Welcome to the Denoisies: Our 2025 AI Awards Show

0:28All right, welcome back to Denoised. we need a little AI theme music. Okay.

0:35Welcome, folks, to the future in filmmaking. Step right up. The past year packed into one shiny podcast episode. Hooray for the Den Wazi, the smartest show in town today. We've got the bots that draft your video. Should I stand up and march? you may bring your top hat and uh and cane and start dancing was that suno that was suno and i could not get it to pronounce denoisies correctly then kept go doing denoisies need a little bit of adr love there i know so well thanks for that i'm in the mood now we're in the mood all right let's uh let's kick it off uh we're gonna start with some obvious ones best image model yeah and so we're just looking at 2025 stuff stuff that came out this year what do you got i think i know you You actually don't.

1:29I don't think so. Okay. So, no shade to Nano Banana Pro, which obviously is a really good model. But I think the best image model for the community, for the creators, for the filmmakers, and the one that'll move things forward is Z-Image. I was expecting something like that. Okay. We do also have open source categories, too. But obviously, this is the best model. This is your pick. So, yeah. Okay. Z-Image. You think Z-Image was the best one? So as models are getting heavier and heavier, bigger and bigger in size and complexity and all that, you still need a model that's lightweight enough to run locally.

2:04And one that hobbyists like you and me can run and experiment with, can build custom models for, can fine tune. And these are things that make somebody that's just sort of messing around with AI into like a bona fide AI professional. because then you learn the inner workings of what the neural network is and how you can tweak it, what CFG you need, what steps you need, how many lures you can run concurrently. And it's like working on a 1965 Mustang. That's how you learn about cars. You can't just go out there and buy a Tesla. Like Google's Nano Banana Pro, I think, is like a Model X Plaid. It's fast, it's efficient, it's luxurious, but it's not going to make anybody fall in love with cars you know what i mean but whereas the image is like that that little it's like a miata it's like that thing is also fast but you need to work on you to put in the work you need to figure it out see okay we're going with this analogy so my pick is not a banana pro yeah and then i'll run her up is like cdream because also i think they've done really good stuff but in that analogy uh you know with the lauras and customizing the models it's like well the nanobana pro if it just does it so if it just gets me to my destiny It gets me there faster in the most efficient way possible.

3:17So, like, we're still getting to my destination. You're, like, tinkering with your car and, like, you know, having fun doing it. Wrenching it. But we're both getting to the destination and, like, I'm getting there. Yeah. You're getting there in style. And efficiently and with, like, the least resistance possible. Okay. I'll give you another reason why Z-Image is better. Yeah. It's better for the planet. It's just a lighter weight model. It's relying on less GPUs for each call. and I think there is always a lever like NVIDIA is always going to want to encourage people to make bigger models. They could sell more GPUs.

3:51I'm guessing. I'm not, obviously. And overall, if you add that up over like millions and billions of calls to the API, like that adds up to a lot of energy, a lot of water that's needed for all the data centers. So if there's another direction of the models becoming wider, like it's a push and pull, then I think we can kind of mitigate the impact we're having on the environment with all that. Yeah, we've talked about this before and they've talked like, I mean, they do come out with the big super turbo models and then they distill them. So like we get VO3 and that takes a ton of resources, but then you get VO3 fast and you get these faster.

4:26Yes. Lighter models, but they're still obviously not running locally. And one thing with Zimage, it's just image generation. There's no edit ability, right? But there is Quen image edit, which is also another good open source. Yes. Edit model. Yes. Similar to Nada Banana, where you can be like, change this thing in an image and it'll change it. Yeah. Good pics on both ends. Yeah. I think it's good pics. Yeah. I mean, I'm glad we don't have the same pics because that'd be pretty boring. Best video model. Kling 2.6. Yeah. That's my... I went with VO3 because I feel like they... As the first one, because I feel like that...

