Val Kilmer's AI Return and What It Means for Film

Val Kilmer has been resurrected via AI to star in the new movie, As Deep as the Grave. This is weird because it's like, I would understand if they had shot, he got sick, they needed to finish it. You know, like Paul Walker, you know, in the Fast movies. But this is from scratch.

Just because that actor looks like Val Kilmer by face, is it really him in performance? Right. Like, how is that... Are they going through every Val Kilmer movie and making, like, a Val Kilmer model?

Probably not. [upbeat music] All right, welcome back to Denoised. It's been a minute. Addy, how you doing?

It's, it's been a minute. Let's go. It was on me. It was on me.

It was on me. I was busy with some other videos and projects, which one of is, is out now, the, uh, One Battle After Another with Beeble and SwitchX, which came in clutch at the end because I was trying to experiment and use Beeble and try to recreate the One Battle After Another car chase with an iPhone [laughs] and initially Beeble and Nuke, and then it hit a limit, but then SwitchX came in clutch, their new product. Uh, I think it's wild. You came close.

I was impressed by the end results. Thank you. Yeah. Yeah.

I'm [laughs] And, uh, you didn't even get an Oscar nomination or anything. I was hoping it would've won the Oscar too, but I mean, I can understand Sutters. You know, I, I mean, obviously Sutters is well deserving of it. A- and, and, and Autumn...

I can't remember her last name. But, um, I still think that, uh, Kugler's IMAX explainer video [laughs] Me too... did so much to lift the- Yeah... everyone's awareness of IMAX and, uh, and, and, and film for that film.

[laughs] He, he should be writing the framing decision list. Did you see that? That just came out from the ASC, um, like the official FDL. Yeah.

It's like Ryan Kugler- What is it?... can write that in his sleep. He just knows all about aspect ratios and formats. I know.

Yeah. [laughs] It's, it's so, it's so crazy. Not to get too much on a tangent, but, like, the Dune 3 trailer came out yes- two days ago or yesterday. Mm-hmm.

And... Oh, let me find... Now I gotta find it. Okay.

So basically- Oh, come on. I don't wanna watch it. I'm saving it for the weekend. Oh, you're doing...

Well- I'm gonna put it on the big TV... the trailer came out... Well, firstly, A, why haven't you watched it yet? And then B, [laughs] the trailer, like, came out on YouTube, you know, regular 16 by 9, and then they also released a bunch of clips on, you know, TikTok and vertical, and, but because they shot on IMAX, someone took the horizontal clips and the vertical clips and stitched them together so that they were [laughs] basically able to recreate what the unaltered 143 IMAX frame should be [laughs] for the trailer.

That's awesome. Th- that... See, that whole aspect ratio game has really leveled up, like- Yeah... for this.

Which also at this point- Yeah... why is IMAX, or why is no one releasing a full uncropped version of the trailer on YouTube? Like, YouTube could handle it. I don't know why IMAX or- Uh, because our de- our devices are still 16.

9 or wider, right? Like, we don't have CRT monitors anymore. Filmmakers- Those used to be 14... on a laptop do not care about the bar, the pillar boxing that they will experience.

They want the full image. We all want the full image. They want the full image. Yeah.

Yeah. We don't care- That's true. Okay... if it's gonna be pillar boxed.

All right. I know [laughs] Maybe it's a, a strategy on IMAX's part, right? Like, the only way you can see the square aspect is you go to the IMAX theaters. See, I mean, anyone that cares- Yeah...

enough about it is gonna see it in IMAX anyways. Like, this is the trailer. We want the full... We want uncropped images.

[laughs] Okay. [laughs] Viewers- [laughs]... we're sorry about this tangent. Let's get back into it.

Well, I mean, if you care enough about this too, let us know. [laughs] I don't think we're getting comments on this. All right. Kinda got a couple stories here around AI actors and acting- Mm-hmm...

and stuff. The first one, this hit the other day. Val Kilmer has been resurrected via AI to star in the new movie, As Deep as the Grave. So Val Kilmer passed away last year, I believe.

Or two years ago. Yeah, right after Top Gun Maverick came out. Uh, so according to Variety, Kilmer was cast in the movie in 2020, five years before his death, uh, but he was too sick amid his throat cancer battle to ever make it to set. Mm-hmm.

Now an AI version of the actor is appearing in the film with the full blessing of his daughter. Nice. So I don't think they sh- Doesn't sound like they shot anything with him yet. It sounds like he was casted, too sick to shoot, and now they're using an AI version of him- Yeah...

in the film. That's very interesting. And I'm pretty sure this won't be the last, right? Like, this is a b- a new avenue for estates and actor longevity after they pass, or even if they're too sick to work.

I mean, first off, like, we haven't seen any per- any of the quote performances, so, like, we don't know how this is gonna play off and how they're gonna film it. You know, it's like, are they gonna film it with another actor and use, um, you know, like, motion transfer to overlay his likeness on another performance? Is it gonna be completely, uh- Synthetic... text or, yeah, generated.

