In short
The episode covers two main threads: (1) the MPA’s MOU with ByteDance/Seedance (C-Dance) on IP protections for generating likenesses and (2) practical video pipeline upgrades like Topaz Hyperion 2.5 HDR upscaling, plus a discussion of IMAX 70mm depth-of-field and the film The Odyssey.
Guests
Addy and Joey (co-hosts). They discuss generative AI model behavior, legal/rights constraints, and video post workflows.
Key claims
MOUs are non-binding but signal progress; the biggest blocker for studios is legality, not model capability. C-Dance inference is done in China, raising data-transfer risk. ByteDance is adding face verification/registration and a permission path for known likenesses (e.g., actors like Tom Cruise/Brad Pitt), with blocking for unpermitted use. Studios may need centralized rights/permission management (possibly smart-contract-like).
Notable examples
Seedance 2/2.5 “rooftop fight” teaser; C-Dance 2.5 being more restrictive; Topaz Hyperion 2.5 converting 8-bit to HDR; IMAX 70mm shallower depth of field; Odyssey theater vignetting speculation.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOOverview of MPA and ByteDance Agreement
0:45 to 1:40
Discussion of the recent MOU between MPA and ByteDance regarding IP protections.
“Okay, let's just get one thing out of the way.”
Understanding C-Dance and Its Usage
1:40 to 3:40
Exploration of the implications of using C-Dance in professional pipelines and the associated legal hurdles.
“with C-Dance used in professional pipelines.”
Data and Rights Concerns in AI Models
3:40 to 6:10
Delve into the ethical and legal challenges when using likenesses of actors in AI-generated content.
“in final output or professional pipelines is something that we will talk about in the next story because that's come up a bit.”
Face Verification and Authentication
6:10 to 9:00
Discussion on the new face verification system for using likenesses in AI-generated outputs.
“If you upload something that has a face in it, you have to register that face.”
Future of AI Models and Rights Management
9:00 to 11:50
Speculations on the future landscape of rights management and the potential for a central system in the industry.
“and what can't go through and how to set that up.”
Exploring Sea Dance and Content Regulation
14:00 to 19:36
Discussion on the complexities of using likeness in generative models and the potential for regulation in the entertainment industry.
“grant permission, a system where you authenticate once.”
Topaz Hyperion 2.5: New Upscaling Technology
19:36 to 22:20
Introduction and discussion of Topaz's new model Hyperion 2.5 and its capabilities in video upscaling.
“And the crazy thing is, like a year or two ago, I was talking to some researchers in my last job, and they're like, yeah, I don't think we can do that.”
Integrating AI in Production Pipelines
22:20 to 25:54
Analysis of how AI technologies fit into existing video production workflows and their impact on quality.
Challenges in High Dynamic Range Content
25:54 to 28:01
Discussion on the challenges of maintaining quality in HDR content and the future of AI in video production.
“I mean, look, we looked at S4, the Bob Lazar story on Amazon Prime.”
Video Input Limitations and Metadata
28:01 to 28:54
Discussing the limitations of current video inputs and the potential for using metadata.
“But also when you can give it high-quality input and it understands the data happening here and, like, maintains that.”
Show all 15 chapters
Depth of Field in Filmmaking
28:55 to 31:06
Exploring the impact of sensor size on depth of field in film production.
IMAX vs D-Cinema: Cinematic Techniques
31:07 to 35:15
Comparing IMAX technology and D-Cinema methods in relation to depth of field and storytelling.
“Yeah, I mean, you could, you know, we could be split screen right now, but you could see it here if you're looking at the video because I'm shooting on Micro Four Thirds.”
The Impact of IMAX on Box Office Success
35:16 to 35:48
Analyzing how IMAX technology contributes to the box office performance of films.
“as well as Nolan's credibility and the story and Matt Damon.”
Production Challenges in Filmmaking
35:49 to 37:23
Discussing the difficulties faced during the production of films shot on large formats.
“Christopher Nolan who bans chairs on his sets.”
Viewing Experiences and Film Projection Issues
37:24 to 39:28
Examining the differences in home viewing versus theater experiences and projection quality.
“Yeah, I think Resolve has a built-in feature that can kind of help with that.”
Transcript
Automatic transcript. May contain errors.0:00Addy:Would you be mad if I saw the Odyssey just on my TV at home for the first time?
0:04Joey:Only if you like put your TV up to your face like this.
0:09Addy:No, I'm going to watch it on my Apple Vision Pro.
0:15Joey:Welcome back to Denoised. Addy. Joey, nice to see you. How's it going?
