GenAI: The Creative Revolution

Okay, thank you all for coming. My name is Martin L. Blanc and I'm the CXO of Magnific. So we were previously called Freepik, maybe you know that name, but last month we changed the name to Magnific, luckily. So, and what I do in Magnific is I'm leading two different teams. One team is called Magnific Studios and what they do is that it's a team of 30 creatives and they help usually enterprise size customers with adopting Genitive AI into their processes. And it's very much companies, you know, big production companies in the US, the biggest agencies in the world. And the other team I'm working on is called Academy, where we take all these learnings that we get from these customers and then we package that into courses and videos and tutorials and so on. So my job is very much sitting in the middle of, you know, learning how the professionals are doing different things and then transferring that into learnings and tutorials and helpful material. Just a quick summary. So, yeah, it's a very quick introduction to Magnific. It's, I would call it like an aggregator. So it's a platform where you can go on and then you have basically all the latest AI models. So whenever there's a new model coming out from a different provider, we test it, we check if it's good enough and then put it into the platform. So you can go there and sign up for that and then use basically every new AI model that's coming out. My intention today is not to sell at all. Basically what I want to... Help you with is like understand where is generative AI now, specifically for video. I'm going to show a whole bunch of practical examples because I think practical examples is a really good way for you to sort of like understand where we're at by seeing it yourself. And then I'm also going to try to reflect a bit on what's going to happen. So if you are managing a project now or you're going to decide where to put a budget, you might have some better options and better knowledge. So my talk is going to be about the new capabilities in AI video, specifically also about the impact on visual effects. I'm going to reflect a bit on the creative process. So I'm also a creator myself, so I use it every day basically. And I have a background also in producing music, doing web design, all kinds of things. So I've been going through the creative process professionally my whole career, and now I'm doing that with generative AI. So I'm trying to reflect on the creative process. And then I have a bullet called staying in control. I know a lot of people, especially if you're working in marketing, you want your own product, or you want something specific. And a lot of people think that generative AI is like rolling a dice. But I will show you that that is not actually the case. And then also just a few notes on where we're going in the future. So one thing that is really interesting in generative AI is that the technology doesn't improve in a linear way. It's very much like there's a lot of design. the world before we had ChatGPT and the world afterwards. If you look at, there's a site that is called Stack Overflow. If anybody's a developer here, they know Stack Overflow. If you look at their traffic, the day that ChatGPT launched, they started going downhill. It wasn't like gradually. It was like that day, November. You can see it in the chart. And this is also happening in Genentech AI here, where we have video models coming out. Last year, for example, Vue 3 was the first model that could produce audio and video at the same time. We had NanoBanana Pro. How many people don't know NanoBanana Pro? Have you heard about it? So NanoBanana Pro is Google's image model. And what it does is basically, it can sort of like, whatever you do in Photoshop, can basically ask NanoBanana Pro to do that. So it's a very, very powerful image model. And then we have new models coming from, especially from China, Kling and Seedance. And then lately, OpenAI came out with a new image model called DeepDance. And I'm going to demo some of these models today and talk a bit about what they're good at. When we focus on video, I think this chart really illustrates really well what has been happening the past few years. A few years ago, there was a viral video where we had Will Smith eating spaghetti and everything was messed up. That's not very far back into the past. We are now in a stage where many of these AI videos with people, they feel a little bit creepy because we are solving in this kind of uncanny valley. But we have fixed the five fingers on the hands. We're now down to maybe there's a little bit of details in the movement. Maybe the character is not blinking in a realistic way, but we're in a very quick progression into this area that is beyond the uncanny valley. And I would argue that within maybe a year or two, we are past that uncanny valley and we can actually create AI generated videos. Where as a human, you can't really tell whether it's AI generated or not. There's a new breed of video models and this is really what is I think powering this whole right now, the quality boost that we're seeing and it's solving a few different problems. There's three principles to these video models that are really important. One is that it's multimodal and multimodal models means that they