Tilly Norwood Signing, Runway's Lionsgate Issues, AI VFX Wins

30 Sep 2025 · 42 min

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

Denoised Podcast Episode Summary: Tilly Norwood Signing, Runway's Lionsgate Issues, AI VFX Wins

Episode Overview In this episode of the *Denoised* podcast, hosted by Addy Ghani and Joey Daoud, the duo analyzes three significant developments at the intersection of AI and filmmaking:

  1. The potential signing of AI actress Tilly Norwood by a talent agency.
  2. Technical challenges faced by the Runway-Lionsgate partnership.
  3. How VFX professionals are utilizing AI alongside traditional compositing tools.

Key Discussions

  1. Tilly Norwood and AI Influencers
  2. Introduction to Tilly Norwood:
  3. An AI character created by Particle Six, developed in the UK.
  4. Potentially being signed by a talent agency, which has caused industry buzz.
  5. Hosts note that the name "Tilly Norwood" sounds gimmicky and question the authenticity of AI influencers.
  • Comparison with Existing AI Characters:
  • Discusses comparisons to other synthetic influencers like Lil Miquela and Code Miko.
  • While Tilly Norwood presents a more photorealistic appearance, the hosts emphasize the gimmick nature and potential short-lived success of AI influencers.
  • They also debate whether fully synthetic characters can build genuine connections with audiences.
  • Industry Reactions:
  • The hosts mention concerns about AI actresses replacing human actors and recognize the need for compelling storytelling to create engagement.
  1. Runway-Lionsgate Partnership Issues
  2. Overview of the Deal:
  3. Runway's partnership with Lionsgate aimed to harness AI to enhance production stages across Lionsgate's film library.
  • Technical Challenges:
  • The partnership has faced unforeseen complications including:
  • Limited capabilities due to insufficient data from Lionsgate's diverse catalog.
  • Copyright concerns regarding the use of actors' likenesses and existing IP.
  • High expectations from Lionsgate executives about the AI's capacity to produce significant content transformations.
  • Insights and Speculation:
  • Discussion on the feasibility of using AI for large-scale projects vs. targeted applications, such as previs, storyboarding, and VFX enhancements.
  • Emphasis on the need for a realistic approach tailored to the industry’s current capabilities.
  1. AI Integration in VFX
  2. Case Study of Freddy Chavez:
  3. Freddy Chavez, an accomplished VFX supervisor, demonstrates the integration of AI tools with traditional VFX workflows.
  4. Highlights include:
  5. Using tools like WAN 2.2 and Nuke to enhance and replace visual effects in older footage (e.g., from *Twilight*).
  6. The successful application of AI to generate and animate realistic characters and objects.
  • Impact on Traditional VFX:
  • The hosts emphasize that blending AI tools with established VFX practices can lead to faster, more efficient workflows.
  • They acknowledge the importance of compositing tools like Nuke in achieving high-quality results that AI alone may not provide.

Key Takeaways

  • AI in Filmmaking:
  • The prevalence of synthetic characters raises ethical and creative questions about the future of talent in the industry.
  • AI is a tool that can enhance traditional filmmaking processes, but its limitations must be recognized.
  • Industry Caution:
  • The excitement around AI’s potential in filmmaking should be balanced with realistic expectations regarding its capabilities and the depth of storytelling necessary for audience engagement.
  • Future of AI in VFX:
  • There is significant promise in the integration of AI within traditional VFX pipelines, suggesting a transformative shift in how visual effects are produced.

Conclusion In this episode, Addy Ghani and Joey Daoud bring to light the burgeoning role of AI in the film industry, discussing both the potential and the pitfalls. As the landscape of media and entertainment continues to evolve, *Denoised* offers insights that keep listeners informed and ahead of the curve in this rapidly changing environment.

---

*Listen to the full episode on [Denoised Podcast](https://denoisedpodcast.com) for an in-depth exploration of these topics.*

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:00All right, welcome back to the noise. We're going to talk about AI actresses getting signed by talent agencies. We're going to talk about where the runway Lionsgate deal might have fallen apart with some tech issues. And a breakdown of real VFX artists using AI combined with Nuke. Yes, sir. More AI, more problems. Let's get into it. More models, more problems.