0:20Addy:Going all right. All is well in generative land. I've been generating in generative land. And yeah, we got a lot to catch up on. There's been so much happening over the last couple of weeks.
0:32Joey:There's been stuff happening. There's been a lot of new models, which I don't even think we'll cover because I don't even remember where we left off. But I feel like the biggest announcement in the last week or two has been an agreement or a MOU between MPA and ByteDance over IP protections and use in CDance.
0:51Addy:Okay, let's just get one thing out of the way. MOUs don't work as we have been seeing with the war.
0:57Joey:It is probably more promising than Straight Outta Four Boos. Okay. That's a low bar. Details are kind of fuzzy, but there's a deal reached between the Motion Picture Association and ByteDance. This spurned off with the first teaser of Sea Dance 2 or 2.5 with the Tom Cruise Brad Pitt rooftop fight thing that, like, everyone went crazy over. And then when Sea Dance 2 rolled out, there were, like, a bunch of restrictions on faces and what you could do with it. and then they sort of eased on that. But now there's an official deal or an official start to getting to an official deal over IP protection, likenesses, copyright protection, and how to use it responsibly with C-Dance used in professional pipelines.
1:43Addy:Okay. C-Dance is not being used in professional pipelines much. There's probably a massive ethical hurdle to overcome. And to my understanding, I could be wrong here, most studios are although playing with c dance are not really using it for any sort of real work like final output type stuff final or close to final or maybe image modifying because it is a video to video model you could push some camera footage through it and get something else on the other end that's modified i mean look if you're a you know not to nitpick but like if you're like a promise or a wonder or um what's britain studio's name asteria like if you're if you're like these type of native ai shops you probably have special uh legal pathways to get to final output and monetize on it but if you're a traditional big studio like uh like a disney or paramount or a netflix like i would imagine that path is much much harder if not impossible at the moment are
2:47Joey:Are you talking about from a technical limitation or a rights issue?
2:54Addy:Not technical. I think it's purely from a legality standpoint, at least from my understanding. CDAN's inference is done in China. So that's like a number one problem is you're just sending the data across international waters. Right. So like there's that. And then, yeah, the inference is done over there, and then it's coming back. And it's like a no starter for a lot of big studios. Having said that, it is so powerful. I could see why MPA has come to the table and tried to get an MAU going, because it's pretty dang good. I mean, we can't deny that.
3:34Joey:Yeah, I think there are two parts to that. I think the technical one and why, and when and if it'll be used in final output or professional pipelines is something that we will talk about in the next story because that's come up a bit. But yeah, the right thing, I know that's a concern of when your inputs include IP or likenesses of actors or famous people, what is happening to that, not just CDesk, but any model. Where is it going? What's happening to that data in between when you upload and then when you get it back?
4:10Addy:Yeah, no, I haven't even thought about like SAG protected likenesses like a Brad Pitt or Tom Cruise. Likenesses and performances, just like what's happening to the performance as well, because that's also protected. That's true. It is. Like, I wish I was more of an expert on the legal side of things on this stuff.
4:29Joey:It's all still getting sorted out. It's such a moving target.
4:32Addy:Right. Exactly. So, like, some of the things that I've heard just out and about is if C-Dense were to, let's say, be a U.S. soil-based inference. So, let's say ByteDense gives somebody the weights, like VPland, and VPland sets up a data center in Kansas, and then the inference is done in Kansas. The data doesn't leave U.S. soil. It goes from L.A. to Kansas.
4:56Joey:Right. Right. And like Minimax, even though that's a Chinese company that they have the open weights, you can take that model and load it on whatever data center you want. And it's staying it's staying.
5:05Addy:Yeah. So like getting to that point would then make it 50 percent easier. And then there's still the like it's trained on publicly available data. So like anybody that has their IP on there can potentially sue and go into a lawsuit like that's the other 50 percent. But at least the first 50%, which is like the non-starter part, is like there's no way U.S. companies will take the risk on Chinese inference given, like, I guess the U.S.-China relationship at the moment with all the trade, like, limitations and things like that.
5:41Joey:The other part that they're building out and like assuming, okay, that however they handle like the inputs and outputs and what data they retain and if that's all clear and okay with studios, the actual IP protections that they put in, the details are a little fuzzy. But from what I understand, just also like talking to other companies that are dealing with the API directly, there are new protections put in place where there's basically going to be more face verification. If you upload something that has a face in it, you have to register that face. Even if it's virtual and a made-up face, there's like a virtual avatar bucket and you just kind of register it there and like that's cool and then it processes.
