Why Sora 2 Launched a Social Media App

OpenAI's Sora 2 launches as both a powerful video model and controversial social app. In this episode, Joey and Addy dive deep into how Sora 2 handles IP permissions with its opt-out approach, test its uncanny ability to generate likenesses, and examine the innovative Cameos feature that lets users put themselves into AI-generated scenes. They also discuss the implications for filmmakers, debate the quality differences between the free and pro versions, and share their actual test results with screenshots from the app.

--

The views and opinions expressed in this podcast are the personal views of the hosts and do not necessarily reflect the views or positions of their respective employers or organizations. This show is independently produced by VP Land without the use of any outside company resources, confidential information, or affiliations.

Transcript

0:00

All right. Welcome back to Denoised. It is time for the weekly roundup of all the crazy AI news. SoraNotSora. SoraNotSora. This is your episode. Let's get into it. Alright. So if we're putting the, the show together today, uh, would say, not a, not a crazy volume of AI updates, but one big. Overarching story this week. So two, probably one of the funnest ones we've had to cover. Regrettably, I did get sucked into the fun. Let's cover it. Yeah. Okay, so, Sora 2 was announced. I would say my initial reaction, we were text about this, I was just like, uh, okay, whatever. But there were a couple things that were weird about this. Okay. The new model came out, you know, bigger, better. Higher quality. Yeah. Does video, uh, does audio with the video generation, the thing that stood partly, maybe I feel like they fumbled the initial launch so hard. Okay. Because last year, you know, it was announced and it was like, oh man, this is like game changer. Yeah. Huge leap in quality. And then they kept it private for months and months and months. And by the time it finally released Kling, Veo 3 wasn't out yet. But, uh, Kling, Runway, everyone had pretty much caught up, Luma, with quality. Wan 2.1. 2.1. Yeah. Uh, and then by the time Sora was released as a product to the public, yeah. It was like, okay, well we've got like five other models that do the same, if not better quality. Yeah. And it seems like forever ago, but when SOA was announced and we saw the initial videos were, it was like, wait. AI can do video. It was that sort of reaction. Yeah. It was like, oh wait, this is ai. I think I saw some of the clips before and I was just like, okay, what weird clips. And then I realized like, oh, this was generated. Exactly. It was like, like with the reflections of the woman in the subway. Yeah, it was, that was, it was quite mind blowing. Yeah. Yeah. I think in the newsletter, I called it the, like the Lumiere train crash moment. Okay. Of ai. Yeah. Where it was just like, you know, people are freaking out of like this new boost in technology. Yeah. But where it never really. Shipped to the public and no, it wasn't their hands. Kinda, yeah, they totally botched that launch. I mean, I think they had, yeah, internal safety testing concerns, which fair enough. But it just took a while before we delivered as a product. Yeah. Anyways, fast forward to now, and then this comes out. Sora 2, but not only is it a new model. Mm-hmm. It is also a social, social media. A social networking, TikTok style app. Yeah, totally. And a bigger focus that looks like more on memes and media, social content yourself and making. Short, social, funny videos versus a, a filmmaking tool. Yeah. It seemed to me it's equally as strong of a video generation model as well as a story building narrative building model. Like an, think of it as an LLM, if you will, because Yeah, some of the, and we will show you some of these examples, like you give it, Hey, I want to make a video about me going to Costco and finding a giant roll of toilet paper. It turns it into a funny bit. Yeah. It makes a ten second bit. But it's not just like a single video output. It is No multiple shots. Yeah. Uh, sound design a script. If you, if you give it a script, it'll follow the script. If you say, I wanted them to say this, it'll usually make them say that. Yeah. Uh, and then if not, it'll just generate some random audio that sort of fits in the story. Mm-hmm. I'm guessing it's something similar to like what we've seen with, uh, Seadream. Yeah. Or Seadance. Um, whatever the video model. Where they realize like in the same generation you can create multiple shots. Mm-hmm. And so you're in that same laden space and you get that consistency mm-hmm. Of having that same world and just having multiple shots generated inside of it. Yeah. Although, I will say the shots sometimes don't line up correctly. Right? Yeah. So like you're in one place and then make shot over, although you're in the right universe and we'll show you the gladiator example. Yeah. Here in a completely different part of it. Yeah. The other thing that was big with this is this new feature they're calling Cameo. Yeah. Which is. Basically how you can scan and insert yourself into the videos, but it's sort of an interesting take on granting permission. Yeah. For the AI model to use your likeness. So you have to like take a video of your face, read some numbers to sort of verify it to you and like look side to side. Yeah. I compare it to like the apple face Id calibration a little. Yeah, I can Easier version of that. Yeah. Yeah, it's, I think it's a really neat way to control your likeness and especially one, you know. One thing opening AI is super careful with this deployment is getting into deep fake territory, unintentional deepfake