Seedance 2.0: Make 100 AI Ads in 33 mins

0:00Yeah, so my friend Sirio, one of the world's greatest AI creative minds, just took me through C-Dance V2 and it blew my mind. It blew my mind because the people that are going to understand how to use this model, and this is the world's greatest AI creative model on the planet, they are going to be able to create AI influencers, faceless accounts, original movies, literally, ads that convert, ads in any language on the planet. This is the creative AI model we have all been waiting for. In this episode, Syrio takes you through a bunch of these use cases, show you how to do it, the prompts, the tactics, everything you need to know about C Dance V2 is in this episode.

0:45And if you stick to the end, you will be a weapon for how to use C Dance V2, how do you use AI video to make money, create content that gets you followers, and more.

1:06I've got one of my most creative friends on the podcast, Serio. Serio, by the end of this episode, what are people going to learn? Ooh, a lot of things. First, why are all these image models, video models, and API providers so important to your business if you are starting some sort of AI app or if you're trying to solve for issues in the creative space with AI tools? we're going to talk about all the use cases cdance 2 is here so we're going to try and explore all the use cases and how we can build on top of cdance to solve particular issues and then productize around those workflows i love it yeah there's tons of tutorials and videos about okay cdance 2 is here look how cool this is but this is going to be a more practical guide to actually, okay, great.

1:59How can you build a business around these models? How can you make money from these models? How can you create creative assets that are going to transform your business? So that's my hope out of this episode. And Cereal, if there's anyone who can deliver on this, it's you. So excited to get into it. Thank you for having me, Greg. And I hope I can do my best here. All right. Okay. So Cdance 2 officially launched today. You can access it anywhere in all of your favorite AI tools. And something very interesting about CDens is that it's probably the first AI model that allows for multi-input generation.

2:43So what does that mean? We usually, if you're never, or if you've used AI tools, you know that you can either use first or last frames and you can generate videos based on those two inputs. But now with C-Dense do, we're able to generate videos with multiple inputs. For example, we can add up to two images, we can add up to two videos, we can add an audio file, and then what C-Dense is going to do based on what we are prompting and what we're trying to achieve, it's going to combine all those inputs together and give us a final video. That's something very interesting here, because it allows us for way more control.

3:25And to show that, I'm going to go into my demo page. Give me a second. So this is what it actually means, right? In here, we have a video, like a green screen.

3:45So, and this is AI generated, completely AI generated with C-Dense, by the way. and let's suppose that I want to change I'm a production studio, I'm creating this game and I want to put some sort of a demo on my social media or a quick video on my landing page and I want to replace these two people with two different characters but at the same time I want to replace the background. Traditionally this would take a very long time but also it would cost a lot and what we're doing here is that we're using the multi-input feature inside Cdens and we're going to have our character one, our character two, and then we're going to have our background image.

4:24And since this is again multi-input, we're going to reference all these inputs in the prompt by tagging them. We're going to hit generate and it's going to take about 60 seconds for our video to generate here. And again, the purpose of multi-input, as I said, is to get very creative with our editing process. C-Dense 2, it's not only a video generator, it is a video editor. That's how I see it. It's almost like Nano Banana Pro, whereby the use cases are unlimited. It's not just producing an image through text, in this case, a video through text or a video through an image, but you're combining multiple inputs to produce an output that's way more complex than traditional Image 2 video models.

5:15You can do something very similar with Kling 3, but the quality of C-Dense 2, based on my testing so far, is unmatched. And we're going to see all the use cases and all the demo videos today. And I hope that you make the decision on your own. But this is the video that it generated.

5:43so this is pretty crazy let me try and pull uh the original uh video input that uh the green screen here

5:58this one over here is what cdent the the motion control is crazy here yeah and it's simply from a prompt you're literally telling it to control the motion to keep the motion of the original video exactly the same this is all natural language first of all this just like exceeded my expectations um i think this is beautiful two questions for you one is from a prompt perspective did you just manually create that prompt or is that something that you used an lm to optimize you you can definitely use an lm to optimize. I think that Claude does a phenomenal job and it's the best by far, especially the 4.6 version, Opus 4.6.