can take like an input audio. Audio. Or a video or images. So you can actually give it, let's say a voice recording of yourself, and then you can ask a character to speak and say those words, or you can give it a video and then say, make the cartoon character move like I just recorded myself with an iPhone. So you can give it many different types of media and then it will transfer that into the final video. And then it also has what we call multi-shot. Multi-shot means that instead of trying to generate one little clip and the next little clip, I'm going to still clip and you prompt everything. You basically see, say, talk about a whole scene. So the difference between a shot and a scene is basically during a scene, it can be like 20 seconds. You have the camera moving, you have a closeup shot and so on. And the new video models can basically create scenes and not only shots. And then everything is reference based, which is a different concept than having start and end frames. Let me just illustrate what I mean by that. Um, so let's say we have. This is our character and we want to create a video with him. He has on his back, he has the number 23. Um, if I was using the old version of GMT-i models for video, I could basically control that, you know, that video model using a start frame and end frame. And then the video model will try to figure out what is in the middle of those two. But let's say we have this guy and he's walking towards the camera and then he's walking into a bakery. At no point can I give the video model the information that he has a number on the back of the t-shirt. So when I'm generating the video, basically it does this. So it goes from one frame to another frame and the signal from each of these frames is either decreasing or increasing during the time. We will probably see the back of that guy in the middle of the video, but the video model don't know anything about what the guy looks on the back. So it will look like this. He will walk past you and then to the bakery. And there's no number on the back. So it's like this kind of approach is basically fundamentally flawed. You will never get, you know, professional and good quality videos because there's so many things that you can't do with this approach. When you work with references instead, you have a set of images or video, any material, and every frame in the generated video will be impacted by this material. Let me show you how that looks. So instead, what you do is that you create. For example, it's just an image, but you can create like a character sheet where you say, here's the guy from the front, the side, the back. You give it all the details, basically like from all angles. You also give it the reference images of the locations. And instead of working with this kind of start and end frame, you're now affecting every part of the video. So now the video looks like this instead. So as soon as the guy turns the back to the camera, you can see the number on the back and he disappears in the end. So just the fact that it's now reference based is now elevating really what you can do. And this is the other part I was talking about. Single shot prompting is basically you get all the blue clips and you try to combine it. It's really hard. It takes a lot of editing. It's very hard getting continuity. The new way of working is basically you're prompting for the pink elements here where you're building multiple shots in one generation. Let me just show you how that can be combined. So I made this example where I wanted a like a yeti style character. And the tool you're seeing here is called Spaces. It's a little like note based tool. Maybe if you work with video production, you are very familiar with note based tools. But you basically have, you build your characters, you build your props, location and everything. And then you can basically pull and say like, okay, we want the data to flow into my video generator from there. What I do is also, this is done with dbt2. dbt2 is really good at understanding text and it can also create text, bright text. So I'm actually generating, for example, here a storyboard and saying, okay, I want my yeti to return to his cave and then he finds a flower. And then it's splitting out into a storyboard. And then in my tool, I then put all my reference images, character. You can actually in the character sheets, you can actually. Also create little notes about his personality and so on. But here I'm creating, sending all these images into dbt2 and asking it to create a storyboard and then finally into a generated video. So the prompt for the video basically looks like this. In the top, I asked for multi shot scene. So that's what I'm asking for. Not just single clip, but create the whole scene. And then I'm saying at, and then I'm referring to this specific image is my troll and this other image is my location. And then I have. The flower, which is the prop. And then I'm also telling it, this is the storyboard. And then with all that information, the video model is then able to create the following video in basically one shot. What is this? Decrease. The. I think that there's from a, like a creative point of view, it's really interesting how. Like editing the video and directing and art direction becomes a very much integrated