But yeah, I mean, I, I saw the, you know, the con side of this is, like, they didn't film anything. Val Kilmer, you know, brought his performance and acting to the performance. You're losing that. Mm-hmm.

Mm-hmm. Which I understand. Like, I don't think... I think you lose a lot when it's not a human performer, even if it's, like, a motion capture kinda setup with, with clean motion transfer.

This is weird because it's like, I would understand if they had shot, he got sick, they needed to finish it. You know, like Paul Walker, uh, you know, in the Fast movies. But this is from scratch. Yeah.

I, I do see your point about the physical performance and all the minutiae of facial performance that Val Kilmer will provide that is only him, right? Like, a sub-in actor is gonna not be able to replicate that 100%. So just because that actor looks like Val Kimmer, Val Kilmer by face, is it really him in performance? Right.

Where, like, how is that pr- Yeah, is it... Are they going through every Val Kilmer movie and making, like, a Val Kilmer model? Probably not. You know, it's probably not like this.

I don't think so. And even then, like, you're still... Uh, I mean, the script and the performance would be so unique to this movie in itself. Like, like, you can't take Val Kilmer from Heat- Mm-hmm...

right? Build a model off the motion and the performance and the look, and then expect that to sort of translate- Right, and then Jack Kilmer from Top Gun... to this. Like, completely different- Yeah...

characters. [laughs] Exactly. His performances. Yeah.

I'm thinking now, it's like the other... From a filmmaker, producer perspective. So if you had Val Kilmer on board for this film, I don't know if this is an independent film, if this was, like, low budget or, you know, if this was, like, a passion project that they got Kilmer on board.Having a, you know, big name actor signed onto your project hugely helps with the marketability and being able to sell that film.

Yeah. So I also don't know- Right... if it's that kind of story where it's like indie film that's getting produced, I'm speculating, uh, that they got him on board, you know, he was too sick to do it, but his estate, his family still, you know, still wanted to do it, believed in the project. And- Mm-hmm...

recasting and losing, you know, Val Kilmer to have in the marquee, it- Mm... greatly hurts, like, the chances that you have to sell that film. Absolutely. Yeah.

For an indie film, that could be make or break, right? Yeah. 'Cause you need, you need to name, like name sell films for in the indie market. At least one.

Yeah. At least one to anchor it around. Yeah. How many times have you seen like ton of smaller indie films that has a big name, and then you watch it, and then they're like in two scenes, but it's like, "Hey, but we got him for like [laughs] two scenes.

It helps, it helped market the film." [laughs] I mean, I, I do get it from audience perspective because we don't have anything to anchor ourselves to the indie film, right? So when we think, oh, Val Kilmer, I associate him with like three really amazing movies. This has gotta be worth something.

He wouldn't just do any project. Yeah. So then you naturally gravitate towards that project. Yeah.

This could be like a very elaborate licensing play, right? I think that's what you're referring to. Like just having Val Kilmer's name on the marquee and, uh, his face on the poster could be enough, and maybe performance is a secondary sort of giveaway. I mean, and we also don't really know how big the role was, you know.

Is it like... I'm assuming minor supporting role, not like lead actor role in the film as a guess. Mm-hmm. Um, because that would also be a lot of heavy lifting to have a fully synthetic or however they're gonna do it performance in the film.

I don't know, man. Uh, I'm seeing a lot of AI videos online. It doesn't seem like a heavy lift to me. A heavy lift in the sense of something that is cohesive and watchable for 90 minutes [laughs] like and- Oh, I mean, yeah.

Yeah. If he was a primary character- And then you don't-... that would be, that would be hard. Yeah.

And you don't have weird, uncanny valley for like, you know, 10 minutes in, and then the re- other 80 minutes you're just like, "Ugh, what?" [laughs] It'd be so distracting from the rest of the movie. You're just focused- Yeah... on like uncanny valley stuff the entire time.

Yeah. Yeah. Exactly. Okay.

So that was one story, you know, kind of just hit. So obviously a lot we still don't know yet, but a lot of opinions around that. I'm guessing a lot of bad opinions. Yeah.

Yeah, of course. [laughs] Yeah. Look, I mean, also you have the blessing of the, the estate and stuff. It's also not like, "Hey, we're gonna take Val Kilmer and like run with this."

Yes. And, you know, even 'cause he signed off, even though if his family's like against it, you know, the family, the family is for it. Well, I don't see it any different than Ian Holmes and Alien Romulus, right? Yeah, I was thinking that too.

Yeah. Yeah. It's like instead of AI, they used, uh- Right. And where-...

body scans of Ian- Mm-hmm... and traditional VFX and CG. Yeah. Where is the line when stuff like this does happen when it was just using more manual methods but still the end result is the same?

Now, I mean, the argument for Alien was like, that was based on existing character. We know what that character is and kinda who he should look, what he should look like and how he should behave and act and perform. Yeah. You know, whereas this is something completely new that has no- That's true.

Yeah... reference. For Alien, that's more story and lore continuity. Yeah.

This is, uh, it feels like more for, uh, marketing and sort of, uh, initial- Right... uh, vibes- Besides-... around a movie... we had him signed on before.