6:25Joey:but then there's another route for known likenesses and faces um which this actually i mean it's a benefit because there's basically a path to authenticating people like someone can i think it's a phone or so there'll be some phone app or something kind of like we saw with Sora a little bit but a system in place where you can authenticate yourself and then give permission for other people to use your likeness with cdance and it's a good path because it's like what happens in these cases where like you do legitimately have permission to use an actor or actresses likeness in your project. This was always an issue before where you're just either completely locked out or blocked from like using it.
7:02Joey:And there wasn't really a good route to be like, hey, we like have permission. Can we like generate and use it on this stuff without getting blocked? So now there is a path where like, if you do have permission, you can get, you can get outputs, you know, using, I know it's definitely with faces. I don't know where this stands with IP and characters and stuff like that. But at least, you know, at least there's a path being carved out to like for this when you have permission and then also blocking it if just like someone wants to make a random Tom Cruise video.
7:31Addy:Yeah. I mean, you're absolutely right. Like we need a mechanism because how easy is it for a user to just go grab an image of Brad Pitt from the web and then start like a whole series off of that one image, right? Like it's that easy. You can build an entire persona out of it, different costumes, different time of day, heck, even different ages. You can make Benjamin Button right now if you really wanted to. Having said that, I went down a deep rabbit hole. That one day I texted you two tests, one with C-Dance, two, one with 2.5. And your response was so funny. You were like, what's the difference?
8:06Addy:What's better? I'm like, shit, Joey's right. I don't know the difference. Yeah, you're like, C-Dance 2.5, so good.
8:12Joey:I'm like, what did you do that you couldn't do before?
8:15Addy:And I was like, yeah, he's right, shit. And I need to come back with more ammo. So I went down a rabbit hole of like testing a bunch of things, both on 2 and 2.5. My general sort of observation was 2.5 was a lot more restrictive with just me being in there. It would throw up a lot more flags of like, this is a real person. Inference cannot continue. And that was all sort of done on FAL. So maybe it's FAL setting that up and not see dense. I don't know. But either way, just...
8:45Joey:I think it's also just like this keeps changing rapidly and the API providers are trying to keep up with what ByteDance is allowing. And it's a communication issue of what can go through and what can't go through and how to set that up.
9:01Addy:Yeah, my guess is when C-Dance 3 comes out, which will be, I don't know, tomorrow, it's going to be as restrictive as Sora was with the app. You remember how, like, unless I had your complete buy-in, I couldn't generate you at all.
9:17Joey:How restrictive Sora was after the first week of launch.
9:20Addy:Yeah, not the first few days, but, yeah, like a month in. It got really boring.
9:25Joey:Yeah, we'll probably see something like Sora, something that has permission issues, or permission, a system to grant permission.
9:34Addy:Overall, it's good. It's a move in the right direction. And that's like the MPA is so smart that they sort of are reading the reading the room, right? Like they already know that to go back to ByteDent and say, where did you get the data? That's like so three years ago. Like that argument doesn't hold up today. The models here, it's already doing inference. It's already across the globe at many different places. The win that can be had now is like, OK, so you can generate it. But can we have some level of control, some level of licensing, some level of money generating mechanism in play so that we're all happy here?
10:13Joey:Also, the focus is on the outputs, which, you know, we've talked about a bunch. And it's like the focus should be on the outputs and what you're doing with it and if that's infringing on stuff. The input argument, it's getting murkier.
10:25Addy:And it's kind of like the boat has sailed on that, I think.
10:29Joey:Yeah, the boat sailed. any of the cases that have kind of come through and had some decisions, you know, the anthropic stuff where it was like, well, you know, if you bought the books and trained on it, like then it falls under fair use. You know, that's one where it's like, yeah, the Pandora's box opened.
10:41Addy:Is that why they're buying the books, like the antique books,
10:45Joey:the stuff that we hear about? We didn't talk about the books. I mean, they're all buying books and scanning them and training on them. No, but apparently, I didn't look into like how true this was, but there was like an X thread that was explaining why they destroy the books. And it was because there is some weird copyright law where like they have to destroy them. They can't resell them in order to fall under. I don't remember what it was specifically, but there was some reason that they had to destroy the books in order to comply with like some copyright law. And it was very weird.
11:21Addy:Can't you just put it in like a seller and throw away the key? I mean, those are probably priceless books.
11:25Joey:That's the other thing too. It was the headlines were that they were rare books, but it was rare in the sense of word perfect 1995 user guide. Like, like. That's how my dad's house right now. Dreamweaver 2.0 had it in 24 hours. It was stuff like that. Look, I'm not trying to get on a whole anthropic defense case here, but it was this was a threat. and it was like a lot of this stuff and it made sense where it was like you know hundreds of thousands of books are destroyed every year and tossed out from publishers and elsewhere so it's not like they're the first to like destroy books it's not like no it's not like they're getting gutenberg's bible and are like we see on this thing all right shred it no it was like the random old computer guide stuff that's technically rare because they don't exist they're out of print But no one's knocking on, no one's collecting.