territory, sort of. Yeah, we'll talk about that in a second. Okay. And the other thing that you mentioned with Sora one launch was, uh, alignment and safety was a big concern, right? So they had to kind of step back and figure out, mm-hmm. I think they nail the alignment here, although there is a lot of precious IP that you could generate that well. Oh yeah, yeah. When you're saying alignment, what do you, what do you mean by that? Like, I can't use somebody's likeness in here without their permission, right? Mm-hmm. So let's say I take a picture of my cousin and then I upload it on this app and that it's like, make a picture or make a video of this guy doing something really silly. Mm-hmm. We're like, we detected a real person. Are you sure you wanna continue? We're like, yeah. Actually you can't. Mm-hmm. You don't have that guy's permission, so. The likeness and like protection on using, uh, the likeness of real people. Yeah. Is one interesting protection they built in. But then there's the other separate issue of copyright and ip. Yeah. Which they have taken a very interesting approach on. Uh, and they're basically saying that it is an opt out method. For I using IP and copyright from the copyright holder. So if you don't want, uh, people to be able to create Mandalorian characters in their video Yeah. It's up to Disney to then go tell them, Hey, we want to opt out of the Mandalorian character and they have to do it each specific ip. Disney can't just tell them the entire, we don't want have our library in there. It's like, no, you have to say, we don't want Mickey Mouse, we don't want Donald Duck. We don't want, uh, the guru Mandalorian. Yeah. This is a very new approach and interesting. The default state is we're gonna grab all publicly available data unless you opt out, which is, well, they've grabbed it, we, they already grabbed it. It's there. It's in there. It's in there. If you want people to be able to make uses. Generate videos with the, that, that involve these I IP characters. Yeah. Like, so I mean, some examples you've, you could ask for, you know, make a Peter Griffin Family Guy episode or something. Right. And it'll make it, and it'll sound like, uh, Seth McFarland. It'll sound like family guy saying whatever you wanted to say. Yeah. This was the part that also came out and was pushing that just kind of made me feel like, like this, similar to the Higgsfield Steal thing, where it's just like, this feels a bit icky. Totally. And yeah, I was like, Hey. Um, and not a good, um, argument for like, you know, using AI as a tool for filmmaking when it's just like, Hey, you could just rip off any IP that you want. And it's, it's so like, built into the model that unless they put a top layer of do not use on it, it's in there for Right. As long as SOA exists. All the Disney IP will exist within it. Yeah. And we know it's there. Yeah. It's just if they're now their guard wells that you generated or not, which I, we'll, we'll show this in a second. I did some testing and it was very inconsistent with what it let me make and what it prevented me from making a hundred percent. The same experience. Yeah, same. I was like, wait, you made that thing, but you can't do, right. Yeah. But they're in the same category. Well, yeah. Well, let's show some examples in a second. But let me just go back to the cameo thing.'cause I think that's interesting.'cause the cameo, you scan your ID and then it's tied into your quote username, and then you can set permissions of who else can call up your cameo. Right. Is it no one? Is it just friends? Yeah. Is it mutual people? Like you're following each other or is it anyone? Anybody. So like Sam Altman made his. Public, anyone could use it, which I would hope, I'm glad he hope he would, being that he's pushing this app. I was like, whatever. What the hell? I made my public. So if you want to do some crazy thing with me at C47, and I don't know if this is just me like just being in the app for the last 24 hours straight, but I feel like this is the best implementation of digital humans, quality wise. Yeah. Let's leave this one. I made. I said, make me a gladiator. And now it's multiple shots. Uh, some of these shots, the likeness is freakishly. It's good on some shots better than others. Yeah. Some shots good. Some shots not so good. Joe's like, I'll take those chiseled arms. Yeah. But I mean, here, it's, it's clearly imitating that gladiator movie, Russell Crow movie. Uh, for sure. Yeah. I wanted to see where it was like pulling from, uh, different ip. Uh, this one I, 'cause we're both friends and characters. Yeah. Prompted it with both of our usernames and I said, make us a Miami Vice clip. Yes. Uh, we could see in the, and also these are very basic prompts. I didn't really give very detailed prompts. You could get better quality if you give it be detailed prompts. Yeah. But like, let's look at the digital human quality here. Like that. That's me. And that's you. Yeah. That, yeah. Like, could tell. You could tell It's sort of us. I mean, this one, it oh my god. Made me look like a weird, it's a little much. Yeah. Yeah. This running one, I look kind of weird, but the lighting. Yeah. And this one made me have a big giant nose. This one I gave it an image of a car of, uh, my car out in the desert, and it was just one. Still frame, but then, then it made all of these orbits and like the, well, that's not, definitely not my car really. But um, for the most part, I mean, this feels like, yeah, the environment and the space, but it's so playable. It's such just such a playground of what they've built, you know? Yeah. This turned into a look. This one is, this looks like a selfie video in the car. Yeah, like