Like, we don't wanna lose- Yeah... that to help, you know, our film's, uh, success chances. Okay. Other story in the AI acting world, an AI company, uh, well, let's see, Verge's quote is, "AI company wants to harvest improv actors' skills to train AI on human emotion."

[laughs] Did you say harvest? Why, why would they use that word? The Verge's headline says harvest. Yeah.

Oh my God. [laughs] Because they're just going for the, uh, crazy hype. But, um, so it's a company called Handshake, and I believe Handshake is one of the companies, if you hear those stories of like, oh, DoorDash or one of those companies that's filming you do like menial tasks to train a model. I think this is one of those companies that sources people filming themselves doing things to provide data- Okay...

for the a- for the, um, AI companies to train on. So it's like a video data, uh, Uber. I don't know. Something like that.

[laughs] Yeah, I think so. Yeah. Yeah. [laughs] Basically, like, I...

One of the ex- I don't know if it was Handshake specifically, but one example was like the company's paying people to like film themselves folding laundry. Like, and getting thousands of hours of people folding laundry so they could chain, train- Dude, $74 an hour-... the robot... sign me up.

Yeah. So this is a new job posting for improv actors, $74 an hour to basically, uh... Handshake AI is inviting actors, improvisers, and performers to join a paid collaborative improv project, and it's basically Zoom calls where they are filming themselves doing improv. Improvise scenes and dialogue, collaborate with other performers, experiment with character emotion and subtext.

And then this is all getting training data to then train other models or provide this raw data to whatever company wants to train models to have better synthetic performers. Okay. You wanna hear my take on it? Yeah, let's go.

This is gonna fail atrociously. [laughs] Just, it's just gonna hit the wall so hard. Here's why. When you take, let's say there is a script of something happy, and you take the input of 1,000 improv actors expressing the happiness, what an AI model's gonna do is more or less, uh, funnel it down into latent space of that, and all of the minute expressions of each individual actors will basically get muddled down into one giant expression, and that giant expression of happiness is just gonna be so over the top and cheesy that, uh, it'll be more distracting than what you originally intended.

We talked about Val Kilmer, like the reason Val Kilmer is who he is is because of the way he expresses happiness. It's different from the way Tom Cruise expresses happiness, and we identify those emotions with those actors, and there is a very strong vein of individuality to that. And when you are trying to like mash all that individual, individuality into training models, you're not gonna get it back. I, I mostly agree with you.

I think... I, yeah, I still don't, yeah, I don't think this would hit the level of like what a real performer brings in their interpretation to a script, a character, the subtle nuances. I think-Sure, this would help in just general text-to-video outputs and getting some better nuances out of, like, a single shot from a synthetic character. You know, I mean, I also realize as you're talking through that we're thinking, you know, film use cases 'cause that's our world.

But I was thinking when we're talking about... when you were talking about it, like, it could, it, you know, we're thinking- Like a TikTok video. TikTok for one, but even, like, gaming or, like, uh, NPCs. Mm.

If you're, like, playing a game and you're having this, you under the hood AI, but you're having, if you're having interaction with a character that is powered by AI, but then having that character also be able to have better physical reactions that's a, you know, synthetic character in gaming, that, that, that could also be an application that makes sense or- I think so... be level up. I think, I think you're right. I think there, there is room- So-...

for, uh, lesser quality outputs- Mm... and more of it, and more sort of qua-quantity over quality use cases, if you will. The, I do have to just say this because this is just Silicon Valley engineer thinking of trying to solve a problem with- We just need more data [laughs]... the tools that they know.

We just need more data. [laughs] They're like- Yes... "Oh, happiness? Oh, yeah, we can, we can fix that."

Right. "Let's just, let's just get data of people being happy, and then we got it." Yeah. People, no.

We're human beings. [laughs] Yeah. We're highly, like, tuned to detect other people's emotions. Yeah, and bring in our creative interpretation to how you read a line or deliver a line a, a million different ways.

I bet these are the same, uh, this is the same group that sort of, I mean, this is going into what, uh, Cristobal, Runway CEO, said, which I totally disagree with, was, "Creativity is just, um, a controlled hallucination." I think more or less that was his, uh, quote. Mm. And I, I couldn't disagree farther from that because creativity is not just the brain on drugs, if you will, like imagining stuff.

It's a, it's a controlled craft and you build your creative skill set over decades. Yeah, like a mus- like a muscle. Yeah, exactly. So like, okay, I have this minutia of an idea, but then it turns into this, and I've built that muscle in my brain- Mm-hmm...

to turn that little thing into this thing, and then now I can develop on that with character arches and environment builds and all of that stuff. So like that's, that is not controlled hallucination. That's a craft, you know? Um- Yeah...

and, and along the same lines, I think this is, the Handshake guys are, are probably thinking like, "Oh, happiness, ha- those are just emotions. Like, surely there's, if, if enough data is available, we can solve that." Right, we just need more data, or we need data of like, yeah, someone, uh, you know, I mean, I'm sure the improv data helps. Like, uh, here's a scenario.