6:46I've used GPT before, but I do think that Claude does understand prompt engineering for vision models a bit better, at least in my experience, and I could be biased. Um, but this is, uh, the, the more something with C dens is that the more detail you give it, the better it does differently from other models, um, where you can be simple and to the point, for example, clink three, if you're simple, straightforward, you're not using a lot of tokens or words in your prompt, um, then it might do a better job. what I'm figuring out with C-Dance is that you have to be highly specific if you want to get very high quality output, especially if you're doing something that relates to preserving character identity, that relates to preserving particular motions in the video or particular

7:50transitions throughout. So I think that both work. I like to start my prom myself, and then most of the time I will optimize it with Claude 4.6. Cool. And before we go into the next use case, I think, I mean, you're a stylish guy, you know, and I think one of the reasons why... We're in all the way. Yeah, you are. You know, you're wearing your hat says Los Angeles, you know, upside down. I feel like you always got good style. Every time I see you, you've got good style. A part of why this video crushed it was, yeah, C-Dense 2.0 did a good job, but also your reference images are really on point.

8:31How were you able to find those reference images and videos and any tips for people? Everything starts with a very good idea, a very good source reference, source image. What is your vision? You can describe your vision, but the second that these LLMs or these models see a source reference, They're able to understand your taste and they're able to mimic that reference image into something more concrete and more tangible for you. So always focus on having a great source image, source reference that matches your idea. It's like in any traditional art. I'm also a painter. I draw. I sculpt. And for me, in order to visualize my idea, I have to have something in front of me that I can see and I can be, hey, I'm inspired.

9:22I want to create something similar to this. And it's the same thing with LLMs. Like think of them as humans. If they were to be like your assistants or like your friends, that's how they understand inputs. So give it a very good source reference of whatever you're trying to achieve. And then follow it with a very specific prompt. All right. Should we keep going? Yes. Okay. I'm going to showcase a video that I did myself. This is a virtual try on video. I recorded myself out there in Canada, in Montreal. It was like minus 30 degrees. I was wearing shorts. And I was like, oh, I wonder whether AI can put me into this outfit.

10:08So now I want AI to replace me to actually put me put on this outfit and have a bear walk by. and this would be you know helpful if maybe you're doing like what like an ad maybe yes if you want to replace let's say that you have an actor you're you did an e-com shoot and um you want to have the exact same motion of the model and you just want to replace the clothes the clothes that they are wearing because you're creating this very cool transition or just because you want a very clean uh style throughout um your your um e-commerce assets so let's see what he came up with it is minus 30 and i'm wondering whether ai can help me put on this outfit okay how about have a bear walk by look at look at the details like look at like how when the bear walks by and then you have all the um footprint can you stop it for a sec yeah i cannot tell that your outfit is ai like honestly not only that but what i'm very impressed with is that my face is the same if i saw this video myself like i know how to how to i'm very familiar with all the ai models open source closed source everything um and i can tell you which video is what i can tell you if it's generated with cling if it's generated if i can tell you if it's generated with cling if it's generated with uh c dense 1.5 uh with wan um but when i saw this video myself i'm like it looks like me there's no distortion in the face which is crazy um and yes the alpha too like it was able to match the exact look at the boots like look look at the pattern of the um of the pans over here so if you go into our source reference so if you go into our source reference over here yep you see how it has like all these um like this specific pattern this cut that's like dark yep if we go into our video it's here yeah it's crazy okay how about have a bear walk by look at like the footprint it's like looking at the bear it's tracking the bear with, with, with, uh, with the eyes and the head.

12:28Like, so it understands the input very well. Um, and mind you, the input here was very simple. Like I didn't go into any details. I could have been way more specific. I could have actually described my outfit so that the outfit could have been, um, more accurate. Um, so it's phenomenal. And again, it doesn't take more than 60 seconds. And this tool that we're in enhancer, you're the founder of this, right? yes sir i am the founder of this so this you can use enhancer it's not just with c dense 2.0 right you can use other yeah you can use it with with um yeah you can use it with any model another cool use case is translating for everyone that wants to build a translation app that's going to take 30 seconds to translate or not only that but also replacing the character in the frame.

13:23Take a look at this one. So we have this original video in Chinese. She's showcasing the glasses.