this looks like footage. I just shot of a video that's coming out except for the car itself. Not the car itself. But this, that shot is great. Yeah. Looks, yeah. Pretty, yeah. Close to me. Yeah. Also, it's funny because when I did the face scan, this is exactly what I was wearing. A black shirt with the khaki top Yes. Has become my default outfit. And then I, uh, I'll show the things that failed. But I tried to like replicate scenes from movies, so I was just like, okay, gimme the casino Royale chase. And so my initial prompt was me, but make me James Bond chasing a guy through Port-au-Prince. And got the environment right. That one failed. No, that one failed. Oh, it said, oh, you can't do the copyright, content violation. And then I just said maybe a super agent chasing someone through. Port-au-Prince. Oh, so you know how to get around it now? Well, then I just took out the specific names of the ip. Yeah, just don't put the keywords in there for the ip. But here, this is where things got weird. Some of these prompts here, I was trying to make characters and I said, Griga is sitting on top of Gandalf's shoulder as they walk through the, uh, Nostradamus, the alien spaceship. Yeah, I saw that one. Their, their app is kind of buggy at the moment. It's a little overloaded. But this was a weird thing. It it, the one I sent you, lemme try. Find it. Um, it worked once. Oh, did I send it to you? I saw it once. I saw it once. It made once where it did make Grogu sitting on, maybe it wasn't Gand Griffin. Yeah. Yeah. I had another one where I said Grogu sitting on top of Peter Griffin's shoulder. Right. And then it ran it once. Yeah. Successfully. And then every time I tried it again it just disappeared. Yeah. It would say no error. It seems like there's like a read, uh, a right issue on the disc. It's not saving it correctly on the cloud, that's possible, but also more so, not so much where is this video, but more so why did it let me make it once, but then it didn't let me make it again. Oh, avoid the content violation. Right. That's what I'm more curious about. So that's why I'm saying this Content protection, well, A, it'll just tell you it here. The content violates our guardrails concerning similarity to third party content. Yeah. It won't tell you what the issue was. Yeah. So then you have to kind of play the guessing game, but it was literally. The main characters were the same and it let me do it once. Yeah. And then it flagged it every other time I tried to generate it. I think it's, it's updating in real time as people are putting more and more keywords in. It's looking for matches and then just updating a giant database, a lookup table of things to not generate, I'm guessing. Right. But like we're, that's why we we're saying that the training stuff is like, the data's there. Data's there. The guardrails for it to prevent you from using it. Yeah. Are right now hit or miss. Okay, so this one I did Wednesday and Reacher because I had heard Wednesday Adams had was a character that people had made tests with Wednesday and Reacher, walking on the bridge of the enterprise. You got a Netflix character, Amazon character. I wanted to combine as many IP worlds as possible. Into one. Yeah, into one thing. Oh, you know what? It's black. The screen out.'cause I turned screen recording on. That's why it wasn't Oh. Oh. What? You don't want me to rip your thing? You ripped off on that. You have a content, but the screen, if you go back and view the screen recording, it's there. The screen just goes black. Oh this? Yeah. It grabs the frame buffer maybe. So yeah, I wanna show, so I sent you this over text. Yes. Maybe you can overlay it for the viewers. I mean, look, I literally prompted for the Mandalorian, if you can see right there. And there he is. He's in an LED volume and I'm in a mocap. That's Mandalorian kind of sounds like me. Gave me a little bit too much around the waist. But yeah. Excuse me. Take, take, take. Uh, but, uh, yeah, that, that headcount is laughably inaccurate, but it's like. I don't think I've ever felt the AI regurgitation more of real stuff more than this model. Yeah, because it's like, okay, it's got the little elements that it definitely pulled from the behind the scenes footage of the mandalorian of the, like something looking like a desert. Mm-hmm. It's got the Mandalorian Mandalorian in there. It had the two monitors at the, it has the sand on the floor at the brain bar station, right? Yeah. This is, this is just the ads. Some of the ads you tested generating, it's like, yeah. This feels like you could see some of the weird text in the corner. Yeah. Too. It's just like, this feels just like the regurgitation of everything. Mm-hmm. Mm-hmm. Here, this is the other one too. Grow, grew sitting on top of Peter Griffin's shoulder. Yeah. As they walk through the Nostradamus. It didn't get the Nostradamus, it would just put him in a castle, but it got, grow grew, and it got Peter Griffin. Okay. Yeah. And, and then note that Peter Griffin is in 3D, even though it's a 2D character. Yeah, yeah. Brought him in nicely. Oh. It made him say Nostradamus. This when I got the enterprise kind, I mean, kind of a version of the enterprise. It has the yellow shirts and the red shirts. Yeah. So it understood Star Trek. Right? Even though it just said enterprise, I mean, nailed reacher in, uh, Wednesday. Yeah. Yeah. Yikes. So what is, what do, what do you think is OpenAI strategy on content violation and protection? Okay, so the social app thing, I was like, this is weird. Why are they making a social app? Right. It is because. If you have a social platform, you are, there are protections. You are not liable for what the users