4:58Well, for kind of looking at 2025 as a whole, I felt like that was kind of the most definitive, impressive model. Yeah. But yeah, in the last month, Kling has just been on fire. O1's been so good. O1 is so good. And then 2.6 and even last week, 2.6 Animate with facial performance. Yeah. Excellent. Yeah. Like some of the best so far. So they end the year strong. I love rooting for the underdog and, you know, like what's Kaisho, the company versus like Alibaba or Tencent, right? Like or Google. No shade to VO 3.1. it probably is the best model by like detail rendering and um understanding the prompt prompt adherence and all that stuff but clink 2.6 is just it's like the people's champ to me in the like it's it's neck and neck with vo 3.1 and yet it's just like little model from a little company out of nowhere so i like that yeah i think we got another people's champ ish this one's from a bigger well i'm sorry i'm jumping ahead because um this is my pick but best open source release one 2.2 one 2.2 animate specifically animate yeah i feel like across the board one 2.2 has just been impressive as a model of course and we you talked about how even just using it as an image generation model yeah really good right uh video generation model open source animate yeah also excellent yeah but that it's from was alibaba right yes yeah so not a little player but still a great open source model yeah and the fact that they're just giving that out i don't know what the grand plan is but it's good for the community short term we get to again same thing i said about zimage we get to learn how to mold it manipulate it you know we have a working model that we could throw onto our gpu into comfy locally and then we can mess with and do really cool workflows and features and um i haven't seen as much with one 2.2 but i was digging into like some mcmuppets workflows uh this past week and um he had that he had that pencil one that's a new one yeah that And that one, it's kind of the workflows were a lot simpler than the other ones because it was a lot of just Sam, Meta Sam 3D.

7:02Oh, he uses that? To just extract, like, to do the masking. Whereas before, a few months ago, with his workflow, that was, it was sort of like a Runway Olive-y thing, but I thought that was better where you can shoot a real footage and your actor is holding on to stuff. And then you kind of replace the environment with AI, but keep the actor the same. Yeah. And that workflow is built on WAN 2.1. but then you had to mask the stuff beforehand and then you had a kind of whole comfy workflow for that yep and now just a few months later it's just like yeah we don't have to deal with that anymore just sam sam 3d does a depth map and detects the person and extracts them right uh and then his the new workflow it was more of the flip where you shoot a live object and you can use like puppets or sticks to move things and then just erase that and i gotta say that's animate that i love all his videos and uh yeah we'll link to make mumpets down below the pencil test the with the puppeteering quality wise the best out of everything he's done because it takes into account the environment lighting and relights the subject so well and just you get obviously gets rid of the pencil that you're holding those object with and then animates that and then you get that control of like actually like i want my thing to move here and here right and he had a workflow even a few months ago that was click and drag kind of control but it was like a very cumbersome process now this is literally like plus out your camera like it's bring in real world control in there bust out your camera move the thing yeah and then and then you know just make your film if somebody was to make uh honey i shrunk the kids today like that's the workflow that would go with yeah yeah no i wanted to build like a big giant bee that they could ride on a green screen blue screen all right next category biggest ai comeback i would say open ai sora interesting okay how do you see where so they were the first to the video generation market at least here in the US.

8:50That was like the first big splash. And then they went away for a while. And then we had VO2 and all these other models and the Wands and the Clings. And then the app dropped with a meme creation capability. Yeah, Sora 2. That was pretty damn good capability-wise. And then I never really used Sora 2 as an API function too much. But from the results I've seen online, pretty impressive as well. So, yeah, they kind of had first place, lost it, came back to pretty close to first. I don't even know how I'd say they had first place. They had first place in the zeitgeist when they released, when they teased Sora.

9:28But then we talked about this before. They sort of like announced it, showcased it, and then didn't release it for months. And by the time they did release it, it was like we had Runway and Veo and all these other models. And when they did release the first version of Sora, I was like, okay. Oh. Okay, cool. Yeah. But Sword 2, yeah, I mean, that did take over the conversation again this year. Okay, what about for you? I said Google. I mean, at the beginning of this year, Google had, their model was like Imogen. They had Gemini 2, which was like, okay. I don't even remember if they had a video model, and if they did, it was on no one's radar.