process. It's no longer very split. It's like, actually you have to think about how the troll should move in, in when you're writing the prompt. For example, there's one part in the problem. I was not sure if you could see it, but there's one point where I'm asking. It to say, focus first on the flower and then change the focus to the troll. Like you can actually ask for those kinds of details. Um, in the prom. So it's very much, um, process where all these steps are collapsing into one. Let me show you a bit of like, what's also happening with visual effects. So this was a animation. Um, I'm going to show you like, this is real footage and let's say you are a filmmaker and you went to Switzerland or Norway, and then you film this. And then. Back home in Denmark, you find out, oh man, this is actually the wrong time of day. And then you go to your visual effects studio and say like, how much is here that to change? Uh, and that can be quite expensive. And then also, for example, here, one thing that's actually quite difficult to do is let's say we regret that we had that red color on the dress. We want to actually have a yellow dress instead. Right. How do we fix that? So the process is basically, um, if you're using generative AI. You extract a few frames of that video, and then you're using one of the modern image models to edit it. So for example, here, uh, I'm using nano banana pro, and then I'm asking you to change the time of day, change the color of the dress. And then I'm taking that image along with that original videos. And then I'm asking the video model to say, you know, take the, all the colors from that image, but all the movement from the video. So combining those two things. And then you are basically done. You're getting this. So, and I think like this is done with a video model called, uh, Kling, Kling 3 in, and it's a native in, in 4k. So you don't even need upscaling. Basically, if you can create a video in 4k, that is also the resolution on, on Amazon prime and so on. We're going to see like a pro, um, a project that is actually using this technology. I think it's really interesting that some of. I think it's really interesting that some of these models, for example, the Kling model I'm talking about here, it's available to consumers. This is just, these are the same models that they're using now in Hollywood and the big productions. It's basically in the past, the technologies of the professionals and consumers was very different. Now you can see professionals and consumers using the same technology. Let me show you another example. This is a very typical problem in, in Gen2vi that you create something, it looks beautiful and you want to keep it. But there's like one detail that is just wrong. Right? So in this case, uh, I was doing this kind of like animation style video. And I really liked this shot where this robot is like running like a big dog. And then there's a, this little guy controlling it. Um, but what you notice is that there's a copy of the little guy. So he's actually there twice. So he's like, you know, disappearing and appearing and so on. If you just see the video, you don't really notice it, but if you have a video where you have too many problems like that, it kind of like starts feeling, feeling off. So basically I want to get rid of this. Right. And how do you fix something like this? Um, what you can do with these, uh, video models is also, you can also use it to edit a video. So I can ask and say, remove the boy from the foreground. I just listed some of the video models that can now do it. They are all from 2026. So Gemini Omni is Google's new model. That one is coming out, I think in about a month. We have Alibaba's Happy Horse came out last month. Seedance 2. I think three months ago, LF2 last week. These models are so new, but they can basically do video editing in really high quality. So I can basically remove that boy. And then I have my video without the boy in it. And, but I still want that little guy to be in the foreground. So I go back to my reference images and then I ask, pull out that little guy using Nano Banana Pro, isolate the boy in the foreground to have a picture of him. And then I ask the video model. And then I ask the video model to animate that little guy's running and then stopping and looking. So now I have like two different videos and then using a combination of tracking. And this is all like more traditional video editing, right? Where you can attach a video to a specific pixel in the video and then remove the background. And then you have actually the clip where the little guy, he's actually moving correctly. So just an example of like, how you can also, you can combine it. You can do, if you're really good at like doing compositing and you know, DaVinci and you know, everything, every trick, you can still use Genitouria and just, you know, really be creative and try it together. Let me just show you, play the final video here. This better work. Please work. Okay. Listen. Welcome. Listen, got that. Look. So you see in the context of a video, you don't notice that this was edited, right? So you can really just blend everything together. So I think generally in visual effects right now, more steps can be done with Genitouria in this, you know, quality that we requires. And