What's the person's reaction? How do we like, you know, cause and effect to have these pairings and train that? And I'm sure this will help for some stuff, but yeah, I don't think this would ever hit the level or be something that we would wanna replace, like actual performers in a narrative fictional piece, you know? Right.

Maybe on something where like the scale of having a performer, like a game, is unrealistic. Maybe it helps with that. Yeah, like GTA VI, you've got 3,000- Yeah... NPCs.

Yeah. You know, you gotta just fill them all with the appropriate emotions. And we know the current jokes, you know, where the NPCs right now- Yeah... are just like, "Ugh.

How you doing?" And people, you know, reenact them and make fun of them. Yes. So like, yeah, if that interaction can feel more realistic, you know, 'cause that would just physically be impossible to have some 3,000 [laughs] people on a motion sim- [laughs] motion tracking simulator interacting with, [laughs] with millions of users of all, in real time, maybe it'll work there.

[laughs] Yeah. All right, what else we got in the actor world? Okay, this one's just in the weird why does this exist. Peacock is launching an AI-generated version of, of Andy Cohen.

[laughs] Remind me who Andy Cohen is. His face looks familiar. He is the face of Bravo. He's, uh, Watch What Happens Live and- Okay...

also co-hosts my favorite New Year's Eve countdown show with Anderson Cooper. [laughs] Okay, so he is AI-generated for a bunch of Bravo shows. I don't know why this exists. You're, the, the Bravo-verse, I mean, I, you know, obviously he's like the, you know, their big bread and butter is the Real Housewives universe.

Oh, okay, okay. And then he has a talk show, but a lot of the stuff on Bravo, you know, is I think still Real Housewives is like the main thing. Yeah, I just don't, I don't know why. Is there a need for people who want to?

Is there not enough Andy Cohen in the world? Like, is he so busy he can't shoot all these things at once? [laughs] I don't, I don't, is it, I don't, I, I don't know. [laughs] But I mean, somehow I feel like we're gonna see a lot more of this.

I don't know if it's like an A-Andy Cohen chatbot and it's like, "Hey, I wanna get Andy Cohen's take." Obviously, we know that exists in like Meta and all their stuff with, you know, where you could ask what... I forgot who they launched it with, but like where you can, you could chat with historical figures in chat. They had that feature for a bit.

I think you could chat with like Abraham Lincoln for a bit. [laughs] Who would, who would want to do that? If you're, um, if you're a- If you're a Civil War reenactor, then yes... No, if you're what's-his-name who built his own AI George Washington.

Glenn Beck. If you're Glenn Beck and... Did you see that? [laughs] Oh.

[laughs] What? [laughs] Yeah, I'm on, I'm on the internet too much. Uh, Glenn Beck built his own AI George Washington and then had a podcast interview with him. [laughs] That's interesting.

I, okay, that's interesting. I mean, okay, but so is that, is, do you think Andy Cohen thing is along the same lines, where it's interactive and you can- I think it's a chatbot. I think it's just a chatbot. Okay, okay.

All right. You know, where it's just like, "Hey, I wanna a- you know, get, uh, Andy Cohen's... as if I'm chatting to him or get his like take on, on something." Are you looking up the Glenn Beck thing?

Yeah. Got, I, I gotta see some footage, yeah. Can you put the volume up on that? Oh my God.

[laughs] It's George Washington in a T-shirt. [laughs] Yeah, it's George Washington- Oh, he's, he's so naughty... in the manosphere. [laughs] He needs the Handshake guys' emotion plugin.

[laughs] Oh my God. Yeah. So, um, yeah, that happened. Okay, okay.

[laughs] All right. Look, I mean, uh, the bright side and the silver lining to all of this is the media companies are experimenting. I mean, without experimenting, you're not gonna hit on gold, right? So let them experiment.

Yeah, and look, I mean, we're making fun of this, but I'm, I'm sure there is a... I mean, Bravo is huge and Real Housewives is huge. I'm sure there's like, there is a market of people who would be like, "Yeah, I would like to have a-Chat with Andy Cohen. I mean, people, people have AI girlfriends, so that, that's not [laughs] It's not out of the realm of possibility.

Yeah. Yeah. And then it... there's, there's also probably that clientele who likes vertical dramas.

It's probably a big Venn diagram- Yeah... of those kinds of people. Yeah. And, uh, yeah, I, I think they would welcome a chatbot of some Real Housewives characters and Andy Cohen and all that.

I was gonna skip this because I thought like, okay, maybe it wasn't relevant here, but because we just talked about AI George Washington, this is another... This is probably a rising issue of weird stuff of like using AI to like feed it data based on a person and then, quote, interviewing them, and then acting like you really had a conversation with them or that it is like factual or based in like truth. This is so ridiculous. Is our show becoming a comedy show?

Because this is all [laughs] This is crazy stuff, man. [laughs] We're just living in a, a, a weird universe. But yeah. Okay.

So basically, and this, this was a- another story that Vanity Fair wanted to write a feature story about Dario, ah, CEO of... whose last name I can't- Antropik... Ah, Amodia, Amodia. I think, I've never...