0:24Welcome back, Addy. Welcome back, Joey. Good to virtually see you again. I'm not sure if I like this or not because I didn't want to drive on the 405 for everyone that just listens to this it makes zero sense because you just hear voices you don't notice that there's any difference whether we're in person or recording this virtually on Riverside I think the audience can tell I think there's a little bit more lag between our deliveries our interactions and also we're not drinking coffee together I just have my paper cup of coffee right now so yeah there you go cheers man cheers yeah all right so our first story people are kind of flipping out over an ai actress tilly norwood potentially getting signed by a talent agency if you're a long-time listener denoised ai actresses and actors being signed by talent agencies actually not really news because we've talked about this for a while and made a bunch of comparisons to this before but yeah let's talk about the story because also i've never heard of this ai actress slash influencer tilly norwood and why are the names always so gimmicky i don't know well she is a british uh british and developed by a british company so you know i think it's got to sound pretty british yeah it just sounds like a made-up name it is i mean maybe that's the point it is a made-up name

1:44tilly norwood yeah all right so this came out there was a panel deadline had some ai panel or something in zurich and the developer behind the company that developed tilly norwood uh elene vandervilden revealed that they were in talks to potentially have signed the actress the ai actress with talent agency and also i noticed some photo on this panel is uh verena that's running luma ai's uh hollywood incubator studio which we are still figuring out what's going to pop out of that yeah i see some snippets of that lab on linkedin john finger posts here and there seems like there's a bunch of people in there working away yeah incubator yeah so that'd be cool okay but uh back to tilly norwood so it's a little bit confusing because uh in the article mentions that uh van de velden behind the company that developed her it named one company as a xcoia and then it named another company as particle six so i'm not quite sure i think particle six is the company the production studio yeah so i went on the particle six website and it looks like they check all the sort of relevant and modern ai boxes so they have the AI avatars.

2:53They have AI VFX. They have AI drama reconstruction. So we talked about all of these use cases for AI before. Looks like a two-person team from what I could tell. So it's Vander Veldin, as well as another gentleman. I can't get his name from the website. Right. Or probably two head people and then a team of AI developers, artists, people. And it looks like they're hiring four people. Yeah. they're gonna they're hiring an ai creative producer ai video motion magician an ai production intern as well as a researcher developer type yeah i mean everything on this list is like kind of all the buckets that ai sort of fits in right now cinematic nature montage showcase vfx drama reconstruction which is what jenny ai has been doing sort of as their specialty commercials yeah it's like all the the entire the entire gambit the influencer thing is just here let's go to her Instagram.

3:46And I know I'm saying her as it's an AI. All right. Instagram got about 16 ,000 followers, which feels low for me for someone for a product that would sign with a talent agency. Potentially, I'm assuming the talent agency deals would be like brand deals, brand promotions, stuff like that. You know, I think probably calling it an AI actress is like the triggering part. And if they call it AI influencer, that might be less triggering. Yeah, exactly. You're encroaching on SAG territory, the precious quote-unquote celebrity territory with somebody that's fully synthetic. And of course, this is going to cause some hair to rise on people's arms and shoulders.

4:27What are your thoughts on this? I mean, we've talked about this before, and it feels like it's in the same bucket as CG influencers like Code Miko. Not Code Miko, that's a different... Little Mikayla. Code Miko is also a synthetic influencer, yeah. Powered by a real person. It is powered by a real person who, yeah, could be because a CG avatar that is powered by an actual human being who plays the character. Little Mikayla, very popular CG, non-existent character, but it's built by a team of people, right, in real and traditional. Yeah, I worked on a couple of Little Mikayla episodes back in the day.

5:01It was a fascinating look at how a digital character was run. And I'm guessing, fast forward to today, it's probably the same thing. There was an entire creative and technical team behind this one synthetic character. The team was like 10 plus people. And it was really interesting because they obviously have an idea of what she should be, what her life should be. Little Michaela was bisexual and had both girlfriend and boyfriend problems. And all of these things that were fabricated to make her really interesting and really appealing to a broad demographic. And then on top of that, there was real products placed within a lot of the campaigning.

5:44So, you know, I think she had a Samsung campaign for a while that paid real money from Samsung to do. Yeah, because people see the posts. And I mean, that's basically what the companies were paying for. It's like, do people see the posts? Does it get engagement? Cool. Like, yeah, product placement. Yeah. Little Michaela was done, like I would say, at the time back in 2019, 2020, right before the pandemic. at the highest level of Unreal Engine quality that was achievable at the time, using some of the best Unreal artists at the time were on that project. It didn't look photorealistic, right? Like, Little Mikayla, you knew was a...

6:20That was part of it. Like, you knew what you were getting into, and it was cool. Yeah, so here's Little Mikayla's feed. 2.3 million followers. Right. I mean, I will say this does actually look more realistic, but you can tell it's like, it has that metahuman feel. For sure. The character is in the Uncanny Valley, but what they used to do was they would composite her into real backgrounds and real photos of real people. Like in this case, they would take a photo of this real guy holding nothing, right? And then they would just composite little Michaela into his back. But you're aware, yeah, it's not like they're trying to pass her off as a real person.