also it also requires less specialized skills. I was talking to a visual effects artist from Netflix and he says that they had used to have a very... like a very fine grained split into the whole visual effects process. And now they can see that now that's expanding. So some of the people who are now were doing only one thing, they're now doing concept art and many other steps. So visual effects artists are really going to be, I think the ones who are going to adopt Genitouria are going to see that they can actually do way more and faster. And we've got to see a little bit of that also from a TV show in a minute. Reflect a bit on like what changes. I think there's maybe this idea that, you know, how do you work with Genitouria in a creative way? The way I found is really like you have to learn, like it's a craft, it's a skill. It's not like a tool. It's like really a skill that you need to learn. And for example, one of the things that you need to notice is Genitouria tends to really go for, if you just ask for something, let's say a car, it will show you a very still typical car, like right in the center. It will give you a very boring picture. It's like, you know, symmetrical and everything. And some of the models, they are, I would say, less random. So they, if you ask for example here, I'm asking for a redhead woman. GPT-2 gives me a redhead woman straight in the middle, white background. And that is just like from an artistic point of view, really boring, right? And then you have many different models you can go for. So depending on where you're on the process, maybe you want actually to, you know, to embrace this kind of randomness and get something completely different. So you can see here, for example, in the extreme, you have Uni1, which is a model that gives very cinematic results. And Midjourney is like very good for illustration work. And you can actually get, you know, use Genitouria as like your wingman, since you want to just be like, let me get some inspiration, like throw the dice and then see if you can, you know, get ideas. There's an example of that where we were rebranding and one of the colors in our new brand is pink. And I was generating this picture. I really liked that picture, but we didn't use it for our brand. But I really liked that you have a picture of a guy, a young, maybe a teenager, and he's sitting in this, what is looking like an art studio. And he's also looking depressed. So I'm thinking, okay, that is like all those elements together sort of like form a story. And I was thinking, okay, why is a guy like that in that studio sad? So I started like building a story. And. These are the actual prompts that I'm using for generating the parts of the video. And I was just writing my ideas. I didn't even bother to, you know, be very technical. It's just like, this is what I'm going to have. I just started with that one image. So it's like a young guy, he's waking up in the morning, fridge is empty. His mom is on the couch. She has been out partying, not taking care of him. And so like, it's kind of like dramatic story. And then the guy meets a girl. And then I'm just like, basically generating a whole bunch of shots with this, or you can call it takes, I would say takes. A whole bunch of takes where I'm just like getting those scenes. And then in my video editing software, I'm then taking all these different takes. And then I'm saying, okay, the beginning of this one really works. And the second part of this generation really worked. And so on. You can see how in the video editing process, I'm mixing and matching all these different generations, picking all the elements that I liked. Until it's like forming a... A coherent story. And you can actually really, you know, bring in AI and just say, okay, give me some randomness. And then, so the process of being creative sort of like changes a bit from the creation to more like curation, where you're actually picking out the things that you like. Because I think the reason why this works with Genentech AI is because it's so fast to create and so cheap to create. That you can actually... Just create a lot and then choose what you like. And you can actually get something meaningful out of it. Let me show you the finished video here. So that's sort of like a video that just came from a single image. Just being, you know, go with the flow and get inspired. There's also the question like, what if you want like really specific things, right? And some of the new things that are possible is basically around, for example, motion control. This kind of like the approach with motion capture shooting. And so on is basically being substituted with AI. Now here in this video, you can see the little original video recorded with a camera of an actor. That is then passed to the Genentech AI video model. In this case, Seedance 2. And then also the picture of the character that I want to move like this. And those two in combination then gives you, for example, here like doing backflips and so on. This has been historically one of the most difficult things to get right in with Genentech. You can also use the Genentech AI because it couldn't keep track of like where is what's up and down on the body that is doing backflips. But that's now possible. And then there's another example here. This