It's one of those things where I never say it out loud, so I don't know how to pronounce his name. Dario, CEO of Antropik. He talks like this. He's always such a New Yorker.

[laughs] Somehow to be a, a C- a CEO of like a big AI company, you have to like [laughs] have to be like the least charismatic person [laughs] between him and Sam Altman. Oh my God. [laughs] They're, they're stand-ins for Silicon Valley. [laughs] Anyways, back to the, the main thing of the story.

Vanity Fair wanted to do a feature about him. He declined to do the interview, so then they just fed Claude a bunch of Dario interviews or stuff or info- And his book... and then also sort of assumed because he created Claude, Claude should know him the best, and then interviewed Claude, Claude Dario, and then wrote an article about this. Great.

If you wanna write a feature story about a dude and the dude turns you down- Yeah... then you either kill the story or you like source, you know, from other materials. It's, it's a provocative piece. It's what Vanity Fair does.

I mean, it's, it's just using the- Well, ironically, I've just been reading clips about it online. I haven't actually read the article. Yeah. 'Cause like why would I want to read an article about like your conversation with a chatbot, uh, making up, impersonating a dude?

I mean, if, if Vanity Fair interview does hit on the, the three or four main things that Dario screams from the rooftop about, which is AI is unsafe, Cha- [laughs] ChatGPT is the unsafest one, and use Claude instead of ChatGPT, like in that order. If the interview kind of covers that, then yeah, I, I guess it, it's done a good enough job. Yeah, but I mean, you could source that info from like doing actual journalism and research. Like Dario's blog posts are like mini novels in themselves.

He wrote like a... What was his last one about rising AI? It was like a 90-minute read. It's not like he's a secret.

He writes a bunch of stuff. I don't know why you need a chatbot to do that. The bigger picture with this is just like slippery slope, especially in the nonfiction world- Mm-hmm... when you're feeding the chatbots historical data and then getting new outputs, and then treating that as fact.

Yeah. You should never treat AI as ground truth or- Yeah... the hard tr- yeah, that's just not... no bueno.

Like- Like maybe it can synthesize ideas, bring up things to investigate, but like to kinda... You know, also with the AI George Washington thing, it's just like, okay, maybe like you didn't really, you didn't really interview George Washington. Like- You know Clay like was like, "Make George Washington more neocon." [laughs] I...

There's another tweet where it's like, "Wow, he made an AI George Washington that somehow miraculously agrees with everything he says." [laughs] Exactly, yeah. You could totally do that. It's like, "See?

George Washington said it." And like- Yeah, you built him from scratch. [laughs] Yeah. I know.

He's trained on historical data. He agrees with me. That was our roundup of AI actor persona stuff. So [laughs] what else is new?

Hope you like it, viewers. I mean, look, if, if this is all garbage, just skip forward. It's all good. Yeah, let us know what you think.

All right. Next up, some actual product stuff. Google on a roll again. Little less in the video world, but the big one they launched this week is Google Stitch, a vibe design platform that basically can transform natural language or crappy sketches into high fidelity, high fidelity designs.

Again, terrible naming from Google once again. Stitch, I just, I just get the Stitch c- character from Disney. Yeah, from Lilo & Stitch. Like there was any other name you could have picked from, but okay, go on.

I believe Figma took a little tumble after this news. Oh. [sighs] But from the look at it, I mean, basically it is iterative hash it out, vibe code design, but you would get the actual designs. Not like an image mockup of a design- Right...

but like app mockups, uh, webpage designs, but like built in actual code and stuff that you could use elsewhere and edit and build off, build upon. I, I think Figma was working on a version of this, but it sounds like Google beat them to it. I'm sure they are. Yeah.

Yeah. I mean, you know, and it's still Google i-i... in their AI lab in the Google world. I've always added my like caveat.

I'm always like sus- not suspicious, but like hesitant to go all in on a Google tool just because I've always been so burned from Google killing experimental products- Right... that I like am hesitant to be like, "Yeah, this is gonna be our br- our, our main new thing." But, um- You mentioned that before. Yeah.

You, you seem really hurt by that. I, I... [laughs] That, decades, man, of- Damn... being burned over, [laughs] over, over products, over things I used a lot that got killed.

Well, you're gonna have to turn, turn off your Oculus, uh, goggles soon too. I know. Yeah. I know.

I said, yeah, the other meta [laughs] that ironically renamed the company after the Metaverse. It just goes to show nothing is permanent. It's shutting down the Metaverse. Yeah, nothing's permanent.

I mean, one day your Gmail might be obsolete, so start- No, don't, don't, don't, don't do that. Don't jinx it. [laughs] Gonna go back to my four megabyte Yahoo account. [laughs] Hotmail, dude.

[laughs] Hotmail had two megabytes. Yahoo had four megabytes. Yahoo was where it was at, and then Google was like, "We'll give you two gigabytes." [laughs] And it was like, "All right.

Bye." [laughs] Yeah, the, uh, the struggle was real. And then the other thing they launched was a sort of full stack vibe coding experience. So this is competing with Cloud Code and Codex and all that stuff.