post. Oh, if you're the host, yes. This is new territory because they're enabling the generation of the content. Yeah. But they're trying to go with, back to that, I think is the DLI Digital Millennium Copyright Act parody law. You're protected on a parody law. No, it's not parody law. This is a basically what protects Facebook from liability from like if people post. Terrible things on the platform that Facebook isn't liable for what people post on the platform got.'cause they're a social network. Got it. It's the same, I forgot the specific law. Okay. But it's the same law I think they're trying to go with Oh, by having a social media app? Yeah. That they're not liable for people posting stuff that infringes on copyright. Yeah. Even though they also built the tool that is enabling people to make the stuff that potentially is infringing on copyright, they're, they're just kind of getting by on a technicality. Yeah. I mean, they're, they're pushing this law to the new limit. Yeah. I'm sure there will be lawsuits. Maybe they want the lawsuits to settle this, to just get this stuff cleared out and sorted out. No, they're like, I'll pay a couple billion just to get this out of the way, and everyone comes to a deal and it's li eventually gets licensed, so maybe, maybe that's what they're going with too. But the social media thing, I'm like, why are they doing this? Yeah. That, that's partly why. That totally makes sense. They're doing it. Now I also think, 'cause I'm looking at this quality and I've seen quality of Sora 2 videos. Mm-hmm. From some creators that had access before when it launched and the quality they posted, it looked much clearer, much more cinematic. Way better. Way better. This, I just have regular ChatGPT. Yeah. I use an invite code. I'm guessing this is more like a SOA two fast model. They didn't call it that, but. I haven't been charged for anything. It has this little watermark and it's watermark. It feels like, I don't know, crap out it. Would you say it's a 480p Yeah. It feels very soft. Very soft. Yeah. It feels very, uh, soft. And a ai e. So there is another version of SOA two that is accessible or will be accessible through. The, oh, like a pre premium version? Yeah. There's a lot of body cam footage stuff, and, and the Martin, the Body, Martin, Martin Luther King ones are freaking the body cam footage stuff's also crazy too, because it still has the axon. The company that makes the cameras, it still has the axon. Oh, really? Timestamp in the corner and their Lugo, because that foot, that logo is on every single body cam footage. Oh, axon. Sure. Yeah. And, and so it again, to the what? The copy paste of like reality. Oh my God. I mean, this thing is like pixel for pixel trying to recreate what it trained on. It knows the stuff. It just like is, is blur regurgitating the stuff that it is trained on? I guess that's the, that is what you have to do in order to be highly photoreal. Right? Like you can't, that's why it feels accurate. You can't like, do, uh, classify free guidance per se. Like you can't creatively generatively fill. You almost have to, uh, re you know, respect the training data a hundred percent, 99%, right? Mm-hmm. To create photo realism. So that's, that's maybe the biggest difference between something like this and some, something like one, two, 0.2, animate where that is made for creativity and. Outside the box thinking where this is. Yeah. In different controls, just trying to replicate the things that it knows exactly. Yeah. Yeah. Um, okay. Here in the Sora blog post announcement, uh, so R two will initially be available for free with generous limits to start so people can freely explore its capabilities. Uh, though these are, these are still subject to compute constraints. ChatGPT T Pro users will also be able to use our experimental higher quality SOA two Pro model on soa.com and soon the SOA app as well. And we also plan to release SOA two in the API. So if you're on ChatGPT PT Pro, and initially when they launched soa, if you're in the ChatGPT PT app, there's like a separate SOA tab where you can, then they had, they did f some cool controls and stuff. Uh, so there is more of a pro level higher quality version of soa mm-hmm. That has more of these. Higher quality demos that I think we've been seeing a lot. Yeah. More cinema that I've seeing on YouTube. Looks nothing like what I generated. Exactly. It looks like it looks much better quality like, like kind of like ray three or kind of like Veo 3. Right. So there is a separate level to soa. Okay. But the one that's getting the most attention is the um, the social media one that phi the world of everything. Yeah. I'm not a big fan of vertical video generation. Anyway, but that seems to be the only thing we got at the moment. Yeah, you can, in the app, you can change it to landscape generating. Oh, okay. It's just still the same kind of lower risk quality. Yeah. I mean, look, uh, this stuff is free or at least you know for now. Yeah. Perceivably, it's free. Somebody's paying the cost somewhere, right? Like this, the running this many GPUs for this, I'm guessing millions of people mm-hmm. Is not gonna be cheap, but OpenAI is biding the costs just to get, I think two things, one. Initial data of what users are using, what keywords are being violated, like what do they wanna make? Yeah. It's just they're, they're beta testing on us. Right. With, with, uh, with millions of people. Yeah. And, and then I think the second thing is like, they are probably gonna turn a lot of this into paid. They're gonna convert into paid customers. So this is a way for them to sort of take the strongest 10% of users and turn them into the pro plan and then