is probably the most prominent example, I think, of a team adopting a Genentech AI. So the producers of House of David, this is a show that's gone on Amazon Prime. They have 44 million viewers worldwide. And they're working with the idea of this concept of this TV series. And he basically got no's from, you know, all these different studios until he could prove that he could actually create the visual effects with Genentech AI. Let's just hear like one minute from an interview with him. Hey, I'm John Irwin. I'm creator of House of David and founder of Wonder Project. With House of David, you know, we had this core problem that we were told our vision was impossible. It is so hard to get projects greenlit right now. And it's a real problem. House of David is an example of that. If we had accepted the status quo and what we were told the show needed to cost, it does not get greenlit. But now with generative AI tools improving every week, we're going to enter a time in our industry very soon where the only limit a filmmaker has is the scope and scale of their imagination and how profound the story that they are telling is. Cool. So AI tools, do they save money? Are they cheaper? Absolutely. Do they allow you to work faster exponentially? But the fact that it's cheaper and faster is a byproduct of the fact that it is a more creative, more intuitive, more collaborative way to work. I'm able to manifest my vision. I'm proud of the level of scope and scale we put on the screen. I do not feel that we compromise. And that goes against what people usually say. They say you never get all three. You never get good, fast and cheap. It's always two of those things. That's just not the case anymore. You can have all three. But what I'm most excited about is they just allow you to work in a more creative way. This actually unlocks a more collaborative, more creative process. So, yeah, they have been adopting AI on a massive scale. It's a team of 100 people and basically built a full pipeline for getting generative AI to work. We're building generative AI video to the quality that is like 4K HDR and ready for Amazon Prime. I was in Cannes last week and actually met with John Irwin and we had just a chat about how I asked him, how does the creative process change for a director like him? And I think the most interesting quote he said was, I can direct an actor in the morning and then he can see a visual effect shot in the afternoon. And for him, he can actually go back the day after while he still has the actor and say, you know, it didn't really turn out right. We need to just do a little bit of change here because, you know, in the visual effects version, it didn't work right. So, you know, and I think if anybody has been like working creative products and projects, if you have really fast iterations, that really, really improves, you know, the quality in the end. It's just some shots from how you can see the difference here, like their actor to the right. And then. You know, the final shot to the left. So they actually actually able to bring in all these people into the into the shots, right? And that brings me to what I call hybrid workflows. So hybrid workflows is this kind of like idea of combining an actor, Gen ZVI, but it can also be done with 3D. So there's a company that is doing a really nice animation series called Tarmac. And they are they're really adapting, adopting AI, but they're doing it differently. They are using 3D software, so primarily Blender to create animations and then taking those animations and then basically using Gen ZVI for this kind of rendering step. Let's just hear it from them. What are you doing right now? The actual scenes of Tarmac are built primarily in Blender, the 3D software. So despite being AI, we are blocking out the full environment using very rough objects. So that the AI later knows where furniture is, where walls are, where the characters are standing and can create kind of the final quality in AI with very high precision, very high directorial control. Yes, sir. So, for example, this is an example of how that would work in an example. You give it the character, you give it location and then you export your Blender animation. And then combining that, you then have a character that moves very precisely how you made it in Blender. So, yeah, and then also let me move on to this is hyper workflow. Another way of controlling is also, let's say you want to have a specific vibe, like, you know, let's say color grading. And so I was trying to create like this little ad. There was a little ad. It was like supposed to be more comical. But I really like this picture because it had very like teal colors and orange colors and so on. And one trick for getting, you know, keeping that vibe is that once you're building all your reference images, you can actually ask it to keep the color grading. So even if I'm extracting each of those characters, I'm asking for keeping the color grading. So you can see in the reference images. For example, they are both characters are sort of like lit with this kind of golden hour light. And the same for the locations. And once you then generate the video, you are also getting that kind of color grading out. So I think it's interesting this example because you actually in the beginning