10:01Yeah. And then in this year, Nada Banana, VO3, Gemini 3. Yeah. Like, just across the board, dominated. Yeah. It's like the Chicago Bulls in 96 or whatever. they just yeah they were they were they felt out of all the big tech companies it felt like they're a bit asleep at the wheel for ai but then they just like had this massive comeback this year yeah and they're now have like leading models in every just about every category well the video model vo has been in the works for a while because uh and i could be wrong about this but i believe a lot of the moon valley people that are building the mari model are x vo people or yeah or they're X-D-Mind, right?

10:40Right. Well, I mean, it's VO3, so I mean, I'm sure there was a VO1 in June. VO1, yeah. I think that part of that team is an offshoot of... Yeah. So not saying it wasn't in the works, but they shipped a bunch this year. Yeah, and we only cover film and television applications. Like, you know, DeepMind is also cranking out world models, Genie. I'm sure they have autonomous driving capabilities somewhere in their physical robot stuff. I mean, they'll also have Waymo just vacuuming up. Is Waymo under DeepMind? I don't know. I don't think it's at DeepMind, but it's Google. Yeah. No, they're doing their thing.

11:14Yeah. They're getting the data. Yeah. All right. Biggest drop-off. So kind of just like company or model that was hot in previous years, but kind of didn't really hear much from them this year or just not on our radar for us. Yeah. Again, the AI startup world is just so brutal, right? Like you get a little bit of money to do this thing and then the revenue is unclear. And so I really feel it for all of the startups out there. So no shade, no knock on any of these guys. I'm just not hearing a lot from two companies, Leonardo AI and Midjourney much. Oh, Midjourney. Oh, those are good ones. Yeah.

11:52Okay. My picks were kind of like Pika and Crea. I remember Pika was sort of the first, one of the first ones was sort of that like modify video or ad. didn't we do the experiment with them with the squid on the table thing and um that was the first like vfx model yeah where you could like take an existing footage and like add things into it and then i don't think that was like the last big thing i remember hearing about from them um one of the most important lines and uh sort of lessons in the ai world that i and you know i obviously work in the ai world is that you don't have to be the best model or the best ai company you just have to be in the right room.

12:30Like with people that will use it or with whatever. The decisions are being made at an executive level with or without the knowledge of technical capability, right? It's how you sell it. It's who buys into it, who invests in your company and all that stuff. That's where a lot of the success and failures happen in the AI startup community. My guess is Leonardo and Maggiore are both sort of falling behind on that side of it. What did Leonardo was one of the ones i never really signed up and dove into like what what unique things did they have because they're also like a platformish too yeah they're like a free pick where it's an aggregator like they host a bunch of models but they had products specific to special uh like creative needs like this is like two years ago but they'll have like a pencil sketch model or a 3d rendering like your outputs would be in that style yeah but like this is a while and it's products it's hardened like yeah um leonardo was actually my very first ai subscription back in like 22 or 23 2023 yeah and at that time they were the first one of the first with the ios app like you can generate stuff on your phone that was pretty mind-blowing like they were much more ahead than anybody else but you know um a lot of companies unfortunately aren't able to keep up that lead for too long yeah yeah i remember seeing them first at the cinema synthetica that we filmed the first one last year in April of 2024.

13:55And like Matt, one of the guys would like use Leonardo a lot. And I remember seeing him. Yeah, they had a real time thing in there. Like, you know, like five frames a second. Yeah, I remember him using I think they had a kind of face swap Laura thing. So you take an image and then kind of mask the face and replace it with like another face. Oh, that's cool. You had. But yeah, I know you mean that. And I haven't heard much from them. Yeah. Pika. I mean, Kraya can make a comeback because they are, I think, still like ahead with real time. Yeah. image and video and they have a real-time video model kind of announced but i just really haven't seen it pop up i think the creator community doesn't know what it's missing as far as having a real-time solution yeah like uh before unreal and unity the computer graphics community didn't really know what was possible with real-time rendering and then once those two game engines were big and popular all types of things happen you know not just in games but live activation virtual production and all these applications so i think uh it's it's a much more niche and segmented market but if somebody figures out real-time image generation video generation like that's gonna unlock a lot of things yeah and image journey was a good one that you brought up uh i think they i feel like they shot themselves in the foot by not doing apis oh because like they they still don't have an api and you know it's more and more things like more models come out and then you kind of move to an aggregator or you want to build stuff into your workflow right there's no option for mid journey so it's like you gotta go to the website and you gotta use it yeah so people are not they're just gonna move on yeah with the easier tool that can spin up quickly right um and then also i don't know enough about their architecture but like are they just a diffusion model and now as more things go kind of multimodal and have world understanding like not a banana and do they have a video model yeah they added a video right recently i think we covered on the podcast is that still i don't know i don't think they're like world understanding i think they're just a diffusion based or oh gee it's like uh you you know, video gen 1.0 model.