13:33But now your company operates in the United States and you want to showcase the same glasses. You want to have the same asset. You want to have her move exactly the same because you're A-B testing the ads and you want everything to look exactly the same. But you want the language to be different and also the model to be different because you're targeting different demographics. So here's our reference model. This is a model that we generated previously. And now we want this model to replace the woman. But also we want her to speak in English. And this is the prompt that we're using. You can stop and screenshot this.

14:07Go and use these assets. We're inside video editor. We're going to hit generate. Again, you can take your time to read the prompt. And what it's going to do is that it will replace the woman in the first video with our source image. And it's going to translate everything she's saying in Chinese from Chinese to English in a matter of seconds. So let's see how it does. Yeah, this is really interesting. Also, just like creating ads and just creating content in like 100 languages, right? Yeah, it's A-B testing at its finest. Yeah. And getting higher conversion rates, just getting cheaper ads because of that.

14:50Optimizing, optimizing, optimizing. Yeah. All right. There it is. So what do you think she said? She said, I love you, Cereal. You wear my new glasses. Let's see. We translated the original video from Chinese Mandarin. this one's amazing it's flattering and versatile must have so you see the wink let me go back into the reference video it feels like she's selling the glasses that I'm wearing back to me and I want it's so good she's doing such a good job selling it I want another pair that's how good of a job this is doing look at the wink look at the way that she puts her hand on her glasses it's the exact same motion.

15:44This one's amazing. Look at the blur in the camera. The focus. The motion. The focus. This one's amazing. It's flattering and versatile. Must have. Great. Nailed it. This one is very interesting. Look at this video. What we're going to do here. This video is a vlog. this is an ad now we have a package right and this is like traditional just like 3d render there's no branding in the package this is meant for like evergreen

16:27okay like a template you can buy these templates what if we actually replace that package with this image so what we're doing right now again And here's a prompt. You can screenshot this. We're replacing only the package and keeping everything the same. Hit generate. And you can find any 3D asset out there. You can start applying texture to all these 3D assets by combining the source reference with image references and just literally telling it to, hey, make sure that you put the texture from image number one into the 3D render video in video number two. We could do this with Nano Banana in images, and now we're doing this with C-Dense 2 in videos, which is quite insane.

17:15So that template, was that found on a stock video website? That was entirely generated, but you can go into FreePick, for example. I think that they have a bunch of templates like that. I'm not quite sure if it's a video template that they have. But you can take an image that is an image template. You can turn it into a video and then you can put everything together and then you can create this templated video and then you can replace the templates with your source references. Let's take a look at this.

17:51I didn't do any changes. There you go.

17:59the logo is completely consistent and understood like the background that it had to be yellow um kept everything the same is this is cdance 2 the best video model to ever exist for now yes yeah like as a recording for uh no yeah it's v04 um but by far uh it is the best out there in terms of realism, in terms of motion, in terms of quality. It's only up to 720p for now. And when they release their 1080p version, it's going to be a game changer for anyone that's creating digital assets. Cool. Do we have time for a couple more? Yeah, we do. So I want to show you two other use cases that are very interesting that everyone would love.

18:51the first one is extending videos you have a three second video you have a 10 second video and you want to extend it to 15 more seconds while keeping everything the same we could not do this before um now google vo 3.1 kind of tried but look at this we have our three second video here and we don't know what's happening next we can recreate this entire scene here is the prompt you can screenshot it you can take a look at it and then again we're using our video extender feature and generate and what it's going to do is that it's going to continue the the actual storyline based on what we said in the prompt while keeping everything consistent This is use case number one of video extension.

19:45And there's a different use case for a video extension that would actually fill in the middle of the video. This is extending the last bit of the video. And there's a use case that I'm going to show you after we explore this one where it's going to fill in the gap. So we have two videos. It's going to figure out what goes in the middle, which is insane to me. Yeah, I mean, if this could do this, this is big. because this is this has been a pain point for me personally with playing with some of these models like ads yeah yeah exactly with ads literally ads or just traditional filmmaking doesn't have to be ads like there's something that you just you want to have at least three more seconds of that video you cannot do that let's take a look at this so it extended the video from the point where we where the video cut right

20:42you have a complete you have a different scene and it's the exact same last frame um this is use case number one of of uh extending your videos there's a bunch of others but there's one last thing I want to show, which is AI influencers in lip syncing. This is the best model for you to generate AI influencers. And they can do anything you want them to do. Again, as I said, you can screenshot this prompt over here. The prompt is highly specific. This is a source image generated with Nano Banana Pro. You're going to go and use the asset. And the way that the influencers or avatars lip sync is simply you prompt it to say particular things.