6:58Look, also, they're not trying to pass Tilly Norwood off as a real person. that says that is an AI creation, but it also instantly off the bat, this looks photorealistic. This looks like a real person. Not to me, no. Because you know right away. If I just showed you like a stack of images of just like young women posing on with their coffee. And you didn't know. The one bad giveaway for me with synthetic avatars is there's two symmetric. Like the face? Yeah, like 99 % of human beings on the planet are asymmetric. so like our right eye is a little higher than our left eye and all those subtle cues give us away as to you know that we're real you know and synthetic generations just tend to be too perfect in this case you can see here like but look i mean i'm saying they're like they're not they're making this as a character it's like a cg character that they're using ai to power it they're like this look they're make they're making this as photorealistic as is possible with the technology today is that like it's good enough like i'm saying i'm talking about the intent i'm not talking about how does it look and how does it look because you like identify this stuff i'm talking about the intent sure the intent is to make it photoreal i think it's indistinguishable from reality yeah like if you're scrolling and you like didn't stop to look and read the stuff you would think this is like a girl posing with her coffee generic cup of starbucks it would like most likely to 99 people would not stand out to them but something was weird about this right we talked about the ai influencer stuff and the actor actor stuff where where are you thinking about this this this is definitely obviously like synthetic looking yeah so here's my opinion on this i think it's it's very gimmicky and it may have a short-term win with a couple of maybe big product campaigns or perhaps even a placement in an ai movie but in the long run this is not something that will stick and we'll forget about it a year or two from now and i think what will be more interesting is taking a real person and turning that real person into many different likenesses.

8:59For example, if you take you, if you take Joey, I would really want to see a 15-year-old Joey and a 65-year-old Joey doing two different, completely different things. I don't want to say that. Just taking an example. I think it's far more interesting to attach a synthetic avatar to a real person and tie it back to a real person than to have fully completely synthetic avatar creation. Yeah, if it's part of their thing. I mean, I'm also thinking of, you know, they've got those hologram boxes. I forgot what the product, what the company name is, but you know, the big box that creates like a hologram lifelike version of like a real actor, and then it has some AI programming in it.

9:37So you could talk and ask questions to them and they sort of respond based on the context of your question, if they have enough material to generate that. So, you know, know, it sounds like what you're saying is sort of in that realm of real person. And it's like, oh, maybe they can be in more places than once or replicate themselves to kind of do more things than they could physically attend or be at because of scheduling or just feasible issues. Like you can't have a 15 year old version of an actor that is 40. I mean, if you're looking at everything that you're showing on the Instagram page here, like everything here seems so uninteresting and bland and generic like there is nothing in here that makes me want to go back and watch this again even the way she looks she looks very generic you know like you can't pick her apart in a lineup of a bunch of ai avatars yeah i mean also like what is her backstory what is the story because i'm not saying like you couldn't pull this off it's like we obviously as humans fall in love with fictional characters all the time of just like getting engrossed in their story and root for them so like i'm not saying it's not possible to create a synthetic character if you develop like a whole story around them or some background or something around them that sort of is another version of storytelling and you're just doing it through social media with synthetic characters versus making a short film or a movie or a tv series i don't see that with this i mean maybe maybe there's more there that i haven't dug into but that could be something that could potentially work in the synthetic route of story and world building with synthetic characters yeah i think that's where you and i agree to disagree i i i don't see it that way i even if you're generating a fully synthetic film like what cave on the kid did echo hunter in that you have uh brock i forget his last name but you have a real actor who's used yeah and it was approved by i think that is really interesting because you instantly attach that notion of brock and who that actor is to the performance that's being displayed on screen and you just have this natural connection to it versus looking at fully synthetic backgrounds fully synthetic actors everything's synthetic you're just like what am i even looking at yeah i'm not saying i love it but i'm saying i think there's an audience for it if you have the story and you have the connection and you build up the connection to that character through old school traditional storytelling techniques and not just like what you're saying with this hype with this reel with a bunch of just random shots of this ai character in a bunch of random scenes where it's just like what's the point of this like I don't have any fictional connection to this character who's just doing a genre.

12:15Yeah. One thing I did maybe think about what you're talking about with the sort of like flash in the pan aspect about this is part of this and the signing and building up these AI actors or talent. But part of it feels like the 2025 version of Bored Apes, where that was like NPC, Hype Reel, Flash in the Pan, you know, tons of money moving around. And then, you know, what are the Bored Apes? People that spent a couple hundred thousand dollars on them. Now they're worth a couple bucks, maybe. Is this the sort of AI version of like Bored Apes build up the character, build it up, try to get some money inside?