have them generate away. Do you think that's gonna be any significant amount of money for like the scale of like what. Chat of what OpenAI of all the things they're doing. Yeah. This is what we talked about a few episodes ago. It's like, uh, you know, content video generation, in my opinion, it feels like such a small slice of the pie versus the entire AI util utility pie. Right? Like we talked about ChatGPT by. As an LLM being used for legal paperwork, real estate paperwork, finance paperwork, reviewing security cam footage, whatever. Right? Yeah. Right. Like those things being so much higher in utilization than like generating Mr. Rogers. Right. I think the meme videos are like cutesy, but it'll like die off. I think this, it feels like the companies that are making generative video models are kind of falling into a couple different buckets and like some are going seem to be going. Initially more after the high end Hollywood production. Yeah. Like this is gonna be like the future of like movie creation. They wanna win the big studio pipeline contracts and that the company is dialed in for that. And that's Runway, that's Google, that's Luma in some parts. That's Luma sort of, I think Luma's doing both. And that's the other bucket that I wanna talk about. But Luma, Luma Ray three is for film, right? No, Ray three for sure. Yeah. And they have their, their, their Hollywood studio. Yeah. Uh, AI studio that they're opening up. So Luma I think is straddling both worlds. The other world. Being personalized video at scale. Um, massive video content. Yeah. Uh, U gc UG, right. Or, or computerized AI generated content. Uh, at scale. At scale. And that was what Amit and, and the from Luma, the CEO and the interview, 'cause he was talking, that was one of the other use cases he was talking about, of just like, oh, what everyone has their own personal kind of video feed or their, like, personal video generated for them mm-hmm. Of like the news stories or what they want to hear, what they wanna see. Mm-hmm. Uh, so this. SOA and especially the SOA app feels more in that bucket. I know that they are, you know, they have reached out with filmmakers and have tried to kind of fit that in the pipeline, but I think it seems like for them they're kind of growing more of the route of video for everyone. Generated models. Yeah. And business strategy, although, I will say, because the model is so good, uh, particularly at human anatomy and physics. Mm-hmm. That I think the folks in Hollywood who are testing with ai, they can't look away from this. Like, I have not seen digital humans of this caliber. From like a single scan, single photo of what Cameo does. No. Of what? Yeah. Yeah. Like also I did mine in the back of the car yesterday of my authentication. So I'm also wondering too if ah, if I did a, if I do a better, I didn't realize that the authentication was also how it was like scanning my face for the, for the video and your, and your costume too. Yeah. I was in my tank tops and now all my generations I'm like, oh my God, I look terrible. The other thing I wanna say about Sora and that, going back to my initial, when this first was announced and I was like a bit. Like, okay, cool. But whatever the, whatever part was also, yes, the model looks better and the stuff looks more realistic, but at its core we're still doing a text prompt or an image prompt to video and I'm just waiting for, I know something to evolve and my theory is and the pieces, are there something from Google that is combining Genie 3, which was that real world model? Mm-hmm. Where you give it an image or prompt and you're like moving around in a space, something like that. Combined with Veo 3, which would be like Veo 4. Yeah. Where you can like be in a space, you can move around in it. Yeah. And you can find your shots, quote shots like you would traditionally, something that moves beyond, like I gotta give it a text or an image prompt. I can actually be in the space and find the shots that I want. I agree with you. I don't think that solution is gonna come from big tech like a Google or an OpenAI, I think that is gonna come from somebody who's deeply embedded in Hollywood. Somebody like, uh, Moonvalley or Luma, maybe Moonvalley or, uh, Runway or, or someone. Yeah. Uh, using or, or, I mean, Runway has sort of laid the vision of that. Yeah. For Aleph. Yeah. And, and, and, and, and, um, Cristobal audio demo. So in, in the same way I think, or a comfy with like ion, uh, 3D world, which also just came out as well. Could be something hacked together, we're, we're gonna see like a type of control being innovated on the m and e side. Mm-hmm. And then big tech come and grab that, essentially copy it. But I don't know, I mean, I still feel like Google is there because also like. I mean, to make a real world model, a world model like that, also you need like a lot of compute and money. I'm not judging by the, the quality of the model, the size of the model. Google has that. You don't say the specific use case.'cause like the, the actual, I'm actual user, like where I just went, I wanna have my, my fake camera. In my AI world, it's the ux, it's the user journey. You know, like how is a user actually gonna be able to control it? And you're absolutely right. Like no matter how much you text a prompt, you can never get the thing that you want. Yeah, you can write a thousand sentences and I'll get the thing that you want. Right? I'm gonna be in a world and be like, oh. Put a chair here, put a person here. Okay, cool. Gimme my virtual camera. Okay, let's, and then next shot. Da da, next shot. Yeah. And then that, that was kind of the thought process behind JSON prompting. Right. But that's