12:20Addy:But, like, look, I think any chance to paint AI in a bad light, the AI haters will take it, right? Yeah, I mean, I think there's an argument.
12:29Joey:It's like if you're scanning them, then why don't you put the information out at least? I mean, obviously, there's a competitive advantage to scanning them. But, like, at least when, you know, Google did this years ago when they were doing Google Books and they were scanning stuff. But they were at least putting them online.
12:41Addy:Oh, right, right, yeah.
Read the full transcript
12:42Joey:They got into heat with publishers, but the info was there if you wanted it. But yeah, it's sort of like a war chest of like a data war chest between Anthropic, OpenAI. And there was some story where someone put an AirTag in a book that they sold to see where it was going, and they tracked it to like an Amazon scanning facility. So they're scanning too. So, geez.
13:07Addy:Wow. I mean, that's that's the that's the side of the business that I think is is, quite frankly, pretty ugly. Like the data acquisition side going back to MPA. And by the way, they're no longer MPA. They have a building somewhere in L.A. and I drove by the other day and I was like, hey, that last day is missing. So I guess they're now they're now the MPA of the world. What do you think is the two-year forecast for something like this? Where do you see this play out in, let's say, 2028 or something like that? Two years, that's a long time, man.
13:44Joey:I think more models or more systems, maybe there's a standard, maybe there's some central rights management and permission management. Right now, it's sort of been each model or company's figuring that out on their own. maybe there is a central way to manage that, grant permission, a system where you authenticate once. Because also it's like, okay, if you were producing a film and you have permission from an actress, actor, to use their likeness in a model, this is just Sea Dance. Normally you're not just relying on one model, you're doing a bunch of image models and video models, depending what the needs are.
14:24Joey:So then you're gonna have to figure out how do we authenticate that with every platform or not? So, I mean, if there's a central system that can manage this and also track and just have, like, better visibility into what's happening, who's granting permission to what for what projects, that would be great. I don't know if that would actually happen.
14:44Addy:There seems to be a massive business opportunity with somebody that comes in and is able to enforce regulation onto Genov.
14:54Joey:I mean, it's tough to enforce. And then also it's like, well, what do you do with open way models or any other models that you can just run yourself?
15:01Addy:Go run, go run in yourself. I mean, like, who's going to really do that? And two years from now, two years from now, like the front frontier models, like the state of the art models will be so far ahead of open source models. I think that it'll just be a no brainer for studios to use something like this and just make sure the regulation is all in place.
15:24Joey:Yeah, smart contracts, something that, yeah, I think I said it a while ago, I think, I feel like some of the core elements in NFTs might come back again, just because they have the, I mean, a million dollars for JPEG was a stupid idea, but some of the foundations in NFTs and smart contracts.
15:42Addy:I just hope the NFT grows, don't come back. No, just the foundational elements of like, just the technology, like more blockchain
15:49Joey:And more intelligent contracts could come into play again and possibly help some of the stuff sort out. I think there's a big part of ease of use, too. And we talked about this with the Instagram thing a few episodes ago when Instagram added the default ability where you could take a celebrity's image and run it through the AI thing. And then after we recorded, they rolled that back and undid that. But we were like, it wasn't anything that you couldn't do if you screenshotted it and took it to another model and did it yourself. But the fact that it was like easy to use. They removed the friction.
16:26Joey:They removed the friction. Kind of opened up the floodgates. And there was another similar thing when we weren't recording. But like, did you see the Google Maps stuff? Where like Google Maps added Gemini and Nano Banana directly into Google Maps. where you could take like a Google Maps like a satellite photo or like the 3D Earth view. Oh, and then do a fly-through? No, not do a fly-through, like literally like change it. You know, nicely painted version of this was you could take a photo of like the Colosseum, like an aerial like 3D Google Earth shot of like the Roman Colosseum today and then make it look like it would have looked, you know, when it was in his heyday.
17:04Addy:I could use that right now for, I mean, I'm just thinking for productions that need to like ground their environments to real maps but then build upon the map let's say you want 1960s downtown LA like 1970s you kind of have the thank you Joey god that's embarrassing now that I look I'm not going to do that anymore yeah let's say 1970s New York, you know, Little Italy before it became a touristy spot. You do the Google Maps and then you put it through Nana Banana, generate enough keyframe for them to feed it into 3D model. And now you have a 3D model of 1970s Little Italy.