12:53And then... Absolutely. I think there's so much investor money out there. And if you even built a fraction of an audience as big as a real influencer, I mean, the investors are just going to be all over it. What I'll say about what your comment was earlier about, you know, if you can create a real connection to a synthetic avatar, that I think would be a really interesting experiment. It is yet to be seen. And maybe I'll be proven wrong and you'll be proven right. But if somebody can figure out how to make authentic, genuine, mass level connection to somebody fully synthetic with an audience that's massive, then we're really talking about a whole new genre of entertainment and a new window into this world.

13:38Yeah. And speaking of genre, that was their response because they did post a response to the initial story coming out and then a lot of anger and stuff over the creation of the what they're calling it now an AI character in this posted response. uh quote she is not a replacement for a human being but a creative work a piece of art i see ai not as a replacement for people but as a new tool a new paintbrush just as animation puppetry or cgi open fresh possibilities without taking away from live acting ai offers another way to imagine and build stories so yeah they're sort of saying it's a its own genre i think that's what i said before i think calling it a ai actor was a bad label and i think ai character synthetic character or AI influencer is more fitting with like what this is trying to do.

14:23Yeah. I mean, I'm not sure if I see the point that this is not replacing a person. To me, it feels like it is. And if you take a up and coming actress from this town, there's a lot of them here in LA and you put them in front of a camera, you put them in front of, you know, know, whatever set environment that you're building synthetically, or you sort of take the real person's likeness and then you cast this entire AI avatar on this real person. I think that'll go much further along in believability and adoption. Yeah. I mean, to the point of saying, oh, is it replacing a real person? Sure. I mean, for this, I see it's like, I mean, look, it's a social media platform.

15:08It's a level playing field. it's more just like building the audience like everyone's in that same game if you're trying to build an audience and so if you're a person or if you're i mean if you're trying to build an audience around your real self if you're trying to build an audience around a character you create it's all in the same in the same playing field so i don't really see this taking away a job from a real person it's this is they're building out of a brand around something and you know people build brands around themselves and it's just more like who is more successful at building at building the brand i guess your point what happens when there is enough of tilly nor like when there is a hundred tilly norwoods that's why i think it's just going to get so diluted and that's where i think the board apes comparison comes in like yeah sure i mean they're not the first to create an ai influencer character and there's just going to be like it's just getting easier and easier to do that and it's just going to be so diluted that either you're going to have to go back to the basics and like really build a brand and a story and you know something that's a that attaches you to whatever this character is besides just like generating a pretty face or or you know it's just going to get buried with everything else and to what we've talked when we talked about this like a few weeks ago with other ai characters that's why i still think if you are a you know real person like i i think the biggest the biggest buffer you're going to have in this sort of ai future is building the brand and like being yourself and building the brand and authenticity around yourself because as big as ai or cg influencers get they can never do like a real in-person event you can never like actually meet them in person sure you can make all the chat bots and virtual experiences you want but there's like they will just never exist whereas you know if you're a real person you're building up that authenticity like we are real people like people can see us they see us at events we have like i feel like that's the buffer in the future to kind of rise up above like the ai noise that's going to be super easy to for anyone to generate yeah all of all of the above that you said in addition to that a real person has genuine reactions to current events or somebody else's content or somebody else's opinion and an ai can only mimic that reaction not really have an authentic one a real person can have some emotional up days some down days and those things create for lack of better term art right and like as as human beings like those are the things that we also resonate with and connect with with other you know creators so again like there's so like 90 of what makes a social personality quote-unquote real is missing from what i'm seeing here yeah yeah i get yeah i get what you're saying with that all right i think enough tilly for now right you got any more tilly anything else tilly norwood hope to never see you okay okay maybe she's gonna she's gonna hot you she's gonna hot your dreams now goodbye tilly norwood i will see another hundred of you in the next episode i think uh every time you generate something now it's just gonna be tilly norwood appearing it's gonna it's gonna haunt your latent space all right next up this one came out like a week ago but we didn't really get a chance to talk about it and we've we've been texting thoughts about this a lot we both have thoughts about these yeah all right so the rap had a big story on the runway lionsgate deal which was announced last year and they sort of were the lionsgate was sort of the first studio to have announced the partnership you know with runway big ai company and the gist of it was like runway was going to train on lionsgate's extensive film library make custom models they were going to use this for i don't know it sounded like every stage of the production process from like story board to brevis to possibly like generating trailers all sorts of stuff and we haven't seen anything come out of that and uh the rap had this kind of deep dive into all of the tech issues that have happened yeah everything that i'm gonna say about this is it's gonna be purely speculative obviously we don't know what's happening behind the closed doors at lionsgate right aside from whatever was reported in this yeah and also we don't know what products or what r &d projects that runway is throwing at the efforts internally and from my guess is you know what's probably happening is that Lionsgate has you know very precious IP perhaps John Wick as an example so if you take the John Wick world and that's a beautifully created world that has a very specific style to it the characters look a certain way and there's like this entire film noir look to it I guess what they were trying to do and what they're trying to do internally is maybe recreate some of that, maybe do some shot replacement for the next John Wick installment or the TV show or what have you.

