This Week in AI: 11 Tools Transforming Media Production

0:28All right, welcome back to Denoised. and it's a rapidly evolving world needless to say

0:37all right the first one uh and this one i'm still trying to harness get my head around yeah cloud artifacts and so this actually came out last week but we'll talk about this week cloud artifacts is basically cloud is known for being pretty good at coding this is now you can just type a prompt into your regular cloud chat and if you want to build a reusable web app it will build the web app for you and make it into a form that you could deploy online and just have a link to and share the link and people can use it yeah i think cloud code is the ai vibe coding app from them yeah i mean they're known for being pretty good at coding but usually it's like built into cursor or bolt or one of these other programs that if you're doing more complicated coding you know these apps use now cloud has pretty much kind of just done the whole thing in one i'm sure there i'm not a coder i'm sure there are limitations to how complex this can get yeah but if you look they have a an inspiration artifacts pack and some of them pretty wild and anthropic office simulator so this is like a 3d space simulator that was built from you could they have the prompt that it started with yeah it's crazy that you know this is now just an ai generated thing and we take it for granted yeah just tell it what you want literally five or six years ago somebody showed me a web-based uh office walk around environment visualizer and they they thought it was so groundbreaking that they were demoing it oh yeah and now you can just build it yeah yeah it's like i think that's why i'm trying to wrap my head around like where what little things like i feel like this would be useful for especially if you're trying to do if you're kind of the more technical like nerdy person on the team and the rest of your team is not like that into ai stuff as you are building little apps or something that can help speed up the process but is a easy to use interface that you can just share with the team so don't even tell them right something like oh you i mean we we do use a lot of summarizing and and and writing stuff with the transcripts of this podcast yeah and it's sort of built into well it's built into like all these automations and stuff but it could just be hey upload the transcript text file and then in the back end it adds all of the reformats it into the youtube description and titles and all that stuff and then just spits it out it's a web app that anyone on the team can use yeah it's like a nice front-end GUI into a more complex back-end.

2:49But you don't have to know how to build. Wrap the freaking API. An API wrapper. Yeah, it's a glorified quad API wrapper. Did you know, fun fact, this is what I heard and read on X somewhere, that Anthropoc wasn't even trying to work on cloud code. It was like a side project they didn't really care about. And now it's like the best VibeCode application. Yeah, like accidentally they happen to build the best out of all the LLMs. Like theirs is like one of the best at coding. Yeah. Yeah. I mean, I also think they're one of the best at writing. Yeah. Their LLM in general is really good. Yeah, for sure.

3:25Yeah. I'm looking at they have dashboards. They have I'm trying to think, too, for like editing tasks or something, you know, if you get a CSV out of like a CSV converter or something on your team, if you like get EDLs or CSVs, how can you convert that? How can you clean up data? Simple tools like that. I'm sure there are going to be a lot more complex tools draw a perfect circle game you click your mouse and draw a circle that's terrible and yeah 74 all right like these these are so time intensive and labor intensive to build manually yeah and the problem for this is i want to build a draw a perfect circle game user should be able to draw a circle and claude can judge if it's perfect or not i want this game to be full screen immersive yeah i mean this and that was it that was the problem to build this whole thing needless to say like obviously this is not gonna you know take jobs away because nobody's trying to build a draw a perfect circle out.

4:11No, these are things where it would just be too ridiculous to spend money on a developer to develop this, but it's something that you could spin out in 10 minutes and maybe, you know, it was like, it saves you some time. It's like the ultimate life hack. Like I feel like so much of this stuff is like, reminds me of life hacking when it was getting popular like 20 ish years ago. And it was sort of, the joke was like, you'll spend two hours to find, to figure out a shortcut to something that would just take you five minutes to do by hand. But you're like, want to spend the two hours to like automate that process.

4:39It's like come full circle with AI stuff. It has. Yeah. I would imagine like an individual app developer instead of, you know, trying to hire out or, you know, automate these little components of like a larger, bigger app. Now you can do a little bit of, you know, chunks of your application in AI. Yeah. Yeah. I mean, I'm thinking one of the limitations because I see another demo project is a project dashboard generator pulling data in. And Claude is sort of also, we've talked about mcp where it can kind of use the ai and access other apps on the web and bring ai to them things that don't necessarily have ai built into it that would probably be the next phase of this uh because i don't think it could pull outside data in yet but then if it's like you could harness clod to build an artifact but then also access data that you connect to it yes that's where it gets crazy yeah i think that's like a retriever augmentation generation a rag yeah so that has to tie back into a database of some sort.