15:49Yeah. So like, I don't know if they, if they're just entire architecture as Google caught up and all of these other big models that have multimodal understanding and remote understanding are just going to beat them. Like already have, man. I mean, or if there's even any way for them to catch up or if it's like they'd have to like completely rework this stuff. I mean, I think they're saving grace. They have a meta partnership, right? That's right. Like meta is using them under, in the backend. Yeah. Meta's ad platform. Oh, the ad platform too. I know like the chat or whatever was like, if you were like, I want to make a fun image.

16:21Like it's majority is what's spun up. Yeah. I'm surprised meta doesn't have their own image model. Maybe they just found other ways to do like, like Sam three, uh, three, yeah. They're like, everybody's making this one thing. We're just going to move on and make the other cool, cool stuff. Yeah. Okay. Next one. Most overhyped story. Hated this one. And, uh, stand by it. Tilly Norwood. You know, you can't say this one because you fed into it because you contributed to the noise around this one. I did. We did together. You asked the question to Xavier and I was like, oh, we're going to bring this up in our wonder.

16:57Yeah. You're like, at two o 'clock, Hattie? Well, we're trying to wrap this up. Yeah. Look, I know what I think about it. I'm really curious to know what leadership in AI world think about it. And also leadership in the film and TV world. I think everyone thinks this thing is ridiculous except for Variety and The Hollywood Reporter and Deadline. I mean, hats off to whoever the hell their PR agent is because they did a killer job over spinning a non-story into this ridiculous story. It's so bad. Yeah, and like anybody that works in the AI stuff knows how easy it is to create a synthetic human and keep consistency.

17:34Like, to make a Tilly Nora would take me like 10 minutes. Yeah, I mean, purely because they called this an AI actor. Yeah, you're right. They ruffled everyone's feathers. but this thing i mean even when you went to the website it was like the most ridiculous stupid trailer hype reel stuff that was like absolutely out of train your dragon rip off with her in a yeah there's absolutely nothing in there that would qualify as an actor thing um yeah this thing is ridiculous what's your most overhyped story it was tilly norwood and you weren't allowed to steal it but damn it sorry no it was that i had to i have two other ones i mean tilly norwood definitely takes the cake but um i wouldn't say this was unnecessarily overhyped but But DeepSeek ended up kind of, you know, that happened sort of in January and people were freaking out.

18:18It ended up not being as, I mean, it really didn't have much of an impact at all. But I think it was appropriately, people maybe overhyped at the time, but appropriately hyped immediately. NVIDIA stock dipped right away. Like they lost billions in value right away. People, you know, we thought, oh, look, we can replicate what's taken all these companies like billions of dollars, super cheap. It was$6 million to make. But then we actually really don't know what happened. Was it a novel new architecture or was it just a distillation of chat GPT? Probably distillation. I mean, nothing, what else has come out of that since they first launched?

18:54Like we have a really... DeepSeek's available as an API. It hasn't really had an effect on the industry. The only effect it had is that it made people doubt, do we need bigger and bigger scale? Right. But then that also didn't slow any of the scale down. So nothing has slowed down. Yeah, this was on my radar. The other one, the Trump movie tariffs. We covered that story and it got no views. We're like, we're not doing Trump stuff. We're not doing political commentary. But this was movie related and everyone was kind of freaking out. But then it was also like, how do you even enforce this? And then absolutely nothing happened.