19:51I think that's too detailed from what they were trying to do. What do you think they're trying to do? Well, here, let's cover the article first, then we kind of jump into that. Sounds good. From the article, over the last 12 months, the deal has encountered unforeseen complications from the limited capabilities that come from using just Runway's AI model to copyright concerns over Lionsgate's own library and the potential ancillary rights of actors. The reality is that utilizing just a single custom model powered by the limited Lionsgate catalog isn't enough to create those kinds of large-scale projects.

20:21It's not that there was anything wrong with Runway's model, but the data set wouldn't be sufficient for the ambitious projects that they were shooting for. Lionsgate vice chairman Michael Burns, last month he bragged to New York Magazine's Vulture that he could use AI to remake one of its action franchises, an allusion to john wick into a pg-13 anime three hours later i'll have the movie right okay so that was one of the targets they're going for that's not gonna happen take take an ip existing property we have and then re-spin it into uh spider-man into the verse or something you know high art animation masterpiece no okay that that yeah obviously that didn't happen the first thing when this deal was announced the thing that always stuck out to me was lions gate does not like i can't say lions gate has a um like a visual language that is like like i can't say oh there's the lions gate look yeah like lions gate produces movies there's a whole bunch of movies they've produced they're all look different yeah so when it was like oh we'll train it on their whole catalog of stuff it was like like aside from john wick what you're saying it's like there's what other franchise in there there's no like lions gate look it's not like yeah it's not like lions gate is not pixar right yeah it's not yeah even those they all have different kind of looks and stuff yeah and then so just training on the movie would be like yeah that's not a lot of data to train on and the variants maybe if they can source all of the material maybe you'd have more stuff of like not just the final product of the film but like every element of the vfx passes all of like all of the raw footage of the outtakes and all that stuff maybe you could get more i'd be curious if they that was the plan to train it on all of that data too i remember a long time ago when we talked about that AI training company that was like brokering deals with YouTube creators.

22:08But the thing they were interested in was like the YouTube creators entire raw media library, not just their outputs, but like all of their outtakes because that data was extremely useful for all of the AI companies to train on. So yeah, I'm curious about that. So yeah, a couple of things going on here from the Lionsgate executive side, I think the expectations are just way too sky high, You know, being able to generate a PG-13 version of John Wick just from ingesting it into a runway engine. Like, that seems ridiculous. Like, we're not even close to that. I mean, that seems like the same gist of all of these other executives, like, a year or two ago, when it was like, AI is going to revolutionize our entire workflow, and we're just going to, like, switch everything to AI.

22:48And then, like, now, a year later, it's like, ugh. Yeah, we couldn't do as much as we thought we could. So it's really interesting that the entire Lionsgate library is not enough for quote-unquote training the model. What I think is happening here, and I don't have information on how the system architecture is underneath the runway video generation model, but on an image generation side, the foundational models take billions and billions of images to train. And once it's trained, you can then attach a LoRa or something in the size of, you know, 20 to 30, maybe 50 images to then pull that foundation model to generate in a direction that you want.

23:30And video generation is the same. For like 1, 2.1, 1, 2.2, you could build custom tuned LoRa's using, I don't know, 50, 100 images. and it'll do more or less what you train the images on, right? So if you take screenshots from the John Wick film and if you have enough, let's say 100 images or so, you would more or less be able to generate the environments and perhaps even the actors' costumes in that likeness. I'm not going to say it's going to be usable or production ready, but it will certainly get you there. So perhaps there is a limitation on the customization and how much runway model could be fine-tuned.

24:18And maybe there's a lot of pain there to get the correct output that you want. And the second thing which you nailed is the Lionsgate catalog is so varied. It's just like they've made so many types of movies and so many different genres, so many different types of looks that to train on the entire Lionsgate library is not really going to get you anything specific. Yeah, I made sure it's good, high quality data for just a general model. But yeah, it's not like Pixar. It's not like Studio Ghibli. There's not like a Lionsgate look. There's no look. Yeah. I think the only advantage is perhaps Lionsgate has access to the raw files from the camera or the EXR files from VFX.