like, I mean, that's just the, that's like text prompting on steroids. Exactly. Yeah. But that's not the solution. No. No matter what's the stop gap for now, until it keeps getting better. Yeah. So that's what I'm, that's that's what I'm like, I'm waiting for a new interface. Like I think like some, some sort of sketch to image on a shot level. Maybe get us there. I mean, and that, yeah, that, which, uh, was Veo 3, which people figured out you could Yeah. Sketch to direct. I mean, first frame. Yeah. But then you also gotta ingest a ton of references in there and each reference has to then, you know, you could tag it, I think in Veo 3. You can tag it in other models as well. Like, let's say if we're generating the, the Martin Luther King thing, right? Like you, you cut to the audience and then you want the first three people to be this person, that person, that person. How are you supposed to do that? With this model? Oh, with this? Yeah. I mean, you can remix clips here and you can tell it what you want different, but yeah, it's not, yeah. That level of control isn't there yet. Maybe it's in the pro version. Yeah. Uh, I don't know. I don't have chat. BT pro, I got rid of that. Oh, you're not paying 200 a month anymore? No, I'm not paying 200 a month for ChatGPT, was not, was not worth. I was gonna borrow your login. Damn. She's gonna make a bunch of deep fakes with me. All right. I think we covered everything we can about Sora. Yeah. If you, uh, have Sora videos you wanna share, just, uh, comment and Yeah. Ly you need an invite code. I think we got a couple between us, so you can just leave a comment. We'll forget if we got one still. We'll, we'll hook you up. Yeah. If you wanna mess around with it. Yeah. But look, regardless of how you feel about Soa, I, and of course it's infringing on, you right. Issues, we're gonna hear about this very shortly from the lawyers. From Disney or whoever. Regardless of how you feel about it, play around with it because I think it's really important to know what the capabilities are as of today, as of this moment, you just download the app, play around it for an hour, delete the app. I think it'll do you a a lot of other. There was one thing I did think about because when we were talking about the IP and the, the opt-in, opt out, you know, I think just expanding right now with this opt-in model. I mean, I'm sorry, with this opt out model, you know, all the companies are kind of forced to go like, no, no, no. I wanna like have more thoughts about that. But I could see this as on the flip side, where it's like, oh, if you're releasing a movie and now you want to give people the opportunity where it's like, oh hey, make a video of yourself with the character from the movie. Yeah. This feels like a great promotion option. Yeah. With the cooperation and partnership of the movie and the people involved in the characters and all of that. Yeah. But it's like, well, at least we know that tech is there where this is now possible. Just cleaning that up legally and working in conjunction with the studios versus like, Hey, you gotta like, let us know if you want to get out of this. Yeah. Opted in by default. You know, I think, I think that this is gonna open up a lot more marketing channels and just interactive ways for combining traditional tech and ME movies. That's a great idea. Movies with, uh, you know, new ways of creating marketing experiences. I think younger folks would love to interact in that way with an, yeah. What if you could be in, what if you could be singing on stage with uh, K-pop demon hunters? Yeah, yeah, yeah, yeah. You could be singing Golden with them. Well, I'm just thinking like how it integrate. I think the way, I think Gemini integrates with TikTok now, so you could have some, um. Like Nano Banana stuff in TikTok, if I'm not mistaken. Well, no, I would guess TikTok would be using Seedance and Seedream 'cause that's their models I would be guessing. But then somebody, but no, a friend of mine who's a really Veo 3 is in YouTube shorts. You can spin up Veo 3 from a text prompt in YouTube. So. Sora too, would just kind of integrate and plug into, um, the hood of the car, if you will, for TikTok. So like all the social media wrapper that you're seeing now, this is just to kind of, uh, circumvent the lawsuit thing, right? This, they're, they're not really gonna turn this into a social media app. OpenAI is not in the social media game. I'm saying the Sora technology is kind of. Been built from the ground up to be social media ready. Yeah. And you can usually download any video make here and upload it to TikTok or your platform of choice, or, I think on TikTok, you get to choose the Soro two video generator and then make your own thing. Oh, like it's calling it up? Yeah. Yeah, maybe. Yeah. Yeah, yeah. I think the thing that Sora has nailed this first go around is. Being able to scan your face and have it, have your likeness in its library, nailed it. And permission control so that you could just be like, yeah, add me to this thing on Freepik. They have the digital twin avatar thing. Did you ever do that? It's like 5,000 credits. I didn't it. Yeah. You, you train a character. Have you done that for yourself? I, I've did it a while ago. How is it a, I mean, I don't know how it's building it and I think just giving a single image to Nano Banana works better than the character generator, but, um, I don't, I think it's. I don't know if they've updated that tech to, to be Yeah. Current with everything. That's, that's the thing available now. This is state of the art today. Yeah. But by the next AI roundup we could be talking about something else. Completely. A hundred percent. All