19:36All right, this one, I had to think about this one. Most underhyped story. I have an interesting one. So we covered both of them. The two research papers from iLine, Go With the Flow, and the other one we recently covered. They're underhyped. Here's why I think they're underhyped. There is, obviously, Netflix is the biggest mover, the biggest shaker in the industry, if not one of the biggest, right? It's either Netflix, Disney, or probably whatever combination of Skydance and Paramount. And if they are doing cutting-edge research on AI to use AI in film and television, we should be looking in that direction.

20:13Because that is what's coming down the pipe, whether you like it or not, that technology is going to get successful. It's going to scale. It's going to impact every filmmaker from the highest to the indie level with the type of technology that they're building. So, you know, it may not impact us today, but five years from now, it absolutely will. And just sort of the stuff in general that's coming out of Violine or what Yeah, they have a team of researchers dedicated. And it's not just them. There's Disney Research in Zurich, right? They're doing amazing cutting-edge work. We should be paying attention to these guys.

20:46Yeah, Disney had that really cool, the de-aging, or the just re-aging research paper of de-aging and making people older, too. Yeah, and then we're in the sort of content business. Research papers is not good content. Like every time I cover a research paper, I always caveat it, but like, I'm the least qualified person to like go through this. But here it is. You know, when Sora 2 as a meme launches, that gets way more clicks and way more buzz. Yeah, because people can see the application of it. But the research papers are the ground zero of like where these things develop and come out that eventually get adapted and turned into products and useful things.

21:21But this is where you see how it starts. Yeah, absolutely. And also the timeline from research to product with AI is like months. Yeah. It's not years. which years later we'll see this, we'll see this applied in actual production, weeks or months. What's your most underhyped? I got a few equally. Comfy UI explosive growth. I feel like this year Comfy UI really kind of like, especially in the film industry, like cemented itself as this sort of next gen node-based AI generator, but also like future, potential future VFX. And yeah, really got like, I think went as this kind of under the radar tool for like hardcore AI nerds to like a much more mainstream.

22:01I agree. I think before 25, Comfy was like the nerd researcher tool. After this year, it's the everyday tool for anybody that's serious. That's serious, but yeah. Yeah. You and me use it all the time, right? Yeah, exactly. And we're trying to like, yeah. I mean, as much as the node-based idea has taken off in other tools and like kind of now every online AI tool has node-based tools, they're a completely different node-based system than Comfy. Like Comfy is another level. comfy's been doing it the longest the node based online is just like a different type of ui for the same thing you could do on the wrapper on top of whatever yeah handy for some things but but but comfy is a completely different uh beast yeah i always compare comfy to unreal engine blueprints like that was such a game changer right and unity didn't have that they just had scripting and that's why a lot of developers went into unreal to do a lot of cool stuff it's just easier to Visual scripting is just way easier.

22:56Yeah, I mean, under the hood, Comfy is just a big JSON script. Exactly. But it's nicer to look at the notes than a big wall of text. My other one, Wizard of Oz Sphere, the success. The Wizard of Oz, the AI upscaling and the process like that, probably got overhyped with everyone's reaction to the big faces and changing the frame. So that was maybe on the overhyped side? That part was overhyped. Yeah. But then - The actual, the money's made. The show came out and then it was been doing like record weeks and like, I don't know, like a million a day or so a week. Yeah, I think they sold 15 million seats or something insane already.