24:59And if you combine those things, you can train an upscaler model to have EXR output, 16-bit output, and it's kind of what Luma has done with Ray 3. So there is advantages to having really high-quality data. But as far as custom tuning a video generation model to output a certain style of movie, if I had James Cameron's Titanic, that's a three-hour movie, and if I was able to train on that, I'm pretty sure I could get a lot of the details accurate when I generate Titanic shots. I think also the example you're giving where it's like, yeah, you could train a LoRa with like 1.2.2. The issue that they call that was like, yeah, it's also limiting to just only be able to use runway.

25:46And like kind of anyone that's in the AI space, you're using every tool available because all these models sort of do different tasks better than others or make things better than others. And so just being sort of tied down to runway, even with being able to have their support to train a custom model is still limiting to like what you're able to do with and not have the full tool set available to you. Totally. And I think the approach here is like coming from two different angles, right? Like if you look at what Netflix did with the Eternut, where they did a shot insertion where something that would have been VFX would have been really heavy and complicated, and they just used AI for that shot, and they got it done, right?

26:28But it wasn't at a TV show level or a movie level, like they weren't trying to generate the whole thing in AI, they were just nitpicking and taking a couple of things here, replacing it with AI. if Lionsgate takes that approach it's a much easier ramp up it's a much easier adoption curve I think into getting the filmmakers to start to use AI yeah I think it's one click vision of like we take our movie and then we could turn it into anime we could turn it into some like tv show episodic in a foreign language in like three hours is pie in the sky at least for you know a while at least for now yeah yeah I think it will be for a long time because uh you're not not only having to generate a very convincing storyline that resonates back to the original world, but then there's so much gap filling you have to do in building this environment, this world, this style in detail, you know, put the characters in there.

27:22And if they're synthetic avatars, then you have a whole set of problems to try to make them look realistic and sound realistic. And you just open up a camera in wormholes that you don't know how to fix. Yeah. But yeah, I think what you're saying, and we'll actually show this in our next story, using these tools for specific parts of the process, whether that's previs, whether that's storyboarding, whether that's helping with VFX shots, rather than this large, encompassing approach where you're like, yeah, we're just going to make the whole movie with AI. That's not going to work. But using AI to help speed up specific processes in the traditional filmmaking workflow is where there will be more instant gains to be had.

28:01Absolutely. And just to kind of go back to Runway's capabilities, I mean, they're probably the leading provider of AI-powered VFX solutions. I mean, I can't think of anybody better in this space where, you know, if you take a product like Aleph, like who's really doing the stuff that Aleph can do? Maybe it's not fully movie quality ready, but it certainly is very, very powerful and is kind of paving the way for VFX that's AI-powered. Yeah. But I mean, look, also it is a black box and you're kind of just like, yeah, tell it what to do and it sort of does the thing, but you don't really have that.

28:34I mean, I'm curious what the deal looks like when you have, you know, enterprise support like Lionsgate has and like what additional features or controls they unlock. But I think of the only other kind of thing I've seen that does sort of what Olive does, you know, is the Big Muppets workflow that we covered a few weeks ago where he sort of built that one 2.1 comfy UI workflow where he takes an existing video, can mask out the person, keep the person completely consistent and then change everything else around them. But that is a complete Comfy UI workflow using open source tools. But the bigger thing is, like, it's built in Comfy UI.

29:07You can see every step of the process of how it's doing it. And you can control, you can change, you can fine tune it. You have, like, that highest level node-based pipeline that professional studios and VFX artists are used to having that level of control. Dude, Lionsgate should have just hired Mick. I know, right? I mean, maybe it's not the technology. Maybe it's the people. Perhaps it's a different way to think about it. Or yes, maybe it's the organization too, because it's like, you know, I'm just, we don't know that much, but it's like Lionsgate is a studio office. Like, I don't, I doubt there are any actual like VFX people on staff for Lionsgate.

29:43Like if they make a movie, it's like John Wick. I doubt it. They're going to another VFX shop that is going to do the VFX work for them, or probably a couple VFX shops. So like every movie is a completely different assemblage of teams and people and companies. Yeah. Yeah. Yeah, like Lionsgate has always been a silo. Yeah, I mean, every movie is like a little mini startup company that like a bunch of pieces come together and a bunch of different contracts and stuff, which also, you know, that's the contract thing was another issue of just like dissecting the rights of every single movie. And because every contract is unique, every movie is unique.

30:14And the other roadblock was just figuring out, do they have the rights to ingest some of these movies to train in the AI models? And that's going to be another issue as well with training on existing IP. Also, in addition to what you said, I'll even take it a step back further, which is that with the exception of perhaps Disney and Netflix, which are fully vertically integrated, right? They have an execution arm, they have a technology arm, and they could go all the way down to making the thing and then distributing it. With the exception of those two studios, I would say every studio, like a Lionsgate, like an A24, is more or less a financial institution.