right. Now, definitely enough with soa. Okay. All right. Okay. Next one. Completely opposite of generative AI stuff. Sort of a new product launch, uh, called Mosaic, which they're calling a, an age agentic video editor. And it is well dissected in this thread from, uh, Aish Jane. But it is looking like a node-based, uh, wey looking interface. But for video editing, which I'm trying to wrap my brain around how this works, I would imagine it would be similar to what a nano banana implementation in Photoshop is. That's still like layers and you're like, okay, let me modify the layers. Right. And you're like, okay, I wanna nano banana, modify this layer. Right. But I'm still in my traditional layer based editing. Yeah. You're still back to by Photoshop. Yeah. Okay. Clip one, the canvas, ideate and iterate. On an infinite interactive canvas so you can connect your tiles of your video clips, and it looks like a series of different video clips connected with nodes. And then that, I guess he's running it through some different prompts of like how he wants it edited, and then it's bringing it down into a more traditional timeline sequence. But it was like you, it's like a combination of like prompting for the edit, connecting the clips you want. And then bringing that down into a traditional timeline. Yeah. It's not like rocket science per se. It's just a neat. Neat way to put those two things together. A generation and a timeline. Yeah. I mean, you know, also we're digitally old. I've edited on, uh, non-linear editing tools Your whole life. My whole, yeah, I was gonna say like, definitely more. Yeah. Longer for the majority of my life. Right. That's how my brain works for video editing. Absolutely. Even like cap cut and the newer, uh, iPhone tools, I'm just like, uh, that's what, so, you know, this could be something that resonates more if you are new to editing. And that your brain is still malleable and open to new ideas. Yeah, I mean, I'm curious to explore, but look also kudos to them.'cause I see here in the other posts, like you can build out the cut here. And then also it supports XML, uh, XML exports for Davinci, Final Cut ,Premier. So, you know, I always love when it's like new editing tools. Take a new take on editing, but then also give you the support to bring it into. Other tools that still have sometimes more powerful features that, uh, you just need Yeah. To, to do that you can't really replicate in a web-based interface. Well, what I wonder about these companies like Mosaic is like, first of all, how many users do they have? How many paid users do they have, and are those paid users? Paying because there's such a pain point from what they're getting with Premier that they're willing to pay that extra to go with Mosaic, right? Like, where's the market for this and what is the eventual, uh, game plan for these companies? Yeah, I mean, I think usually they kind of find their niche more in like people who are trying to deal with video editing, but they're not video editors and like that's where like De Script or V or uh, you know, that's where they, I think they kind of found their sweet spot where it was like, oh, we can hand, you know, podcast editing. Uh, vlogging presentation stuff. Um, yeah, the, uh, where it's an internal small team, it like falls into the lap of the marketing person. Uh, they're not a video editor that wants to figure out how to learn Premier Resolve. Yeah, sure. Let me clean up my stuff. But like, like you said, Cap Cut, Veed,io and to some extent Freepik as well. Like these guys have Freepik's been building out a video editor too. Yeah. Yeah. All, a lot of this stuff is already almost there. I just wonder like why there's so much overlap between the AI startups in the world. Yeah. Uh, I think on the app generative I AI integration too, which also makes sense in the video edit. Yeah. Where you're like, I have been wondering when Premier I know is integrating some of this stuff, but like when, even if you're just like, oh, I just need to have a placeholder shot and I'm editing sometimes that's felt like, yeah. I wish that was more integrated into some of the editing tools. For sure. It's funny, uh, Adish Jain is I think Jain is, the same last name as Amit. Uh, the Luma founder. Oh yeah. I'm not sure Jane related. Jane. Yeah. Yeah. Anyway, his, in his words, in a world tending towards AI slop, create something real. And you gotta, you gotta, X is a, X is a, a game of hyperbole to cap to capture attention on X. What the heck? Yeah, I, I'm curious to see where, yeah, I'm curious to mess around with this. Yeah, look, I mean also, you know, I'm excited about a new take on video editing.'cause I looked at a node based thing and it was like, how do you. Edit that way, but also in fusion and resolve. Mm-hmm. It's node based, it based compositing, which initially took my, took a little bit to wrap my head around 'cause I used to like just layer based compositing. Four man's nuke After Effects Fusion. That's my nuke.'cause I can't afford real nuke.