of the process, you're actually choosing the color grading, not as like something you normally would do in post-production. So it's like whenever you're generating a video, let's say five seconds or 15 seconds, and you're giving it a very short script, the video will be very slow. The actors will talk slowly and so on because you're not asking the video generator to create a lot of storytelling during those, for example, 15 seconds. But you can also use it as a creative tool. So you can say 15 seconds, but then give it a long list of things to do. And then everything will be really rapid. So I had this idea of like these two people gradually. these two people going on a date and he's like this crypto guy and he's just rambling about technologies and so on and she's like zoning out. How do you achieve that, right? And then my idea was to basically to give it a whole list of technologies that he should just be hammering away, right? And she's just like looking at around. So you can actually use this kind of like approach. You know, some of the things that the video model is doing, maybe not on purpose, but you can actually use it as a creative tool. So this is the first shot where you just have like the guy talking and then she's like leaving. She's leaving the cafe and he's walking after her and still talking and so on. To create that extension where first 15 seconds where he's rambling, you know, I want to have an extension of that in the next 15 seconds. The maximum is 15 seconds. So I need to have two videos if I want 30 seconds in total. The second part of the video is basically the same reference images, but then I'm also putting in the original video as one of the references and then starting video two and saying continue this video. And then the video model will then understand what's happening in the first video and then basically just continue from there. So that's a good way to actually like, if you want to just build infinite long videos, you can do this trick. And so in this, there's a second part where I'm continuing the video and then I'm just going to play the video and you can get the impression of the vibe of this technique. I'm into crypto, biohacking, AI girlfriends, that kind of thing. You mean like Bitcoin? Bitcoin, Ethereum, Salama, Cardam, Polydoc, Kusumos, Cosmos, Terra, Avalanche, Arm, VanVleet, Nier, Mina, Offsmoots, Hubs, Hutter, Classic, Terra, TwoPoint, MCP agents, vibe coding, that kind of thing. MCP itself is fine as a protocol. I'm going to go. Why are you leaving? I'm building this startup. It's AI first. I'm getting an Uber. No employees, no anything. Goal is to be the first one person unicorn. You know, there's so many interesting projects. Taxi! Yeah. It's us. You can actually... So here's another example of this kind of hybrid workflow where it's a slightly different approach. So I generated the video first and then I went back and fixed some of the problems. So I have this guy coding his room and then a monster appears in his closet. Here's the mini-re 1980s Windows I N meditation yourself with groverancy! in a desenvolvement in your consciousness ompcountry because it's it's a lot Benovs. It took it from here not the core 17. It's okay. You can pronounce it. It's a real cute little monster. So if you notice, there's like problems all over the place in the background. So the characters look fine. The dialogue is fine. But basically like the room changes all the time. And this is really like I did the video fast. Normally, you would do a process that is more like what I'm going to show you. So here are some of the problems. For example, he's sitting at the desk, which is to the right of the closet. Next shot, you know, there's not a desk. And then later in the video, there's a door, right? And this is like a problem where you kind of like the genitorial models don't really have good idea about the room they're in. They don't build a 3D version of the room. Luckily, there are tools like this. This one is a technology coming from a company called World Labs. World Labs basically built a virtual scene. So you upload, for example, a picture of this room. And then you give it a prompt and say what's actually in this room. What works best is to give it maybe like a fisheye version of the room. But then you get a 3D model where you can move a camera around. So what I'm doing now is like I'm moving the camera around and I know roughly where I want my shot to be. But this room never changes no matter where I'm looking, right? And then every time I'm hitting that, you know, that picture button or capture button, it's creating like a reference image for me. And then with all those references, you know, reference images of the room, which are, you know, structurally correct. I can then give them back to the video. And then you can now see like how you can actually change this. I'm just going to show the video here. So you can see here the characters are actually moving in the same way. But now I have my desk to the right and posters on the wall and so on. So you can go back actually also correct it. It's definitely best to do this kind of like process before you start generating. But you can actually also go back and fix your shots if there are problems. So world models. And 3D, what you saw with the other guy, is