30:57And what I mean by that is you just got a bunch of executives and a bunch of people that make big money decisions, right? They'll give the John Wick filmmaker$100 million to go make John Wick 4 or whatever. And then it's up to that filmmaker to figure everything out. Now, those folks that are making these big high-level decisions with money are not really going to understand the nuances of what is and is not possible with AI. So I think there is a fundamental difference of those two worlds. Yeah. I mean, maybe Lionsgate has a tech arm that offers support. you know i know i think like there's some similar stuff even with the vertically integrated companies like a netflix where if even if they bring on a you know they produce a a movie with an outside production company they sort of have a tech division where it's like hey we can provide support and help you with problems and stuff because we've like you know see this happening across a bunch of different films that we're producing and so we could help you however you need help with but yeah i mean to some extent amazon as well right amazon has the technologies in in-house and stuff yeah yeah so it's like i don't know if they were trying to do something similar with that and it's like here's this sort of wide-ranging support for all the projects that are happening but so yeah what you're saying lionsgate is an office building that is producing movies like getting the financing and funding and distributing them and stuff but uh yeah they're bringing out their every movie is a completely different mesh of teams and people and other companies so i'm just kind of curious what the initial vision was and then like how this is gonna play out in the future like i don't see why this wouldn't work for pre-vis or storyboarding or concept designs or poster mock-ups I don't see why this wouldn't work for that There's a lot of opportunity for a generative AI company to build real value and real efficiency within a movie-making pipeline.

32:39Yeah, an existing pipeline. Something that already is there and how you can fit into that and save time and money. Well, I mean best of luck to Runway. I'm a big fan and I hope that they have a few wins at Lionsgate and that's all you really need to get the steam rolling. Yeah, I'm curious I think it probably just hit the roadblocks from the initial vision, but I don't think this is a fail. I think there will still be good things that come out of this. I think it's a big misunderstanding, if anything. Yeah, I think this is the same. This is the film industry thing in the bucket of the CEO that goes, you know, we're going to replace our entire tech support staff with chatbots because they can handle everything.

33:15And then realizing, like, they're not quite there yet. Yeah, it's the Duolingo thing. Yeah, yeah, exactly. The Duolingo of Emi. Yes. All right, last story. And this one kind of goes to what we've just been talking about, blending traditional VFX pipelines with AI. So we saw this ex post by Freddy Chavez almost, and I looked him up on IMDb, a very accomplished VFX supervisor, I think based out of Montreal, if I'm not mistaken, worked on movies that you'll recognize, such as The Meg, Annihilation, Deadpool, Chappie, American Sniper, Sin City, A Dame to Kill For, and The Mandalorian. So there you go.

34:01This is somebody who is deep in the world of tier one filmmaking, understands the quality that's required, and is also really familiar with sounds like all of the AI tools we have on hand today. So what happens when you give somebody with this level of skill and this eye for quality, the tools that we have. Take a look. So basically what he did was he showed comparisons with some original footage and some footage that was already had traditional VFX, like some older shots from Twilight, and then redid them using modern AI tools. His tech stack is WAN 2.2 Animate in combination with Nano Banana, Beeble, and Nuke for VFX enhancements, replacements, and full head de-aging.

34:45And we can see some of these comparisons, side-by-side tests of some original footage, like this Nick Cage concept of Superman. This was the Twilight shot, where it was just the actor and then replaced with the wolf. Yeah, there's three examples. And, you know, each of these three examples is what I would call very classic VFX problems to solve, right? So the first one is relighting and sort of head rotation and maintaining lightness, which is the Nicolas Cage Superman example that we're showing you here. So it looks like he did a second pass using Beeble to relight Nicolas Cage and perhaps maybe even generate him from scratch and then composite it all back together in Nuke.

35:28And it's a really tricky shot because his head and his body is rotating, so you can't just have like a plastered face. You actually have to have spatial awareness of that move. and then the second shot this is from one of the twilights where you know um you're essentially have a stand-in that gets composited out and then replaced with a cg animal a wolf and it looks like one 2.2 animate was used here to create the wolf and then all of that was composited with nuke yeah so what do you think the workflow would be so you think it would be track the body with 1.2.2 animate? So I think he used Nano Banana to generate the wolf.