5:36Yeah. One thing I was trying to do lately, and I hit a wall doing this, I'm surprised there still isn't a tool that does this yet, is I wanted to use one of the AI chat bots to look at a Google Doc and come up with some revisions, but then edit the Google Doc in the doc. Oh. And none of the tools will do that. They will yet, I'm sure this will eventually happen, but all of them can access the doc, pull the data in, but then I'm like, okay, thanks for the edit update the doc it's like oh sorry dude i can't do that you got to copy and paste it in and even gemini yeah even gemini you know which is built into google yeah won't let you edit the doc with gemini it'll like give you updates but then it's like you got to copy it in there's permission issues all over the place yeah but eventually yeah the the chat gpt's of the world still live in this like island it's still not connected to the rest of our lives i mean i understand we also have a lot of risks with that of just you think unleashing it into the world into all your docs and be like hey change everything where did all my docs go yeah i rewrote everything since day one yeah i mean it's cool with people build with that um if you have found any cool solutions with uh the cloud artifacts please let us know in the comments so curious to see what is getting built in the world yeah especially in the video yeah if you're video space if you're vibe gunning and building apps just share it in general we'd love to see it yeah all right next story from hey jen hey jen known for kind of creating those realistic avatars talking avatars.

7:03Not the James Cameron avatars. Human avatars. Yes. So they launched or are launching, it's in a private beta right now, a creative operating system. Yeah. This kind of levels up from where they were, right? So previously, H &N avatar systems, you upload a script or an idea and they have these realistic humans that you can pick from or you can build your own. And then it animates the human with perfect lip sync, perfect hand gestures, or holding a product, and then you have your video. And then you go and edit that video, then post it on social, and so on. It's supposed to sort of be replacing influencers, I guess, or augmenting them.

7:44Or I think a lot of it's like, oh, you want to do internal videos on your team, you don't want to hire a person, or no one on your team wants to do it, you don't have the budget for it, and hey, you can have this AI avatar read your script. Now it seems like they're wrapping a whole bigger service on top of this video generation, avatar generation model, which is the COO creative operating system. And this reminds me eerily of what we talked about in the last episode with Meta's ad platform. Yeah, similar idea. Yeah, this I mean, it's a bit more give us some give us some scripts, give us a footage.

8:14And we'll generate the A role at the B role where it makes sense kind of do the editing and a lot of the fine polishing for you. Well, I don't know how polished it gets, but we'll do a lot of the creative for you. Yeah, the AI generator thing. I mean, that I'm curious how they differ and behave. I mean, also the meta one probably can only you're probably stuck keeping those assets in meta whereas this is like yeah this is like oh you can generate it and and you know you have the assets to deploy however you want yeah it also reminds me of descripts underlord feature which is their their version of give us some footage kind of chatbot tell us what you want and we'll like edit it and trim out stuff for you you you talked about descript before yeah exactly and you know it's funny that uh and a lot of people will say this not just me is like we thought when the ai revolution was coming it was coming for the tedious like stuff that nobody wants to do but not the highest level of creativity and creative output and that's what's happening first like that's the first thing that's being augmented and potentially replaced yeah i mean replaced also you know i wonder too of if if a company is able to just generate a bunch more assets it's like this is the kind of stuff that's sort of just low bar cheap to produce stuff that they either would have never hired someone to do it in the first place or would have never had this much output in the first place right and so it's just more of like additional things that exist in the world that just wouldn't have existed in the first place or is it replacing some jobs it's probably going to be somewhere in the middle yeah it is yeah and look at this is not gonna be you know end all be all it's not gonna like replace people right like it the quality still needs work and we all know in the creative world especially in the film and tv world that 90 is just not good enough you got that last 10 is the hardest yes yeah and this in the ugc or in the like we just need like different things to test on tiktok when we run our hundreds of ads yeah yeah this will work for that i just worry that you know are we gonna be just inundated with ai generated avatars the influencer stuff is already a ton it's it's a lot to handle like instagram is practically full of influencers now you add another million accounts or something like that of ai influencers right like what does that do to yeah i mean we talked about this before i mean i still think if the real human influencers like it's not necessarily like i'm following an influencer because they're just talking about products all the time it's like i'm following them because they like kind of have a shtick and they have sort of like a funny story or a personal yeah or it's like a setup or it's like they kind of cover a topic and then every now and then you know they're on a sponsored trip you know for cruise or something and it's like sort of an ad but i'm also kind of like yeah good for you i'm glad you got that ad deal like because i enjoy your stuff and like yeah you love it go you know yeah go make that bank yeah i guess i don't actively think about it but the favorite my favorite youtubers the ones that i really follow closely i follow them because of their personality.

11:10Yeah. Yeah, I'll give you a good one. There's a colorist, DaVinci Resolve colorist by the name of Wakas Kazi. Okay. He's pretty big now. And, you know, I could care less about his coloring. Like, I just love his personality, his content, and his color, no pun intended, on stuff. Yeah, so I guess AI won't be able to replace that bit. Right, it still comes back, right? Are we talking about real human or are we talking about this? Is it like, yeah, AI, sure, it'll replace the UGC looking things of just like, oh my God, you'll never believe like I found this product type thing that are being pushed as paid ads.