17:46Joey:I think you're overestimating how powerful this feature was. It was literally like you're taking a screenshot of Google Earth and just, you know, style transferred it to whatever you want. So it was something that you could already do today if you took a screenshot and put it in Gemini and just style transferred it. But because it was already built into maps and it was such an easy tool, and then people started abusing it and making images that look like real satellite images but have geopolitical issues like a port, like making a port look like it was bombed and it wasn't really bombed. All sorts of stuff that they then pulled that feature out.
18:29Joey:And again, it's like none of this was stuff that you couldn't already do if you just screenshotted it and knew how to like load up another model. But because it was so easy to do, it removed the friction and then it caused issues. And so the same thing with C-Dance where when it's so easy to do and if you could make a video of a celebrity without any friction, it's problematic. It seems like if you remove some of that ease.
18:52Addy:Yeah, so like it has Brad Pitt's Patreon QR code.
18:57Joey:you just like send them$2 and then you could use likeness for one generation.
19:08Addy:Honestly, it all comes down to money. The celebrities know that there is money to be made. MPA and the studios know there is money to be made when we have generative media just scale the ecosystem entirely. but then how do you figure all that out i think that that is the next phase of gen ai in hollywood is the next couple years we're going to see things that kind of work and fail and some things that really take off on those terms on the regulation and the monetization side all right next up topaz
19:49Joey:uh to pass got a new model hyperion 2.5 this is a new upres model that can take 8-bit video and turn it into hdr either prores 10-bit or uh exr 16-bit or h265 10-bit finally
20:06Addy:finally so i was waiting for that i was like do can we even make a model that can do that because isn't that really difficult? And the crazy thing is, like a year or two ago, I was talking to some researchers in my last job, and they're like, yeah, I don't think we can do that.
20:25Joey:And Topaz has done it. You get the training data. I haven't tried it yet. I want to try this. But this also joins a group. Like, they're not the first ones to do this. So this joins a group of other up-res models. SwitchX has their own up-res model, HDR model. Beeble. Yeah, Beeble, SwitchX, HDR. It's a dedicated model of Switch X that is literally just for uprising and adding more bit depth into any video. I call it upranging. Upranging, that is a good term. I've heard that one's good because it gives you slider controls over, it gives you some control over the uprise process. I don't know how much control Hyperion gives you.
21:05Joey:if it's just like take the video, it gives you the up-res version and you take what you get, or if you have some control over how you want the exposure to come back. LTX has an up-res model. That one I think it's take or leave it, an HDR model. And then who else? I think Luma has it. I know Luma will give you HDR outputs. I don't know if they have an up-res model specifically for like give it something, give it an input and you get an output.
21:31Addy:Well, isn't LumaRay 314, or I think it's 32 now, it's video to video, right? So technically you can give it your video.
21:40Joey:I mean, video to video, but are you going to get the same video you gave in out?
21:45Addy:Correct. Yeah, yeah, right. See, one thing Topaz has always done well and to the best of their ability is, like, respect the original input.
21:55Joey:Oh, yeah, that's great.
21:56Addy:To an extent. Yeah.
21:57Joey:Yeah, I mean, all of their generative upscale models have been excellent, especially the latest version of Starlight 2.6. Yeah, I just used that a couple of days ago.
22:07Addy:It's fantastic. I mean, going from like a, and, you know, I always use it on like generative AI stuff, but I'm sure you can use it on restoring old things as well. 1080p to 4K is like really convincingly good. it's maybe not as good as like native camera footage that shot in you know 6k and then down sample to 4k because it is still missing some textual detail but it's such a step up from 1080p
22:34Joey:yeah this goes to the question that i brought up before so when we're talking about kind of professional pipelines and also this like hybrid pipeline the issue currently is okay let's say you're shooting with actors, you're using like a pretty decent cinema camera or some, you know, any decent camera, you've got that good footage. All of the current pipelines that are some form of hybrid, uh, that are easy ish is, uh, you know, you're taking your footage and then you're running it through like people switch X or you're running it through C dance too, or, you know, any of these other, um, Omni, I need these like video to video models that are really good at editing slash modifying video you get your output out now but now you're basically just took your like high quality cinema footage just like squashed squished it and you just got like a web output you know potentially then maybe you could run it through a model like this and then get the range back it still seems kind of counterintuitive to me because you're still like losing the camera data and then making
23:39Addy:it up again you're missing one key step so let's say the original camera footage is um you sitting at a coffee shop having a coffee and you want to modify that to you sitting in a bar having a beer right but like you are the same it's just everything around you is changing and the
23:59Joey:drink that you're holding is changing so you go ahead and do your generative ai version of
24:07Addy:the thing that you want and then when you bring it back in um you just cut a hole where joey is and retain the original joey and then just swap the background and the drink with the
24:21Joey:general yeah so there is this route where you are creating layers and then you're recompositing that back in and then you still have your camera original footage that gets murkier if you're trying to like really push that source footage and like relight it um yes which yes you can kind of do with uh biebel switch light the original model on the source footage but then basically you're kind of getting into like a hybrid traditional vfx pipeline yes exactly and
24:50Addy:the advantage so with the topaz stuff like let's or the biebel stuff right like you get to the same resolution or near same resolution of your camera and almost the same dynamic range as your camera footage and now you're doing an apples to apples compositing and that will yield you much better
25:08Joey:results yeah you're saying if you take the outputs and bring it back in and composite the this stuff back together with the real camera the more the more pixels from the real footage you
25:18Addy:retain the better your result will be in my opinion because camera pixels are still superior to generate AI pixels.