23:34Yeah, and it's been a really big success. Yeah. And like there's an audience, people are going to it and paying the 200 bucks the seat and trying to like, you know, figure out the next movies and stuff to go through this process. I think they did select another one down the pipe. Gosh, I don't want to feed into the rumor-ville, but yeah, they're working on it. Harry Potter was Cosm, right? Was Harry Potter announced? yeah that was a cod another cosm yeah okay but yeah i mean i just showed like there is a market for this interactive fun reimagining of older movies classic movies yeah i agree um more on the sphere later i actually have them on another list okay all right next one biggest acquisition or deal there's one that came to mind but it's i have a save for another category so the one i put was like figma and weavy yeah i mean it's it's not a small deal that's a medium level deal no and i think that's also like taking something that's semi-established tool in the traditional graphic design space and now bringing in a big ai player yeah and do it and merging the two together i think and that felt slightly bigger than like adobe and invoke or more more impactful in the design world yeah i think the figma folks this is just a guess but i i feel like they have a better chance of extracting value out of Weavey than Adobe does with Invoke.

24:52Yeah, and building it into their core product. Yeah, yeah, because Adobe is just so much more legacy. Like, they've been around for so much longer. The ecosystem's fully built out, paved, and, like, it's just, like, rigid thing. To put Invoke and sprinkle Invoke onto all of their products, that's going to be a really difficult job. Yeah, and I think it would probably just either be, either they integrated into Firefly, which is already kind of this newish experimental thing. Right. But it's like they're not going to build it as a legacy products for maybe years, if ever, like once they see adoption in the newer, more experimental products like Firefly.

25:25Right. Whereas Figma is a much newer company, more agile. And I mean, they're big, but they're not Adobe big. Yeah. And they could incorporate this better into their system. So it's good for us. It's good for all of us. I'm trying to think of more acquisition of deal. There's, I mean, the big one is I have saved for the next category. Yeah. what's that what was yours i mean you're probably going to say it on the next one the disney open ai partnership yeah i'm about to just merge these two because my next category was most surprising partnership well i have a surprising partnership okay it's suno and warner music group oh that's a good one that's a good one okay dude how is suno going from the most hated to actually like yeah let's shake hands it's all about money in the industry yeah yeah well there's probably a rev share model i don't i i didn't dig into that as much as uh yeah i probably should have but did they in that agreement did suno have to like neuter anything they have to like tone back what they do i mean i just used suno to make this cheesy song and um i mean i i didn't well i mean i didn't have any roadblocks or anything but i wasn't trying to copy anything that was like suno's been in a lawsuit from the day like yeah but i mean in this partnership and with warner that Suno have to like put more roadblocks in or tone down like what they could like styles they could replicate or like props they'll reject no I don't I don't think so I think it's the opposite is Suno will probably get access to training data from all the Warner Music Group artists and their songs and then as it as that stuff is trained and it's used for inference they'll probably kick back some of that revenue back to Warner Music Group so like it's a way for them to generate income just from the library they're sitting on.

27:05Yeah. Yeah. But I'm curious on the consumer and you're using it. Does that change what you can do with music? Does that change what kind of stuff you can make? Yeah, that's a open ended question. Yeah. I don't know the answer to that. But yeah, that was very surprising to me going from like the bad guy. Yeah. I mean, same for me, you know, Disney and Sora. I mean, I think that's also the biggest deal this year too. But it's more surprising. It's not surprising that Disney's trying to find a footholes and just kind of take a gamble on like where the future of this stuff's going and uh when Sora was first announced and we talked about like oh it would make sense if like a big entertainment company kind of partners up so they can put their IP in there and you can like make cameos with famous characters sure yeah I think it's just a big lot of where is this where is this going where is this going to go how is this going to play out is the AI generated stuff that's going to show up on Disney plus are people gonna like it care not care well that's the least I think that's the least of it um I I think and I we covered this on our open AI episode was like every business segment of Disney has different needs for AI right like imagineers need a certain type of AI versus television versus film versus gaming versus whatever animation and to just have a broad Rob Strokes, OpenAI partnership across really limits you in terms of what you can use and what capabilities out on the market.

28:30I don't know if that means that they're restricted to like, they only have to use Sora and OpenAI models, or if they're still able to. They'll definitely be probably incentivized to. Yeah. And I mean, they'll have access, I'm guessing, to their developers, maybe make custom solutions and stuff. But I would guess if it's like the R &D or ground level stuff at Imagineering or Disney, that they're just messing with everything. Yeah. Yeah, I don't know what the actual intent for Disney is in partnering up with OpenAI. I thought it would have been Google. Like, that made more sense to me. And then the day after we recorded the episode, I found out about the Google lawsuit.