really like the process right now for keeping that kind of structure intact in the environment. Yeah. And I think one of the important things about this kind of whole Geneseo IO is also when we're talking about filmmaking and video is really, I'm talking to a lot of filmmakers and they start, you know, removing things from their scripts because it's not possible. It's not possible within the budget. And it's often not about the creativity. They have the ideas and it's really about like, okay, the cost of doing visual effects or the cost of doing shootings in a specific location is not there within the budget. But I really think like with GenoTV Eye, you know, we will be able to be way more creative with our scripts. And just to summarize, the roles editing and directing are now merging into workflows, single workflows. The hybrid workflows is what is bridging right now. You know, acting, 3D animation with GenoTV Eye. And finally, GenoTV Eye, I think, can now make projects possible that was not possible before. That's it. Wow. Thank you. Thank you. Thank you. I could have listened one more hour to you. Thank you. Thank you. Before we go to the next session, we're a bit behind with the time, but we wanted to do a short Q&A session. So who's up for questions? There. Yeah. Thank you. Editing. Was there anything virtualized? And what do you use? I use DaVinci myself. I think editing is actually one of the areas where it's not, that is not affected that much. I think actually editing skills, understanding how you tell a story through editing is going to be very valuable. Because you might get something from GenoTV Eye where it's creating those scenes, but you need, editing is really about like, how do I tell the story and how do you make it exciting? So all the tricks about like, you know, mixing audio and the visuals in the correct ways. I think that those skills are really needed. I think, you know, GenoTV Eye will affect it eventually. But right now, it's not, it's not where I think like it's going to be substituted. I think actually those. Are in high demand right now. Yeah. There's a lot of questions in the media. We've seen it. You know, human artists did flakes. Do you think it's a situation where that's really nice? You showed, like, say, examples of creativity. Didn't you, like, people please respect? Do you think it's backstage? Creativity. Yes. I still, I think it's like, it will be cheaper to create, right? So we will have more content, but, you know, the attention of the audience will not change, right? So we will, we will, we will have to figure out, like, how do you, how do you do something with these new tools where you stand out? And like, which projects are going to stand out? I think that is going to be driven by, you know, a person with a great idea, right? And I think, for example, John Irvin here is a good example of like. He's not less creative now, right? He's actually more creative because if he didn't have the Genetive AI, he would not be like having a TV show on Amazon Prime. And I think like, you know, the human factor will be, continue being the thing that will differentiate you. The thing that you can't, nobody can copy. So, but I also think that if you don't use Genetive AI tools, you're going to be left behind. If you try to create a film in two, three years and you're not using Genetive AI. It's not going to work. That's for sure. Yeah. Let's say for video making. I think it depends a lot on the skills that you bring beforehand. If you are used to, for example, editing or if you're used to doing visual effects, it will be much easier for sure. If you have to learn everything about filmmaking. But if you learn something and practice, nothing will make it work for you. Then you hate it. But I think, I'm not sure anyone wants to do. That's good. promise. Yeah, it's cool. a lot of rock tracks and then it's okay. Getting to Eric Clapton level is really difficult, right? Genitive AI is like a little bit the opposite. It's very easy to get something that is 70, 80% of the way. But getting those kind of like last 20%, that is really difficult. True. Because that's where you kind of... One last question here. Yeah. Yeah. From a video, from a fictional and creative standpoint, or the opposite of that, from a corporate standpoint, or from like a B2B marketing, where do you see this being like applied or used? So I think one of the things that is happening now is that when creation, for example, video is probably the best storytelling tool we have. You know, it's like, it's just super effective. It's like very captivating. So I think like from a marketing point of view, you have so many options now to tell stories about your product to your customers using video. Because it doesn't cost a lot to do, you can also maybe use... What we're seeing in our enterprise customers is that instead of using agencies, many of these companies are actually in like making sure that it's their own team who are creating the videos. And one of the benefits is that they can move incredibly fast. They can almost use video as like, you know, commentary on something that is happening. And so this kind of like using the speed that you can now get, you can then, you know, be way more relevant because you can actually like join a conversation using video.