very different and of course you know for better or for worse there's like there's a big kind of dialogue around AI in movies and what that means and you know how we feel about it um but to your point like what we get excited about internally is there's this dialogue of like you've got on one hand just like day-to-day people like myself that maybe oh now I'll be able to create a movie.

Um on the other hand you have the far end of well now studios will be able to make the same movies for much cheaper. And that seems to be where a lot of the focus is to us.

The area that we're the most excited about is that middle where you essentially have this band of millions of creative people in the world like artists, people who have real taste and talent and ability but they don't have necessarily the access and the infrastructure to do that right and that's not necessarily like you know we have a um one of our alpha users is this uh he's a a filmmaker in Sagal and like he's been he's been a filmmaker for over a decade um his sting his thing or you know his his focus is on uh doing like music videos for local artists and and it's like this kind of funky like Afrobe style uh you know West African kind of flavor.

Um, and now he suddenly started making those music videos using generative AI and like the production quality of these have soared and he's had a number like he went from being just like a very obscure, you know, person in his local community to he's had some videos that have gone up like 10 million views on YouTube now and nobody has any idea.

They just think, oh, this is just like a sick artist that like I haven't heard before and you know, cool visuals. So, it's that middle layer. It's like the independent filmmakers who today, unless you're friends with one of the top five studios in the world, you have no capability of making a big budget production.

Now, you'll be able to do that, right? And that's not just, you know, obviously this individual is one example, but for instance, we have like somebody like Natasha Leon who we work really closely with. She's, you know, an industry insider.

She's a, you know, one of the the kind of top people in the industry today, but she's working on this new movie that's like she's been leveraging AI to help do it because this has been something that she's wanted this is a movie she's wanted to make for well over a decade, but she just couldn't.

She w like she talked to the major studios and it's like, hey, it's going to cost us $30 million to do this, right? And and we just can't we can't do that. Now it's like, well, if I can suddenly potentially do it for $15 million, now I can, you know, I can make this thing happen, right?

And and so it's there's there's there's this like idea that you'll be able to do, you know, movies for cheaper, but really what we're seeing in the studio is a existing movies you're now just doing more than you want expected to do before. You're now having to compromise less.

Like directors aren't saying, "Hey, I had a $75 million budget. Now I'm going to do it for 50 million. " They're saying, "Hey, with my $75 million and my team, I'm going to go and now do all the things that I couldn't do before unless I had a hund00 million budget. " So, that's like one thing we're seeing.

And then the flip side of that is you walk into any studio, for every one good production that's live, there's 10 that weren't green lit because they just couldn't get the budget. Right now, suddenly, you'll have a lot more of that.

So that's like high level how you know I think that the way that this stuff is actually getting implemented it's happening in a different way than I think where there's been a lot of fear um and and you know I think it's totally justifiable fear but I do think that ultimately it's the artistry that wins out more than like you know budget requirements or budget constraints.

Yeah. Yeah. And the I mean the exciting thing from my point of view is if movie studios keep budgets relatively the same because there is this incredible demand for content, but then you can take a $und00 million budget and it can now go to 10 different films. You get 10 more shots on goal.

It's 10 more teams and maybe the underlying teams even can create, you know, better margins themselves. So very exciting. Yeah. Yeah. This is awesome. Thank you so much for stopping by. Hope you have a great rest of your day. Yeah. Come back on when you have this. We'll talk to you soon. Awesome.

Thanks for having me, guys. talk later. Um, and that is our show for today. Jordy, do you have any other breaking news you want to share? Breaking news, Brandon Jacobe trade deal. Well, I guess now's a good time as any for some personal news. After Wild Chapter X, I've officially wrapped up my time there.

I'm extremely proud of the work we did and you'll see more of it soon. More to come on what's next dot dot dot. Uh, Brandon has been a dear friend for a long time and uh I'm excited for his next chapter. Me, too. And uh I'm gonna miss being able to text him when I have bugs. Uh but uh Tyler, you're up next.

Expect some bug reports. But um yeah, fun show. I'm trying to think if there's anything else that we that we missed. I don't think so. We hit all of our ads. Um uh Chimath was saying that Meta switches to Sonnet for coding um instead of using Llama uh that they had uh fine-tuned on Meta's own codebase.

Now they're just oneshotting everything with Sonnet. So very interesting. Since the change code suggestions are generally better, engineers can change back to Llama and occasionally do when the fine-tuning makes a difference.

Internally, this is a big change given how big how heavily Meta has invested in the Llama project or product. Uh, this move officially acknowledges that anthropics models are currently far ahead of Meta's own LLMs, even when fine-tuned.

I suspect Meta will double down and try and make their next llama versions more capable for coding, but until then, it uh it doesn't want to hold back its engineers. So, very interesting that they're uh they're using anthropic and just kind of letting it repro. Uh, last post to end it uh from Jarvis Best.

He says he's sharing a screenshot of Linda Yakarinho saying, "After two incredible years, I've decided to step down as as CEO of X obviously has a much longer post, which we covered earlier. " And Elon just comments, "Thank you for your contributions. " Jarvis says, "Lmao, cold as dry ice. " Um, anyways, uh, cordial.

Yeah, at least they were cordial. Uh, leave us five stars on Apple Podcast and Spotify.