29:07That they were suing. Disney was suing Google. Google for IP infringement. Yeah, the same day that they partnered with OpenAI. It's like, basically, middle finger to you, handshake to you, right? It's like, that's not, I mean, that doesn't move the industry forward. No one can use our Darth Vader images, except Sora. Yep. Yeah. Well, yeah, we talked about this last time. We'll see where that goes. But surprising. I mean, yeah, I think the big thing is that he was trying to hedge, put a bet somewhere. And yeah, Disney is also something they can invest in. They're actively building out their AI team called the Office of Technology Enablement, OTE.

29:42So like these companies are trying to figure it out. And they're partnering up with who they think is going to give them a leg up. So I don't blame them. you know at the same time yeah no i gotta yeah i mean all right biggest missed opportunity yeah so for me it was capitalizing on the success that the sora app had like the fact that it was the number one downloaded app for that week and millions of people used it there was all types of like ip initially and it was it was like in the zeitgeist yeah but then like everyone was like what are you doing and then they like completely yeah toned down what you could generate with it and then everyone was like, this is not fun anymore.

30:20Exactly. And so I think that was the biggest missed opportunity. You think they should have just come out with like slightly more IP restrictions? No, I think they should have went after licensing rights and just paid Disney for the Mandalorian. Yeah, just like, hey, we have 15 generations of the Mandalorian take$15 or whatever, right? Like just start paying for IP so the platform gets bigger and bigger because the fun in the generation is you taking a selfie next to Tom Cruise or whoever, right? Like that's the cool stuff. And when you take that away, it's kind of like completely neutering it.

30:55Yeah, I mean, yeah, definitely. I mean, look, after we messed around with it, like the first week, I haven't really touched it again. Yeah, exactly. Yeah. Mine was, I was trying to think of something here, but basically like shutting up about your AI spot. So, you know, we talked about a lot of the commercials that came out and like they're AI generated but then like the companies the people behind them just like keep talking about like the AI process or just talking about the AI stuff and then it's sort of like you know turns people off or they revolt whether the spot is good or bad I mean we both think the McDonald's one in general was just a bad spot in general but then it people have this reaction and then it kind of gets this backlash rightfully or wrongfully like I think netflix did the smart move with uh the uh the eternats um yeah where the show came out you know it was out for a bit months later watched it yeah much later like oh by the way we used ai on some of the vfx shots right and then everyone's like oh okay and it's like you already watched the show you liked it don't didn't like it whatever but then there was no like revolt when the show first came out of like oh i used ai like yeah it's also not their like tier one it's not stranger things no right and so and it's not telling you that they used ai on the week the the show drops like that has a big negative impact exactly i'm like they were smart about it i'm saying like put the ad out and then later and be like oh hey we did this thing with ai and this was the process used uh i think a lot of these ad agencies and the companies themselves like shot themselves in the foot by kind of doing this whole like we made the spot what we can do Yeah.

32:29Look how we replace humans. Yeah. All right. Biggest surprise of the year. I mean, for me, it was the first time I used Nano Banana Pro. I was like, holy shit. We can do this now with AI. the level of text adherence, propped conformity and details and just diversity in people and age group and frame composition, camera language, like it just nails preserving everything. like changing one thing, like not messing or redoing everything else. I was not expecting that this year. I think I would have been more in line if it was next year. Like, oh yeah, you go from Nano Banana 1 to that in about 18 months.

33:14that makes sense yeah but that was like within three months yeah yeah that was the biggest surprise for me mine was uh some of these i'm trying to remember what i put why i put these down cloud code oh sure you tried it i tried it i use a lot now like just as like ai well like claude's models i've always been a fan of but then just ai that can run on your computer and like actually do things with files and like rebuild stuff or just like ai chatbot but like turbo charged. Yeah. And it's got a like a UI, a developer environment. And yeah, you run it in terminal. They have been moving it. There is a Claude code now in the Claude desktop app.