11:44But is that a character that you're going to follow? I mean, we know that will exist and that does exist with like the little Michaela world already. But I still think, I don't think it's going to like just completely push out human influencers. For sure. But the little Michaela personalities are so arduous to build over time. It takes a team, right? versus somebody with genuine personality that is very magnetic. Like, that's natural. That's going out on Instagram and YouTube. And that really is the value of that person. Yeah. Anyways, coming back to agent. Yeah, so it has a private beta. But I mean, I think the agentic AI editing stuff, we'll just see more of that.

12:24But yeah, it'll probably get you 80 % of the way there. Maybe, you know, someone on your team isn't a dedicated video editor. They're the marketing person who can now edit videos or put some stuff together. And this is also where a lot of the money is, is just getting brand stories and brand products out there. So they're for sure developing stuff to monetize. Tools for e-com brands to help push out where they're like, they just need volume of stuff, of videos and content out there. Correct. Not necessarily quality. Okay, next story. So this is from Hunyun called GameCraft. And this ties into an update-ish from like a week or two ago, which I don't think we talked about, but Runway entered the game world.

13:00Cristobal, my bad. We'll talk about it on the next one. I know you're out there watching this show. We love you. Yeah, thanks for retweeting the AI, inside the AI studio clip. You should come on the show, Cristobal. Yeah, we'll get you on. We'll get you on. Anyways, they launched Runway, not this, not Hoon Young. Runway launched Game Worlds, which was sort of their first push into gaming and ties into the kind of real world model generative stuff that they've been doing. But right now the Game World, the Runway launch is sort of like it kind of generates a text-based game. but eventually it'll be like a you know you can move around in a generated world in real time hunyan already published something called game crap which kind of already does that ish and so the demo is moving around a generated virtual world in real time now it is just generating each frame so i don't think it's a persistent world so like if you travel forward and then loop back around it's not going to be the same oh it's procedurally generating that space it probably will hallucinate and not like recover the old world i right i think so like if you walk out of a building and then turn around you might not see that building again but it's generating stuff in real time in a 3d world and they got samples of like i mean roman world a car driving through a street yeah we're gonna show the video here look i i take the chinese videos that come out with a grain of salt because there's a lot of uh wizardry under the hood to make this look as good as possible like we had that one volumetric solution and then jim gadolta kind of exposed that there was like a real volumetric well i mean to be fair to them it was in the paper oh okay it was he exposed all of the uh the the x hype guys yeah saying that it was revolutionary not they didn't they weren't hiding it but they weren't hiding it but come on it wasn't i am looking at an x hype post and so i also am not looking at the actual paper and so i could also to be perpetuating a but like inaccurate just looking at like the boat going by the sea here i mean dude this is right up unity and unreal territory isn't it real-time rendering of a 3d environment in for gaming usage that's unreal engine like yeah either either we see a big neural renderer adoption within unreal engine and because you still need all of the packaging around unreal like the epic online services uh game physics logic decision trees like you have to build a proper game engine it's not enough to just have the visualization right like there's the mechanics of the game that has to be built out and either unreal adopts technologies like this or these guys build the bits and bobs that unreal has had for a decade but the future of gaming i think for sure is going to be neural rendering from the developer perspective or the gaming perspective like the user perspective or probably both the user doesn't care what generates the beautiful like if cyberpunk but i'm saying like if you're a user like i could see what different like genre of games where you're like you want to explore some new world that doesn't exist and you're a solo player and you're exploring and this world's being generated on the fly based on what you want to explore but then the other end of like riot or games or whatever developing a new game and using generative AI to help build the world, but then locking that world off so that when it ships out, same with what we've talked about with film, it ships out to users, everyone in a multiplayer sense is in that same physical world.

16:22Yeah, it can't be a random draw. Like you can't just keep procedurally generating random stuff. Right, I mean, that could be a new genre of games. This ties into what we've talked about with movies and stuff where it's like, oh, does it make sense to like - Choose your own adventure. Choose your own adventure, the movie that's just made for you that we talked about this more in an upcoming episode with an interview. But yeah, that it's like, how exciting or interesting is that if it's just like a movie just made for you that only you care about? Yeah, it's certainly unconventional and I think it will have a little slice of the action, if you will.

16:52But gaming right now is a shared experience, right? Yeah. Like nobody games in a vacuum. When you sit down on your PS5, that's connecting to the internet. You sit down on your desktop, that's connecting to the internet. And for everybody to share in a game, that world has to be precisely engineered and replicated across all the gamers. Right, so anyone that goes into this world, you're all in the same world. Yeah, and that's what Unreal is really good at, not just rendering the world, but making sure player X is in the position A, player B is in position B, and so on. There's a coordination effort over the internet that has to happen.

17:27I'm saying this is great as far as visuals go, but the bottom of the iceberg is the game mechanics. So if they can figure that out, or if Unreal acquires these guys because they've already got to figure it out, right? Like that is the feature of gaming is a hybrid between traditional game mechanics and neural rendering. I can also see this being useful for virtual production and stuff because you can just give it an input image and tell it what you want. And so then if you can have that consistent virtual space, which I remember we saw some demos last year with what's called World Model or World Labs, I think, where it was like you give it an image and it would build a faux-ish 3D model space that you could kind of move around and explore.