But making it easy, I think a little bit more user-friendly, 'cause, like, Cloud Code and Codex, you still kind of have to either run it in terminal or know a little bit about what you're doing. This one, I believe it's all web-based. It will automatically call up a database and, like, kinda some of the back-end stuff that you need to run these things- Right... if you are trying to vibe code an app.

It has it in a single platform- Wow... for you. So you can have this web experience to kind of vibe code an app. You have this new design tool that they built to help design it.

So you have the front end, which is the design tool, and then you also have back-end infrastructure capability, like you mentioned. Yeah. Configuration of database connections and authentication flow. So it has, uh, like, Google Sign-in built in, so you can kind of add it quickly, database stuff, so you can add it quickly.

Stuff that you still have to kind of manually do a couple steps if you're using the other tools, and kinda know what you're doing, built into, like, one single platform. So it's - definitely seems like it's a tool targeted at people who wanna vibe code but don't have coding experience and want a little bit simpler interface while still not giving up on the powerful features that you could vibe code, potentially. Man, I, I don't know what this does to, like, full stack developers who are, like, the most sought-after type of developers because they sort of know the, the architecture from the very bottom to the, all the way to the top.

There, there are so many people that make a living being freelance and contractor and project-based, uh, full stack developers, right? Wonder what it does to them [laughs] with this stuff coming out. I mean, I think, you know, I think it is a dangerous slope, and I think we talked about this before of, like, what happens when you're a entry-level or junior developer coming into the job market, and if those roles get eliminated, or a lot of them get eliminated, but you still...

Like, these things are cool, but it's like you still need a, quote, like, "an adult in the room" [laughs] that- Yeah... knows how this all works and comes together. 'Cause I, I can vibe code all this stuff, but, like, I don't know what's happening underneath it. Like, I wouldn't trust that, like, it's not gonna blow up or, like, you know, something wrong is g- something's gonna derail.

Like, you still need people with this expertise- Yeah... knowledge. But if we kind of replace the entry-level roles that would, like, feed up to being a senior developer, where is that gonna put us in ten, fifteen years? Yeah, how do you become a senior developer if you can't get your foot in in the first place?

Right, getting hands messy and doing this kind of, like- Right... lower level coding work that you would've done that's gonna, being replaced by AI. Yeah. I, I don't know if it was Dario, but one of the big CE-CEOs just said that, uh, the class of twenty twenty-six coming out of college this year are gonna basically more or less experience, uh, one-third unemployment rate, so thirty- Yeah...

p-percent or so. Yeah. Yeah. I don't remember who said that, but I did see that quote.

And yeah, that's also cr-grim and crazy. So crazy. So crazy, because when I graduated from college, it was the... Like, there was more, more jobs than college grads.

I graduated in- Yeah... like, '08, so it was also a pretty shitty time. [laughs] Yeah, around the same time. Yeah.

'08 was a pretty crappy time all around. Everyone was... [laughs] Oh, my, so I, I graduated, like, couple years be-before- Before the crash. Yeah.

Okay, yeah. Yeah. Yeah. [laughs] When everything was still optimistic.

[laughs] Damn, I didn't know you graduated in '08. No wonder you're in film. Yeah. [laughs] You're like, "Forget the job."

Yeah, when I graduated- Economy... like, jobs. Like, [laughs] nothing seems secure now. Like, might as well just freelance.

Yeah. Equally as secure as- Might as well make content... trying to get a job. Yeah.

I don't know if you saw this quote from Sam Altman, this tweet [laughs] that was, "I have so much gratitude to people who wrote extremely complex software character by character. It already feels difficult to remember how much effort it really took. [laughs] Thank you for getting us to this point." [laughs] What an asshole.

[laughs] First of all, he's not a developer. He has no idea what goes on in there. Yeah, he's like, "Thanks for the code. We'll, [laughs] we'll take it from here.

[laughs] Thanks for all the hard work." Like- "We're gonna automate this all"... these guys wonder why nobody likes them. You know, it's stuff like this.

Ugh. Like, keep that to yourself, man. [laughs] I mean, not as bad as, uh, humans take up a lot of energy too. [laughs] Yeah, it's like, oh, uh, yeah, humans need food and shelter, and they need about twenty years to become intelligent.

Why can't we just have AI do all that? Yeah, humans have a soul. AI has a soul.md file.

It's the same thing. It's the same. [laughs] Ugh. You can't...

Dude, nonfiction is way stranger than fiction, right? Like, what's the same? I think the moral of all of, everything we're talking about is, um, Silicon Valley needs a reboot. That's pretty much the bottom line here.

I have a friend who was on the first season, and, um, he's ready. He's ready to be recast. [laughs] All right, now kinda we'll go power through this. Quick roundup of some new stuff.

This Apple Lito. This is yours, Addy. Tell me about it. Yeah, so Apple, not known for big LLMs and image models and video models, right?

They're letting everybody else play that game. What they're doing I think is really smart. They also need to do more, but what they're doing is they're making models that nobody else is making. Very niche- Mm-hmm...

very sort of bespoke and custom stuff. So Lito, or Lito, is essentially a 3D latent space where you're generating a 3D object out of a prompt, right? Mm-hmm. Let's say, you know-A camera.