25:26Joey:And it's original data. You're just like, you're not, you know, you put it into this. It's doing a best guess at what it thinks it is. It might work. And look, I think a lot of these pipelines will be fine, more than adequate for web stuff. If you want to do a web series, this is probably all you need. Maybe streaming series. Obviously, it depends on quality control requirements of streamers. But, like, it could potentially work there. You know, but then when you start talking about 4K theatrical distribution.
25:53Addy:Yeah, no. No, theatrical is like the last hill AI is going to climb. I think for now it is. I mean, look, we looked at S4, the Bob Lazar story on Amazon Prime. Yeah. Like that is where the technology is at today. And that is like a timestamp of where we are in mid-2026. And like if you look at the quality of the generations and, you know, I'm sure they didn't do an HDR upscale or uprange there. But it kind of serves the purpose.
26:22Joey:I mean, as usual, it depends on what point of time they did this stuff.
26:28Addy:Exactly.
26:29Joey:They probably did an up-res at the very least. I don't know if it was a full HDR up-res, if that was even out at the time that they finished this.
26:36Addy:Right. Yeah, I wonder what the Amazon distribution window is. How far ahead do you have to submit everything to them for them to then push it onto their network and then give it to all the prime users?
26:50Joey:Yeah.
26:50Addy:Or just like if it's like three months ago, then three months in generative AI timeline is a long time. Million years. It's been 84 years.
27:00Joey:I'm curious to try this. I'm curious about the pipelines for this. I've probably talked about this before. We need to have a shootout. We need an HDR shootout at this point to see what we get. And to compare it with like original stuff, you know, I'm thinking like you're shooting on 6K, Blackmagic, B-Raw. Even at a pretty higher compression, you're still going to get a lot of latitude in this.
27:19Addy:Yeah, generate a candle next to a street lamp with AI and then shoot the same thing with camera and then put the two side by side, publish it 10-bit on YouTube and then look at it on your 10-bit TV. That's one way. So that's the other thing is like viewing and actually doing QA on high dynamic range content is difficult. I think most of our displays are now 10-bit, but the authoring of this is 12-bit and beyond.
27:49Joey:I feel like the alternate would be when the model gets to a point where they will either give you like 10-bit 4K output, which I think Switch X2 is going to have once it comes out, you know, or 16-bit, or does they give you that high enough output? But also when you can give it high-quality input and it understands the data happening here and, like, maintains that.
28:10Addy:Ooh, I like that. I like that. Which, yeah, that's... Could eventually happen. You better give your guy, Manji, a call and let him have that idea. Yeah, that's a great idea. Yeah, free idea. You guys can take it.
28:23Joey:Maybe read the metadata off the camera
28:27Addy:and just retain that, whatever that came in as.
28:31Joey:Yeah, I mean, everything's limited right now because it's like all the video inputs are, you got to give it a H.264 file. Like nothing's going to take a ProRes input. EXR, just drop or, you know, camera raw files.
28:43Addy:How about that?