18:09I have a hot take on virtual production. Irrelevant in the next five years or whatever? Virtual production is not going away. The combination of putting physical actors in virtual spaces is never going to go away. LED volumes will go away. What do you think? And be replaced by either just like a empty space and we're compositing and being able to... like what light craft is doing but imagine an ai version of that well i mean it's a very expensive selling point but the selling point of led is like well everyone can see what's happening like the actors too and that it's like that's the like level up from the green screen compositing world also the lighting the reflections you do get the right yeah maybe there will be other solutions like we did see some demos at nab with projection yeah where it's like okay you can kind of do the same effect right but for a fraction of the way less of the cost you just roll the big projector blasted on the wall you don't have to do the entire build of the led wall shout out to ramiro and chris yes yeah awesome demo so yeah i mean i think i think it'll probably be like a lot of things like yeah maybe led will just be reserved for the like top top ultra high budget movies yeah but i i think the world already has enough and i think the numbers will continue to shrink until there is a steady state, like maybe 10 productions in the world at a time can use enormous volumes like the one at Studio Ulster or the one in Nant here in LA, something like that.

19:36But for the rest of it, if Lightcraft or tools like that are good enough and even highly empowered by AI, then do you need the electricity, the real estate, the HVAC system of... Well, what's also the deal with like just development of improvement of led in general like would it ever get to the point where like the panels can be way cheaper way lighter way less power sure like absolutely so could it that's our could we have a curtain green screen dish curtain but it's really just an led curtain you know some futuristic like led fabric i want to i want to decouple the two things so led panels is a huge business and it's only getting bigger because not just for live events and concerts but now you have location-based install stuff like the sphere and then everything is getting a slapped on led panel you go to buildings in new york they have led panels on the side right like so that business is constantly growing but the economics of film and tv production as you know i don't have to tell you this is not looking good and it's only going to be smaller and smaller so the economics of that is just going to force some of the led vendors led volume vendors to pull out of the of that game Right.

20:46Or why I spend so much focusing on building like high quality LED for the lighting. For film and TV. For film and TV, for the very specific needs of film and TV. Because you're only going to sell 10 ,000 units of it. Right. And it's like, well, we can make these cheaper if we don't care so much about that. Because for displays and outdoor venues, the color science doesn't matter as much. And that's exactly why you have like the Rho BP2V2 panels, the Absinthe panels. Like some of these premium elite panels are still five, six years old. Like they haven't iterated on it because there just hasn't been enough revenue for these companies in the film and TV world.

21:22But they will build new panels that are more fire resistant, water resistant because they're just going into more places. I mean, I'm just thinking like, yeah, there's just gonna be something that pops up that is so small, so much smaller, so much lighter. And that use expands beyond M &E to like anything like any type of display that could just be cheaper and easy to put up and light and, you know, blend with the real world. has value yeah for sure so like maybe something like that pops up in the next five ten years and it's like okay well led is like so like as cheap as slightly so you think it was a projector or something we're gonna see like uh almost like uh a slowdown of led volumes and then once new panel technology comes around we're gonna see a resurgence almost that's my one theory kind of like 3d tv where you know the hype was over and then all of a sudden with apple vision pro and everything we're seeing a re-emergence in stereoscopic stuff slight re-emergence dude they're dropping the avp air soon apple vision air mac rumor i mean that that is a rumor i mean we know something will be coming when i mean what's the time frame on that uh let me call tim cook like i don't think we're seeing anything this year maybe not even next year to your point yes i think real time generation will be widely adapted i still think there's value and everyone and be able to see what's happening, but unless it comes down drastically in cost.