The ob- the issue with most three D mesh generation is the texture is just atrocious because, um, the texture doesn't really follow the three D object accurately. It has no awareness of lighting. Mm. Um, so when you, when you see the mesh, you're like, "Ah, th-this does look like a camera.

I think I could use it." And then you put the texture on it, 'cause without the texture, it's kinda useless. You're not gonna retexture it yourself. And you're like, "Oh, now it looks low poly and crappy."

So Apple has sort of resolved that by coming out of the latent space with a light, accurate, um, two D rendering as well as a three D mesh itself. So it's combining both, uh, the look and feel of nice two D generations with the versatility of three D mesh. On the, on the left, you have a much more accurate lighting and texturing of that apple and versus, yeah, on the right. This is what you generally get with, uh, automated AI mesh generation, is it'll slap on a texture that's just totally not what, what the object really looks like, and that's because it's using two different processes.

It's using a three D mesh generation, which is completely decoupled from the texture- Mm-hmm, mm-hmm... and the i-image generation. Here, this is all combined into one pipeline. That is interesting.

Yeah, and it seems like everything that Apple does come out with, machine learning or paper-wise, is rooted in the Vision Pro- Yes... application space. Like they had that other paper we talked about like a month or two ago of, um, what did you... I think it helped generate- It was a VLM, right?...

worlds. It was like, um, Apple does have a model, the, that local lightweight model that can look at the real world and identifi- it's like the ter- T2 Terminator, uh, uh- Oh, is that... Didn't they do something else that was, um, like two D image to help expand to three D images, like three D space or some-something like that? Sure.

I could see that. Uh, I think basically everything they've been coming out with is like in the - seems like, "Oh, that could apply to Vision Pro in that use case." Yeah. And, and like they're, they're making stuff nobody else is making, which is really interesting.

Yeah. 'Cause they could really, really... I mean, them being Apple, th-they have so many brilliant minds at work. They can make anything they want, and they're choosing to go after these very specific lanes.

And I know there is an internal strategy on why they're making these things. I just, I just wonder what that is, right? Like, what's the end game mean? Well, I mean, it's a Vision Pro now, but then also obviously the, you know, the, the holy grail is like a realistic augmented reality.

And so something that you could have, you know, project things into the real world. So like having objects that can match the lighting of like our actual environment, if it's being projected on our glasses and we have - like, I see this apple on the table here, but then it looks real. Mm. Or like lighting seems like it fits my environment.

That, that seems like the play down the line. Yeah. I wonder if they're feeling a little bit- I think, uh-... vindicated with- Yeah...

Meta shutting the metaverse down. 'Cause they're like, we see AR, VR, but we don't s- they were never about like immersive worlds or metaverse type stuff. They were just like immersive headsets for work stuff, but not like a collaborative place. I wonder if they're- Yeah, it's like a, it's a functional approach to AR, VR, where metaverse, uh, Meta's metaverse was more of an entertainment approach.

I, I, I do wonder like what the future of the metaverse is in its entirety. If you take both Fortnite and Roblox into account, I know both of those platforms, although they have millions of daily active users, they're struggling with seeing growth year over year. And, uh, one of the, the big problems with both of them, because they have young audience, is that as they age out, as, as a gamer goes from like twelve years old to fifteen to eighteen, they'll start to kind of at-attrition away from the platform.

And so there's a natural sort of falloff with- Yeah... the growth curve because you can't input enough young users, uh, at a faster enough rate than the older users leaving. So I think in about, I don't know, ten, fifteen years, we'll, we'll see what the metaverse is, and if it's even, even around. Yeah, and what that looks like.

I mean, there are just a lot of technical limitations to have that, that low latency real-time experience, uh, that- Right... was outlined- Yeah, everything Apple... Yeah... in that book behind you somewhere, [laughs] that I know is on the shelf behind you.

Yes. Uh, yeah, I, it's actually behind you when you sit here. When I sit there, yeah. Yeah, it's right there.

The Metaverse Book, which- Yeah... it was a good book, but basically the entire book was just like, "Here are all of the technical [laughs] limitations of like how something like The Matrix like is physically not possible and might never be possible." Yeah. Shout out to Matthew Ball.

Yeah. That's his book. Yeah. Okay, other quick update, announcement.

This one's a teaser, so we don't really know much about it, but Runway previewed a developer preview, uh, with NVIDIA that was a mostly realistic sub-one-hundred-millisecond video generation model. Dang, that's good. That's... is fast.

That's really good. It's fast. That still doesn't look like what we're seeing from Krea. So their demo is like type a text prompt, and as you're typing, the video is generating, and then, you know- Right...

you could redo the text prompt, and the video changes, which is cool. It's near real time. It's not real time. Yeah, not like Krea- Yeah...

where you're like moving objects, and your video is updating as you're moving the objects, or you can attach it to a webcam, and your video- Yeah... is updating with the camera. So like, you know, I still think text inputs is like the weakest input. Yeah.