28:45Joey:Yeah, if it knew what to do with that. eventually probably eventually by eventually it's probably 12 months out exactly yeah next year
28:53Addy:last one you sent this to me the odyssey so i've been watching a lot of imax stuff lately as you can imagine my algorithm's full of it and i came across this video by max adams and i haven't seen the odyssey so this is kind of a spoiler for me and anybody else who hasn't seen it but supposedly the entire movie has like shallower than shallow depth of field and it's primarily an artifact of that giant massive IMAX 70 millimeter film in the sensor yeah duh damn it I thought I hit upon gold
29:29Joey:I was like oh really um when I saw it I think you've probably seen the clips there's like the um the uh intense kind of monologue scene from um uh Anne Hathaway and she you know she's like monologuing where she's like getting very traumatic and but her head's kind of moving like you know just kind of like tucking in and out like this and i was watching i was thinking like the key and peel meme with the the with the sweating where like the focus puller had to have been like like really trying to hold focus on that because there were times when like they would just lose focus because they were just moving in and out but yeah it's not a sensor obviously it's a massive
30:07Addy:70 mil is huge and so yeah your your your your depth of field is super shallow from my understanding it's it's just just physics at play because of the way uh light rays converge on a sort of a convex lens um the the range at which uh you're in focus is significantly impacted by your sensor size so like a small sensor has way more play so you get a lot more depth of field but the larger and larger the sensor gets the depth of field becomes smaller and smaller and like what max showed was like how much bigger a 70 millimeter piece of film is than like a super 35 sensor which we're all used to it's like four or five times bigger it's massive so you can i don't know if it's a linear relationship but it'd be like a four or five times reduction in your depth of field versus like a Panasonic Lumix or like a Blackmagic Ursa or something like that.
31:07Addy:That's Super 35.
31:08Joey:Yeah, I mean, you could, you know, we could be split screen right now, but you could see it here if you're looking at the video because I'm shooting on Micro Four Thirds. So I have pretty crappy depth of field in the sense that like, this is pretty in focus behind me. Whereas, what do you want?
31:25Addy:I'm shooting as Super 35 here. Right, and even your bookshelf right behind you
31:29Joey:is like a little bit soft. Things are already out of focus. Yeah. Yeah. So that's a good, that's a very, you can see this example literally right next to us.
31:36Addy:And like that has to do with aperture as well, but then.
31:39Joey:I think we're probably both at 2.8. Yeah, aperture, focal distance, but yeah, sensor size or acquisition size, big factor. I mean, even the 65 mil comparison to the IMAX. It's so massive. A huge difference. You think like, oh, five millimeters more, what difference could that be? But it's like huge. And the way the IMAX is also captured in their proprietary aspect ratio.
32:03Addy:100%. And this got me to think, like, I don't think IMAX 70mm will ever really take off. It's a big R &D experiment for IMAX.
32:13Joey:You know, even just to shoot the dialogue scenes, they still, like, they had the big blimp system. So, yeah, it was quieter and they could record audio. But did you see the mirror setup system that they had? Where, like, the camera's still so big in order to get, like, over-the-shoulder shots they still had to have.
32:26Addy:Yeah, you need a beam splitter to mount it vertically and then go shoot sideways.
32:33Joey:No, it's basically like they had a mirror mounted by the lens and then another mirror so that the actors could have an eyeline. Oh. Because the camera's just too physically big to put it over their shoulder. But if you wanted to kind of have those more intimate shots, they had to have a mirror system so they could still perform to each other and work around this camera system that's still huge and massive.
32:56Addy:The other thing to note is a lot of the big directors, I'm just thinking out loud here, it's like Spielberg is known for having a look that has a larger depth of field. Sometimes he'll have things in the background in focus as well as things in the foreground. If you go back and look at Jurassic Park, a lot of shots are crystal clear in every depth of field. And I think, are you better off with a giant D-cinema sensor like an RE65 and cropped to 4.3 to mimic that IMAX look than actually shooting on IMAX 70.
33:34Joey:I think we'll find that out with Dune 3, Greg, because I don't think that was shot on actual IMAX cameras. I think that was shot on...
33:42Addy:Yeah, I think it's all D-cinema and then converted LIMAX. That was LIMAX. Yeah, and it's like you'll still get the big IMAX presentation
33:51Joey:you'll still get the, you know, for better or worse, proprietary aspect ratio that can only be displayed in 70 mil at the 70 millimeter prints. Are you really going to notice or feel a difference whether it was actually shot on 70 mil film or if it was shot with, you know, D-Cinema 65 mil AR cameras?
34:11Addy:Let me ask you a question. Do you think if Nolan shot this conventionally, The Honesty would have had an equally big box office and just success that it has now?
34:23Joey:No, I don't think so. I think a lot of this is tied to this spectacle and also this, not rare, what's it called?
34:32Addy:Like the technology of IMAX.
34:35Joey:No, the technology, but also the scarcity of being able to see...
34:41Addy:Of the 41 movies around the world and so on.
34:44Joey:Yeah, see the film in the way that it was shot in the special way and that you could only really see it. in this handful of screens. Obviously those 41 screens aren't the reason it went over a billion dollar box office. I think also it's a big epic story. It had a big epic format. It just all like cap lent itself to having this big moment. But I don't think it would have hit as hard if it was just like a regular 35 bill or.