22:46Yeah. Anyways, this is the tangent episode. Yeah. Sorry, folks. We'll edit this out. No, we're not going to leave this in. All right, let's jump to the next one. So this one is not necessarily a new product, but I thought this was interesting. So FreePick is one of the sort of popular sites of kind of AI, all the AI tools in one spot, kind of like a good aggregator, good creator platform, a lot of tools to like edit images, you can call up whatever model you want. they had a model where it was like you kind of play pay for a couple different tiers and you would get a certain amount of credits and whatever model you use uses different credits and it's usually just a conversion of like how much it is to use the api and they convert it into credits blah blah free pick now just made an announcement where they are going to get rid of all the credit systems and just say unlimited image generation um that's crazy so no matter what model you use if you use the free pick free picks built-in ai model or if you use uh imogen dolly whatever other outside model flux yeah all that no credit system unlimited generations if you're on one of the paid plans you know they're like yeah you having the credits limits creativity it gets in your brain i mean i experience this where it's like i gotta use these you know is this worth like these credits to like test this idea out or yeah it's like create with it i spent a hundred dollars gives me a thousand credits but i might only just need 10 credits for this one test yeah do i just leave the credits in there yeah or am i like i don't want to keep like experimenting or messing with stuff if like my credits keep dwindling how are they doing this i'm guessing maybe it's some sort of pareto thing where it's just like they looked at the data and maybe like 80 of the users are not using up all their credits and you know so maybe it's socialism it's credit socialism i mean if someone takes advantage of it maybe a super user you know what like will be offset by the people who like are more casual users super user paid for those credits well i mean the super user starts using beyond what the paid credits are for i mean maybe they use a throttling system because also you know the nice thing about claude or chat gpt well not necessarily claude because there's no image generation but chat gpt if you're making images on it there's no credit system with the dolly images if you're making them through chat gpt so it's like nice because you just keep asking you don't really think about the usage but if you keep using it a lot eventually it's like oh you know you've been using this a lot you got to like you know pause for a few hours like that's happened to me on claude where it's like are you usage is high so maybe they'll do something where it's like oh if you're just like cranking stuff out i mean they must have a low number of users because the sheer cost of infrastructure to support unlimited free generation is so high it's so high joey well i mean i think the thing is who's that like how many people are actually going to like really like overly take advantage of that yeah i mean you're gonna have a modest volume of people around the world like you're talking 10 000 images per minute around the world at any given minute i'm just just my guess like even that equates to you know i don't know ten thousand dollars a minute on aws or something like that right like who's paying here's my it's only for paid people on paid plans so you have to pay you have to already be paid oh so there is no free account no it's not free no that's crazy i thought they were giving you have to be on a paid plan no so their paid plans would normally like you pay bucks a month you get whatever 10 000 credits yeah now they're saying no credit well also to clarify too i think this is only for images so i think video video is right you can't go on this and start cranking out vo3 videos yeah exponentially that'd be awesome but you can't do that so i think there's still a credit system for videos you have to be a paid user and only for images so you're like a vegas buffet you pay at the door right that's what i'm saying like if like i think the people who are paying and are casual users are going to be offsetting the smaller number of people who are going to be the mega users yeah and maybe if someone's like a crazy mega user they'll throttle them or be like oh hey you're you know you've made like 20 000 images in an hour you gotta stop yeah i get you i mean i don't know about you but with me you know i started making like i started using chat tpt once every few days and now it's every day a few times a day with image generation too i noticed that my trend is upwards yeah like i generate more just super random stuff maybe just mess with some friends generate more when you don't have to think about like a oh this is costing me x amount of credits exactly exactly like i use chat gpt to generate they don't tell you and they're not like oh you've been making too many images that's eight credits yeah yeah no so i think it's a cool nice psychological change of feeling more uh open to just experimenting and creating uh joaquin abella the ceo free pick uh it looks like they're a spanish company is the ceo of free pick joaquin come to the show

27:18she'll make a super cut of every time addy's just trying to recruit people on twitter to come on the show no you know what i'm gonna look back at this in 10 years be like see i invited you you never came on and now we have sam altman sitting next to me yeah so yeah i think this is cool and i'm curious if more companies will follow suit and or like if this will shift with the credit model because yeah every finish everything every tool is like you gotta buy credits you gotta use the credits does it cost x amount of credits yeah also uh the spanish government is actually really healthy with tech subsidizations and stuff like they're genuinely now uh really involved in making sure a lot of the virtual production initiatives ai initiatives uh take off it highly benefits the the local population and economy so maybe free pick is kind of getting a little bit of money a little bit of help which is fine cool i mean i mean startup programs exist all over the place.

28:11Government sponsored startup programs. Most countries besides America. But yeah, there are. Yeah. Every time I talk to people in other countries and it's like, oh, I'm running my company. It's like, yeah, I got this grant from the government to help pay for stuff. And it's like, that's cool. But it's funny that my guess is the most number of startups happen in America without any of that. Well, we have the VC land. I mean, look, we also have SBA and we have some other programs, but it's easy to get a loan here. Easy to get investors here. But those are loans that you still have to pay back. I feel like in other countries there's like, grants grants it's just like here's money to go start no strings attached you don't have to pay us back yeah we're not expecting a return we just want you to make stuff well studio ulster is such a good example like a multi-million dollar vp stage in northern ireland it's funded by the uk government direct funding or like tax rebate type stuff no no no because also like we're big uh we're big on the tax rebates yeah i think i think it's direct funding okay yeah yeah there you go all right next up grayscale labs nano digital volumetric diffusion so we saw this video popping around yeah this was on linkedin a couple weeks ago yes or just i don't know i saw some of the clips i don't think i fully wrap my head around it and then now yeah this is really awesome it is a i think it was a just resolve so the first thing it's doing is a depth map estimation which ai models now are very good at doing that and then using the depth map it's able to determine where the light source is and then on top of that add uh diffusion not image diffusion but light diffusion that you see through smoke or through thick air and stuff the ridley scott filter yes now that i think about it gladiator was all this all of this alien you ever watch they're like behind i don't know it was a commentary behind the scene of alien and they would say that like they would have the the the fog machines on the set but like ridley would like not shoot until the fog perfectly like sat yeah they would have to wait to roll camera until like the fog just got like just right yeah i bet the pa's like uh mr scott we're ready to shoot no chat more smoke why do i sound like sean connery i don't know i don't know you need to work on Ridley Scott.