$300 for a lifetime. Licenses. Pretty deal. Pretty good deal for Resolve. I, I have a physical card for that. Did you ever get one? What? Like a dongle, like Resolve dongle? Uh, no, the, the. The key. Oh, the license key is on a physical plastic cart. I did, I do have some of those from like when you bought a camera sometimes. When you get an Ursa you get one of those. Yeah, yeah, yeah. But if you don't have that 300 bucks for lifetime license is a pretty good deal. Fair deal. Yeah. All right. So yeah, we'll, we'll check out more of that, but I wanna put this on everyone's radar.'cause um, for sure new takes on editing and, and solving tools is always interesting. And that's why we do on the roundup. Exactly. We cover new tools also. You heard it here first, this roundup is, was, um. This week was mostly swallowed by soa. There were not a lot of other updates. W we had, we had way too much time suck with Sora with that we had to share with you. And yeah, I think it was more, well, I mean, I think just in general, I didn't really see a lot of updates unless we, if we missed stuff, let us know. Um, minor updates too. Uh, Nano Banana had an update. The big one is now there's much better control with aspect ratio. Great. We had this back and forth. The upload, the last image has to be the upload order matter for how it did the aspect ratio. Uh, now apparently you can just set it and it'll, um, honor it. And then the other it said image only output. I still don't understand what that means. Captain obvious at Google raise their hand like, guys, why don't we just do a dropdown menu of the aspect, right. We just tell it what we want it to do. And then yeah, the other minor update more. Excited.'cause I like, uh, Seedance now also supports first frame and last frame. Oh, amazing. Yeah. Last frame. Always having Last frame is like a super handy for ance. Quality's good, man. I like, yeah, yeah. Don't sleep on those Chinese models, dude. No, don't. I, I, yeah. Talking about, I know Sora just came out, but like those Chinese guys, they're working. I been talking about Seedance and Seedream for a while, but yeah, they, they're some of my favorites. I mean, yeah. I use one all the time. Yeah. Like I said, 'cause it's free, it runs on my pc. Mm-hmm. And damn. It's good. Yeah. Can't wait to use 2.50, I don't think. 2.5, which we covered in the last episode, but the other minor update with that as well was um, you can give it an audio file, I believe, to drive the performance. Oh, nice. Okay. Which is actually something you can't really do with Veo 3 right now. You could, it would just be. So like, let's say it would be like an animated movie. You cut the dialogue together, character A, character B, character C, whatever. Give it that and then give it reference images of each character. I think so, yeah. I think you can include. Similar to like a, like a, like a Hedra or Heygen we can give it an audio file and drive for performance. You can, I believe, do that now in the one, 2.5. Like that stuff is already so ready for all the, all the like the mindless hours and hours of YouTube content for like really young kids. Like toddlers. Yeah. Like spell A, b, c, like all that stuff. Like I don't understand why there's not enough of AI generated content in that category. Probably is or is growing. Money there too. Yeah. You know, I mean, I like look at Coco Melon, look at Blippy. Like these are, but see those are, those are, you know, I mean they built a brand and they built like, you know, sort of a defense where it's like yeah, they have the identity and the uniqueness. Yeah. Um, you know, I feel like with the AI spinning up stuff, you could generate a bunch of stuff, but how do you get something that takes off and, and feels the same, resonates. Yeah. Or that like, you know, like can stick with, I don't really know how much brand affiliation is with, with toddlers, but like that, that, you know, feels like that breaks through and, and becomes a, a, a bigger thing than just the one-off videos. Yeah, totally. Yeah. It's an interesting problem to solve in crack. Yeah. Yeah. I have heard that YouTube children programming is difficult. To monetize and crack into. Oh, really? Yeah. Just because there's, I think a lot of people break trying to get into this space. Oh, it's ultra competitive and uh, and the price, the, the ad sense is different for kids videos'cause there's more restrictions.'cause they're kids. Yeah, of course. A lot more content security stuff. Yeah. Built into it. It's just the targeting's a little bit less specific because it's kids. So I think the ad rates are a little bit lower if you're doing kid programming. You make up for that in the like, volume, right? Yeah. If something breaks through and kids like to watch stuff over and over again. Yeah. You get the, the repetitive views. Yeah, I know the, there's a couple of like unreal engine affiliated creators. I think one of them is called Silly Crocodile. They build, build a whole thing in Unreal. Yeah. And then they put it out there and it's quite successful. That makes sense. Yeah. Or like some blending where it's like, oh, what if you haven't Unreal. Merged with like one studio, cut to studio A real person. Like what Blippy does? Oh no. I was thinking just like even. Instead of animate, like if you build some basics and not unreal, but then you use something like Wonder Studio or some of the, uh, oh, like digital ai, digital character where it's just like, oh, instead of animating, just have people act it out and Yeah. And, and speed up your animation process. Right. I would love to get into that world, um, if I have the time. I feel like it's just such a, like a obvious low hanging fruit with AI right now. Yeah. Because with ai, the quality is not there for feature. Not yet. Right. Even if it too, we're not there. Kid stuff. Yeah. I'm sure people are cracking it. Maybe it's just under our radar.'cause it's never, I don't know. I'm not watching like kids videos that much. Yeah. Alright, cool. Thanks for everything to talk about at Denoisedpodcast.com. Alright y'all, so if you're watching on YouTube or listening on Apple Podcast or Spotify. Please, please, please give us a five star review. Three stars are not appreciated. Only five stars. Only five stars. Makes a big difference on the algo. All right, thanks everyone. Catch you in the next episode.