If you compare like Krea to like Unreal, where it's like instantaneous and real- Yeah... this is more like a, a really high GPU execution of Arnold, like a traditional renderer that's mimicking a fast real-time response, but under the hood it's probably just the same thing. It just has way more GPU to- Yeah, it's just the same thing-... kind of crunch it...

just running faster. [laughs] I'm sure that it's light, they lighten it down, but, uh, underneath it, it, I don't think anything is really real-time ready, and they're throwing the latest and greatest, um, NVIDIA GPU at it, which would naturally give it a, you know, a boost of speed and throughput. Yeah. All right, other quick update.

Let's see. Photoshop added Rotate Object to their beta.Which this was- I tried this... in Illustrator first, but yeah, this is...

How does it work in free? Really good. So you give it a 2D image, and then it'll turn it into a 3D object that you can rotate. Yeah, so you're seeing on the video here.

Yeah, and then the last click was for harmonize. Yeah, exactly. I mean, look, for, for most Photoshop out- users out there, this is a game changer. Yeah.

'Cause that is the one you battle with the most is changing 2D perspectives of stuff. Yeah, and it's in your Photoshop tool, which is already, like, has tons of powerful features, and- Right... is a powerful photo editor. So, like, this is great for, like, really dialing in those first frames or anything, product shots, whatever.

Yeah. If you wanna change the color of the cat, you can then pick the cat and go into Photoshop Nana Banana and change the color of the cat. Like, you have all of that at your disposal at the same time. Yeah.

Yeah. That is a good, that is a good... That's a really good, uh, update. And then last quick update, Firefly.

Adobe's AI image generator can now be trained on your own art. I think we saw this in a preview, but it wasn't out yet. Yeah. If you go to the Firefly website, it just kind of says, "Contact us."

Um, so I'm wondering- That was, that was- That was-... I think like a custom Firefly model. That was like an enterprise thing. Got it.

Okay. This seems more like Midjourney-esque refs where you can give it, like... Or the mood board feature where you kind of give it, like, 20 images of, like, what you kind of like, and then it kind of s- you know, it steers the output to look like that. Perhaps it's, uh, training a LoRA under the hood and then giving you, like, a waiting function to then influence your generation.

Uh, yeah, maybe. Okay. Yeah, you were just there, um, on that website. No, that was the enterprise thing you were talking about.

Okay, okay. Here's their... This is their press release from yesterday. Yeah.

Expand a series of photos with the same look. That's cool. Explore character across different scenes. Custom models.

Ooh. Custom models are private, designed for commercial use. Designed for commercial use. I like that.

So illustration style. It's upload 10 to 30 JPEGs. Uh, okay, so, like, this is a illustration style. Aim for score.

Oh, wow. I mean, it literally looks like it's tr- yeah, like you said, I think it looks like you're training- Yeah... a LoRA. Yeah, but it's, like, so user-friendly.

Yeah. Look how, look how easy it is to visualize your data and then have, like, a score of how good your data is, and then it's doing the training off- offline sometime. I guess it's gonna tell you when it's ready. Yeah.

Now, it does seem like all the demos they're showing are for, like, very specific illustration styles. Sure. So, you know, I wonder how well this would work with something that's, like, a, you know, photorealistic or if you're trying to do, um, something photorealistic with, like, a certain type of cinematography. Kind of like our- Right...

one battle challenge thing that we did. But for illustration styles, this looks great. This is cool. It's, it's usable for a lot of, lot of sort of marketing, social media campaign stuff, I think.

You could- Yeah, especially if you have a design, you're, like, a, an aesthetic- Right... and you wanna just make more stuff in that aesthetic. This looks like a great use. And yeah, so this is consumer-friendly, accessible, custom model training, not the enterprise thing that they offer that is- Joey, we were talking about references versus LoRAs.

This looks like a big LoRA thing. Um- What's up, man? I still think, uh... Like I said, they weren't showing photorealistic demos.

I still think, still think photorealism, you want that... You're, you're going for the reference. Okay. All right.

Um, so- Well, when I get my hands on this, you know the first thing I'm trying is the desert highway. Yeah. [laughs] Yeah. With a Dodge Charger.

Yeah. All right. Good place to wrap it up. So we had a couple of big videos.

If you didn't catch it yet, watch the Comfy UI CEO, Jolin, getting interviewed by us, as well as our coverage of the Netflix acquisition of Ba- Netflix company, Interpositive. We had a lot of comments and thumbs up and high points, and we were partying and giving each other high fives. Just kidding. Joey and I still haven't seen each other in person.

[laughs] But, uh, thank you for your support, y'all. We really appreciate it. Did you see what the supposed price tag of Interpositive was? Is it $600 million?

Yeah. But I wonder if that's part of the whole package deal of, like, for the movie, the d- the movies too. I don't- Ted Sarandos is already spending that $2.8 billion that a year in France with the Paramount deal.

[laughs] He got that. He's already like, "Yo, yo, you can have some. You can have some." He got that breakage fee.

He's like, "Here you go, man." [laughs] Well, thanks for everything we talked about at denoisepodcast.com. Thanks for watching.

We'll catch you in the next episode.