30:24He's British, right? I think so. Yeah. I was gonna say, I don't really know what he sounds like. I imagine Ridley Scott sounding like Sean Connery. Anyways, it is like that. It can add volumetric to feel like fog. Yeah. And it's a resolve plugin. What's lighter than fog? Not smog. What's the other? There's another word. Volumetric fog? No. Isn't there another word that's not fog? It's like, like atmospherics. Like atmospheric haze. Haze. Haze. Haze. That's what I'm thinking of. You can adjust the haze to fog level, but like the lighting works. They have demos of it sliding sliding around the direction of the light yeah i i guess it like to the naked eye this seems pretty legit like i i see it looks like they had a haze machine on set and like it was like thick lighting but if you asked like a greg fraser roger deakins like these primo grade uh cinematographers they're like no that looks nothing sure but also if you don't have a greg fraser budget and you're trying to shoot good looking stuff and you didn't have enough money to have like a huge spotlight or haze machine or the timing to like let the haze sit just right Good solution.

31:23Good solution. Yeah. And Haze machines, a lot of times, like as we know, in VP stages, not a thing because of fire sprinklers and all types of systems. You can't have a smoke machine in there. So this could be a good post solution. I think it is just a Resolve plugin and I don't know what the pricing is. It doesn't say. All right. Next up, Higgsfield Soul. So Higgsfield, we talked about a bit. They started off with doing sort of a pretty good looking pre-built camera movements and other kind of AI video generator model and tool. They've now launched still images and called Higgsfield soul. And these are sort of more trained specifically for in the fashion influencer space.

32:03So like highly realistic looking AI generated photos. I could also do, you could blend it with product images and kind of get these product shots. It looks pretty good to me. Like one of the first things I look at when I see people, AI generated people is skin. It looks like there's a lot of poor details makeup details uh imperfections like freckles and things like that so yeah it's definitely been trained on a lot of actual people yeah and trained on these style these different fashion slash advertising styles with these names so like yeah like and the core so just like a model like like russian looking model in various situations uh gorp core so outdoorsy photo shoots with arcteryx brand oh interesting so they even broke down the sub genre of influencer yeah because It says they have 50 curated presets of fashion-grade realism.

32:51Wow, that's amazing. Yeah, so not just an image model. It's an image model targeting very specific types of images, types of use cases. Tokyo Street style. So it's definitely scraping Instagram data. What are you talking about? Yeah. Y2K posters. Oh, my God. This is actually amazing. That looks really familiar to me. This is every movie poster from the 90s. Is that the Matrix phone in green? Somebody needs to buy that. That is like two Nokia brick phones taped together, right? Yes. That was my very first phone. Oh, yeah. The 8720 or whatever. The little brick thing that they also sold the case for, but now everyone's like, what?

33:28I still have mine. Next episode, I'm going to pull it out for you. All right, pull it out. I think mine is in a storage box somewhere. Yes, so is mine. So, yeah, another kind of new image model, but also interesting that they're targeting very specific use cases. So if you combine this with what we saw with HeyGen, the creative operating system, like now you're starting to fill in the pieces of what an influencer's tool set is, tool chest. I think we're going beyond influencer. It was like, this is like your brand and you need to look like you have a big influencer UGC campaign. And now you could spin it up.

33:59I'll try to find the links. Maybe we talk about this in a future episode. but I've seen people like showcase, you know, entire campaigns, like photo video campaigns for like a fictional brand. And they're like, Hey, it was like a fictional coffee brand. I just made up here are all the creative assets. It looks like a professional photo shoot did this. It looks like, you know, we got the girl pics hanging out in front of the coffee shop, holding up the drinks and none of it's real. It was all completely generated. So crazy. Yeah. How fast this is happening as an exercise? Should we try a denoised?

34:26Oh yeah, let's try it. Yeah. Let's try it. Yeah. yeah yeah which genre do you think we should go with y2k hell yeah no i'll figure so yeah let's try to like maybe maybe there could be like the denoist con and we have a complete fake photo campaign like as if we had like yeah like as if we had like a live stage recording but you know and people were there with their denoist swag but like none of it's real yeah i think our limited viewership will go down to zero after that guys we don't want to lose you guys limited real viewership but the ai viewership will go through the room in the latent space will be skyrocketing okay next up spark 3d yeah so spar spar 3d is a open white paper that a lot of other companies use to build 3d models in the past like generative 3d models yeah so from image to 3d model uh and i guess this is a level up from spar 3d so with the extra c construction 3d sparse representation yeah and construction for high resolution 3d shapes modeling yeah it just seems like overall the big takeaway is that everything is just a higher poly count high resolution a lot of detail yeah one of the big issues with like generative 3d is sometimes or a lot of times get like very blobby soft yeah looking shapes um and the demos are seen here extremely sharp yeah you get what i call melted wax melted candle yeah yeah that's good analogy this is this is awesome Yeah, yeah.