35:12Addy:I agree. I think you're right. The lore of IMAX had a lot to do with the success as well as Nolan's credibility and the story and Matt Damon. It all combined into this massive success.
35:25Joey:Yeah, it's a Greek mythology. You literally have these images of these Sisyphean-looking images of Hoyt Van Hoyt carrying the big IMAX up the hills and stuff, and they're like, we had to literally carry all the gear up the hills because we couldn't get any access up there. And so just the making of the film and the film itself, I feel like, are tied into this perfect synergy.
35:48Addy:What about those Boston Dynamic robots? They couldn't do it.
35:53Joey:Christopher Nolan who bans chairs on his sets. He's going to have a robot carry his stuff up for him.
35:59Addy:Yeah, that's so interesting. I guess we'll wait for Dune 3. But again, it all goes back to how do you create the right type of conditions so that moviegoers go back into the theaters? This is a verification, if you will, of creating those right type of conditions. right and it paid off but on the other end i just heard from a friend the other day that avengers doomsday which comes out on december 18th this year some of the theaters are already
36:29Joey:sold out well their issue was dune already booked imax all the imax theaters and so they weren't going to change their date so they made up the fake um infinity vision format okay which is just a rebrand of the other premium large format theaters like um i think like dolby and cinemark cinex whatever they call it yeah xd so dune has imax for like three weeks and then doomsday can have it i think something like that but anyways the issue was they both have the same release date and dune had imax screens first and disney wasn't changing the date so they made up a
37:09Addy:new thing can you imagine how pissed the production designer on the odyssey would be let's say she walks into like i'm assuming i don't even know who it is like they spent three months like taking a side of a mountain in like some european city to turn it into a beautiful troy or whatever and then nolan shoots it in the end everything's out of focus except for matt
37:35Joey:goddammit's face god damn it we spent the wide the wide shots were all in focus okay okay you do you think they used ai to get them back into focus yeah i'm absolutely sure yeah yeah yeah yeah that's the topaz the topaz do you can you handle uh 16k upscale i know i think there is
37:58Addy:some refocus models now that will bring things back into focus.
38:05Joey:Yeah, I think Resolve has a built-in feature that can kind of help with that.
38:08Addy:That's right. So Nolan used that then.
38:10Joey:The one issue I had with the Odyssey, and I was trying to see it again in 70 mil, but everything sold out or doesn't exist anymore. But I saw it at the LA Live, the new theater they built in downtown LA, not the Universal one. there was like crazy vignetting and I don't know if that was the projection was bad or
38:31Addy:just a bad lens on the projector maybe
38:33Joey:yeah I don't know if it's a projector issue or if like the film was vignetted so if you saw 70mm elsewhere let me know if it was really like vignetted if that was a thing or if that was just a bad projection I think it might have been a bad projection
38:44Addy:the projector is trying to be artsy he's like let me put this lens on
38:50Joey:it was like noticeably I wish I took a photo I'm like, you know, I'm anti photos in theaters. But in this case, I'm like, I should have taken a photo for reference. So I'm curious about that. If anyone else saw it, let me know if there was vignetting.
39:02Addy:Okay. Would you be mad if I saw the Odyssey just on my TV at home for the first time? Like I do with every film.
39:11Joey:Only if you like put your TV up to your face like this.
39:15Addy:No, I'm going to watch it on my Apple Vision Pro. Then I get the...
39:19Joey:Okay, that'd be cool. That'd be cool. Right. I'd be curious if they're going to release the actual iMacs ratio on digital. Because that's another issue, too. They never release the full thing. It's always like crop. So it would be good if they release the full thing, Apple Vision Pro. That'd be a pretty good experience.
39:40Addy:Okay.
39:41Joey:All right.
39:43Addy:Enough nerd talk about iMacs.
39:45Joey:Yeah. Hope you guys hung in there.
39:46Addy:Thanks for being patient with us.
39:48Joey:Thanks for everything we talked about at denoisedpodcast.com.
39:51Addy:We are on Apple Podcasts, Spotify Podcasts, and YouTube on the two podcast platforms. If you haven't left us a comment yet, please do so. We would really appreciate it. Thanks, everyone. We'll catch you in the next episode.
From the publisher
The MPA and ByteDance reached a deal on Seedance IP and likeness protections — Addy and Joey break down what it actually means for studios and why Chinese-based inference remains a blocker. Topaz releases Hyperion 2.5, converting 8-bit video to HDR. Pl...