36:10So you have your reference wolf image. And then you give that to 1.2.2 animate and you input the motion of the mocap actor here that you see with a gray suit on. That's your input to 1.2.2. And then your output is going to be that wolf in that same motion cycle. And in Nuke, you're probably feathering and blending a lot of like that fur looks really difficult to blend and composite in so uh obviously this guy's really good at nuke so he's doing a lot of heavy lifting here that we're not seeing and you think he would have used wand 2.2 and sort of just to create the wolf on like either green background or on something that he could composite out and then layer it on top of yeah you would have to maintain in just like any other composite uh you have to maintain the same camera uh lens and the same camera perspective for both plates so i'm guessing in um nano banana perhaps there is this entire frame in there just to kind of decipher what the cinematography language is and then replicate that in one 2.2 so when you blend those two things together it looks like it's being shot from the same camera yeah also i mean just with her hand touching the fur that also looks really good too and like blends right in yeah yeah exactly it's really tricky but the fact that no computer graphics was used per se like this wolf is not a digital wolf that was built in blender and the fur was done in next gen and it's none of that it's completely synthetic that in that it was just generated in one 2.2 yeah here actually i just noticed in his original in his behind the scenes thing he has where he completely erased the actor and then yeah brought in yeah And then, yeah, this Paul McCartney de-aging.

37:58The Paul McCartney de-aging is nuts. It looks so good. I mean, looking at this stuff 10 years ago, our minds would have been blown. And now it's just like a post on X, you know? Yeah. Yeah, I remember when it was such a big deal when they did it in like Forrest Gump and like inserted him in archival footage. Exactly. Tom Hanks. And yeah, now it's like, I'll just run it on my computer. Well, people still talk about like the Irishman where De Niro and Pacino were DH'd. Yeah. And that was like the three camera rig. It's a massive thing. They had like an infrared camera set up and a traditional camera set up.

Read the full transcript

38:32And it was, it was like a whole thing. And now you're able to just do it in comfy more or less. What do you think? What do you think was that you think like giving old images of Paul McCartney and Nano Banana to replace the face to have that driving image and then animating. kind of similar i think you were talking about before animating it with wand and compositing with nuke yeah i mean this is a he picked a good example because paul mccartney's hands and his torso when he's older match that of when he's younger they look at the hands dh too the hands look a little bit dh as well but the yeah it's a it's a one 2.2 input output example right you give it the old paul mccartney motions as an input and then with that banana you give it a young mccartney as the reference and then output is this and i think the the magic here is the lighting matches really well he's blending in with the couch and the shadows are cast correctly um his face is lacking a little bit of detail uh if you look at the old i wonder if that's a a resolution issue too because i'm wondering what he's outputting wand 2.2 at because like sort of i think it's kind of is in that 720 range to have some usability i mean i think you could crank it up to 1080, but it might take forever and you would need a lot of some beefy machines.

39:50So I'm curious what these, because also I'm looking at this on X, which is already compressed. So I'm curious what the outputs were being done at. My guess is he grabbed the video from YouTube, perhaps. So it's not going to be high quality to begin with. So, you know, input is output and you're not going to have a higher quality output per se. But if the average person looks at this, they'll think it's real, right? Like, we're nitpicking here, and we're trying to find flaws. But really, like, this is well done, and this is acceptable for… No, this looks great. Talk show level, for sure. Yeah, and I think this goes directly to what you were just talking about when we were talking about the last story, where it's like, this is the future of taking traditional VFX pipelines and using AI to help enhance it, speed it up, make it better.

40:41and do stuff like this with that one person just kind of messing around in their free time, making something look this good. Yeah. And I don't want to take any of the attention away from, I think, the bread and butter of what is happening here, which is Nuke. So much heavy lifting is being done on the compositing side, whether that's the edging, the feathering, the blending, the highlight control, the shadow control, skin tone matching, and all the stuff that he's probably doing under the hood. I don't even know. I'm not a Nuke artist. He clearly is. And with AI tools and with traditional VFX tools like Nuke, those two things combined, I think, gives a way to a completely new type of workflow.

41:24One that is potentially far less expensive than traditional VFX. Alright. Good place to wrap it up. Let us know what you thought about this episode in the comments and links for everything we talked about as usual in the show notes or over at denoisedpodcast.com. And just going back to our last video on YouTube. Thank you all for the comments. Sora Not Sora added again read 4109 as well as two new commenters, Oberdofus and Doug Creation Studio. We thank you. Alright, thanks everyone. We'll catch you in the next episode.

From the publisher

Addy and Joey analyze three stories at the intersection of AI and filmmaking: the potential talent agency signing of AI actress Tilly Norwood, the technical challenges in the Runway-Lionsgate partnership, and how VFX professionals are combining AI with traditional compositing tools.


--

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

More from Denoised

All 101 episodes
Tilly Norwood Signing, Runway's Lionsgate Issues, AI VFX WinsDenoised · 42 min
Listen in VO