35:54It's a paper, you know, but I don't know when we'll see it in like an actual product, but I'm sure very soon. Good to keep on your radar. Yeah, I know that in AI companies, researchers are actively reading white papers like literally every day. And it's just a matter of them pulling the algorithms in the white paper into a model. Yeah. And then testing out that model in the real world. So I think within weeks, we're going to see Spark 3D beta products out. Yeah. All right. Next one. High Dream AI. Yeah. So, HiDream is a free generation app. It has a browser-based support. And a lot of people in the AI community like it because of high quality.

36:34So, it's a text image generator? Yeah. Text image is still the bread and butter of a lot of generation. Even in video generation, I think if you get the image right, the image rewrite, then you can feed that into video generation. It's a lot of first frame. You need that first frame. Yeah. And I think I talked about this before. ultimately it the workflows that we settle on won't be text to anything like we're not going to prompt for much longer we're only doing this now because this is this is a start right ultimately it's going to be you know generating a bunch of frames bunch of images and then having precise control over the video generation based on those images or frames and what's really neat about it is uh at the very top joey if you go to the top of the website you can just generate right then in there you don't need an account you can literally go yeah type something in and look how fast and responsive this is one one thousand two one thousand three one thousand so within three seconds you have not just one but several images different styles yeah so if you're out there and you just need to generate something on the fly this is not bad and yeah didn't ask for that is this you did not for that so yeah look you don't have to do a complex comfy ui workflow you don't have to sign up for credits if you're just this is their own model like they develop their own model this is their own model and i believe this is a chinese model okay if i'm not mistaken yeah pretty wild pretty good yeah some of these look pretty photorealistic and so and then you could is there an api to access it if you wanted to i saw there's a subscription thing on the bottom.

38:08Oh, the dev model is free to use. High Dream L1 full model requires credit. So the one I just played with was the lower model. It's not the high quality model. So it's even higher quality. Yeah, I'm telling you, this thing is a sleeper. Yeah. People are obsessed with ChatGPT image journey because it's easy, right? It's integrated into the app. We talk about MidJourney a lot. I mean, Runway is still pretty good for like with the references and they did release an update to their image reference the way they use it and it's gotten a little bit better. But ChatGPT is still I find the best for like just giving it a variety of images and being like, you know, put this person here on like in this location, rotate the camera 45 degrees so we get a different perspective.

38:47Yeah. ChatGPT has just been really good at that. I think because they have an internal prompt engineering step and they use their own LLM obviously for that. And I think that is the magic. Like everything being built by them. They just know what the user wants and then turns it into the prompt that the image generation wants. Like there's a middle step that we don't see. Yeah. Well, whatever they're doing, it works really well. It works well. And ChatGPT5 is about to drop. I saw some blurbs about that. Yeah. Yeah, we'll see how mind-blowing that is. All right, Dopal. Dopal from Google. So what is this?

39:20Yeah, so virtual try-ons is supposed to be a big money grab, a big commercial sector for generative AI. If you think about our interaction with the web, a big portion of that is fashion, is you picking your outfit, buying something online, whether it's Amazon, whether it's, you know, Lululema, Nike, Adidas, like all these big brands are now online brands. Physical stores are going away. So, one of the biggest problems with apparels is returns. Because everything is online based, you're also returning a ton more than you would in a physical store. You're not trying stuff on. And, I mean, these are billions of dollars that they have to worry about end of the year in returns and they have to take returns so to mitigate that even if they spend a good chunk of their money on virtual try-ons hopefully that shaves a lot of that billion billions of dollars a waste down and google knows this anthropic knows this everybody knows this so this is this is a hot as we keep coming back to you e-commerce is a very large market way larger than emily yeah yeah so google introduced doppel which is an app when i have it downloaded so here we're gonna play back you'll see some addy try-ons so yeah i'm a big fan of addy addy fit check should we post the clown outfit one maybe not so yeah here's one of me wearing a pink suit here's one of me wearing a gucci mains outfit from an nba game i like that one with the long socks all i did was i downloaded the app i uploaded a picture of myself and then i grabbed gucci main outfit from the web uploaded that and google doppel made not just this image of me but a moving video of me so i could see how the fabric falls on me and how to right now is the sizing accurate too like because like that that would be a big issue for return so you're like oh i want to see like how does this fit on me you know and and so the sizing of the actual item needs to figure out how does it actually match to that's the number one problem in real life it's like what do you look like when you wear the size small versus the size large yeah and like how much is it sagging yeah my issue too especially with like guys and shirts is like either with guys and shirts t-shirts you know it's a small medium large but every company cuts they cut them differently and so like i like blue lemon i can't wear the shirts because the sleeves like it's built like the hooks on my shoulders a lot of time you get the gun show dude no it's just the way they cut it here where it's like kind of it's just like the cut is just weird and it like picks up the whole shirt but it's like an issue i have with like you know just oh how do how do they try the different shirts but you know multiply that times yeah i don't think this is there yet i'll tell you yeah like i think we're just that beginning step where just putting the outfit on a person and then having generally fill in their back and so on step two is gonna be fabric physics fabric fit and stuff like that and step three i think will be highly photo real stuff like you can't tell that from like yeah and the fin the fin looks yeah yeah i mean and this is an app like you from google and you can kind of use it on photos of products in your camera roll but if this was being deployed by like an e-commerce site uh you know the issue too is just like what is the generation cost yeah for each of these yeah it'll definitely offset the returns at the end of the year like that's how much they're willing to spend this yeah yes like if i can offset even 10 of that billion dollars right that'll pay for all of this yeah what do you think the cost is now if you were to like build a product to like generate video generation is very expensive yeah you know you're talking one dollar video typically exactly from someone just yeah browsing around the site and maybe they buy maybe they don't yeah that's uh yeah that's one of the challenges with these now because of yeah i mean but to be fair the video generation models will get more efficient and perhaps because the stuff is running on a mobile phone you don't need the high resolution You can crop and make it vertical and save costs that way in the video generation.

43:21Yeah, yeah, that's true. Yeah, you don't need to have a full 16 by 9 widescreen image. And speaking of virtual try-ons, so AWS is no stranger to all of this. They have their own image generation model called Amazon Nova. Does Amazon sell stuff? Do they still have an e-commerce store? Are they interested in e-commerce? I think it would just roll into Amazon.com. Yeah, that was a joke. That was a joke. for a minute i thought how they launched the fashion label and you just completely separated aws from actual amazon from from their parent company they've won because that's exactly what they want you to do anyway sorry for my lame joke uh yeah so amazon nova canvas another virtual try on yeah so amazon nova is the image generation model and nova canvas is the virtual try on on demand so your brand can actually build on top of the aws infrastructure and have virtual try-ons right now.

44:16Right. And also I seen their demo video too, not just try-ons, but furniture, seeing furniture in your space. Quality-wise, I think it's still a bit easier. Yeah. Furniture is way easier. Anatomy and humans is probably one of the hardest problems. But look, it just still looks very AI-ish to me. Like the quality is not as good as the Google Doppel product. Yeah. But because this is ready to go and scale, this is a far better proposition for brands. Yeah. That's advantage of aws if you're already all your stuff's ready on the platform you just tap into uh bedrock and connect it to their other ai programs and yeah and most big brands already have aws integration like they have teams that are already building on top of aws so for them to turn this on would be way easier yeah again i'm so i'm curious what the what the cost is for like each time someone's trying out a new thing like how does that how does that cost compute chat gpt computes it really well so you can prompt it like hey i want a video 12 87 20 30 seconds long at you know 100 videos per minute how much is that going to cost on aws and chachi bd give you actual prices oh like getting a budget from chachi bd to figure out what yeah all right all right cool all right what you got last one that you sent over adobe firefly sound effects generator so yeah yeah this one so firefly not to be slept on is now including on top of video generation image generation a sound generation model where you can uh mimic something using your voice because we're really good at ad-libbing with our voice you could either text prompt what you want and then you could also drive it with your voice yes very cool for you know because like i've gone through sound effects libraries like thousands of sound effects and you're looking and it's like i mean they're pre-recorded they're canned and sometimes they don't always match up and you got to do a lot of massaging So with this, it's like, I need this, something that sounds like this, but let me be a Foley artist and just like into the mic, just kind of.

46:10So intuitive. Mimic what I want. So much control. I love this. You know, Adobe is always so focused on how creators actually use it. And they know that text to sound is not going to be the thing. So, you know, it's like I have a car scene, car going past the camera. And I want to do. Yeah. Time to the actual video. Yeah, exactly. and put that in there as your voice and then you'll get an actual car sound yeah and yeah and again this ties into like yes there are foley artists that do this but that is also like top tier like it is like a regular creator cannot get a foley artist no uh with their awesome room and millions of props and supplies yeah and i don't think a foley artist is interested in any of this because there's not high quality enough for like a dune or a movie of that quality generate a sound effect it's probably just be an mp3 file but uh for something you're making on the web yeah great solution super cool yeah i wonder how it does with like trailer kind of sound effects like um impacts and you know the risers christopher nolan yeah yeah yeah can you make the interstellar sound effect on it not interstellar uh inception yes yeah so that is our roundup and tangents for today links for everything we talked about as usual denoristpodcast.com and also in the show notes we'll have links to all of the apps and updates that we covered thank you for all of the reviews we've gotten.

47:29We haven't got one in a couple of weeks, actually. Yeah, so you can leave a review, please. Yeah, and we'll shout you out. All right, thanks, everyone. We'll catch you in the next episode.