21:39In the prompt, you go and say, hey, she is saying. And then you go and say, give me a second. Or she's saying, this is what I mean. And what do you call these? Quotations. In quotation marks. So everything inside quotation marks is what the model or the avatar will say. Very simple, natural language. You just tell it what you want it to do and it's going to understand exactly. um now of course there's ways to prompt things so they look and feel more realistic especially emotions um how to control emotions how to prompt emotions you do not prompt emotions by saying hey the character is sad or the character is happy you have to describe the muscle movements right because just saying character is sad okay there's not a lot of control like sad how there's thousands of ways for someone to be sad.

22:35But by describing the muscle movements, by describing the transition in motion, transition in tone, in body language, it's able to achieve more realistic results. So this is what we're going to see here. That's why this is a very long prompt, because I'm being very specific with what they're saying, because the aim for this video is so it doesn't look AI. Okay. Yeah. Give me a second. This is what I mean. The way I breathe, the way I talk right after moving, it's all generated inside Enhancer. It's crazy. There's, let me show you another one. I have goosebumps on that one. Like that, that looked real.

23:23No, let me show you another one right here um so this is our sort of reference i say that we want to generate ads right and we have our product and our product has some sort of text and again one of the main flows for other video generators is that the text was usually wrapping or was changing as the video was generating right um then we have our prompt over here again we're very specific with our prompt with what they're saying how they're saying it how we're structuring the prompt We're going to hit generate. And then we're going to have her talk about this product that she never tried because this AI model does not have thoughts or does not have taste.

24:01She doesn't exist. And the product was never sent to her. That's the beauty of AI models because you can create a version of yourself if you want, or you can create a completely different IP. And the brand does not have to send you the actual clothes that would cost them a few bucks to actually ship them to you. And now multiply that by like thousands of AI influencers, it becomes like very costly. Now the brand can just be like, hey, can you just place this inside your image with Nano Banana Pro? It's going to keep it very consistent. Can you just generate it with C-Dense too? Or maybe we can do it for you if you want.

24:38And there you go, like unlimited content, very cheap. There he goes. Okay, quick taste test.

24:50huh wait that's actually nice it's not super sweet it's really clean i wasn't expecting that yeah i drink this look at look at the text like the text is quite on like spot on it's not changing insane all right syrio this c dance v2 this feels like the best creative model to ever exist with it being so good just walk us through quickly how to think about like think you know why would we use any other model are there any benefits using another model or as of recording this should we just make this the default i think that at some point uh what will happen is the same thing that it did with nano banana where it became the best ai editing image editing model cdense to me seems like it's the best by far.

25:41However, and the reason why it's the best by far is because you can practically, you can animate UX UI. You can animate like logos. You can place logos within a video. You can do so many things that other models are not capable of. And you can generate like very good lip syncing. Now, of course, some others are very good at other things. Maybe emotion control. Clang 3 does a very good job at that. And there's other models who are fine-tuned so that the images look way more low-fidelity, like more realistic, or it doesn't look like they have the cinematic feel. Kling 3 has a cinematic feel. When you look at the video, you know that it's very good at producing cinematic videos.

26:28I'll show you an example of another video model that we fine-tuned inside Enhancer. It's called Enhancer V4. And what this model does, again, it's not the highest fidelity video model. It does not produce like this crazy transitions. It just might not keep the character extremely consistent. It might not have multi-input references, but it produces. Hi, guys, this is crazy. Like, I am not even real. This is an answer version. Probably cannot do the same thing because it has different type of color schemes that it produces, It's kind of different in depth, different ways that it treats the background, different ways that it treats the subjects.

27:07And this is a different video model that is fine-tuned particularly for this exact use case. For that's talking head video generator. So I would not say that C-Dense 2 will replace everything that's out there because it really depends on what you're trying to achieve with the model and how you are using the model. But I think it's going to be for now the default model to generate and edit videos, especially editing videos, maybe not video generator, but it's phenomenal and will be the state of the art video editor out there for any use case, really. What's the best video generator? Well, I mean, for now, it seems to be C-Dense.

27:55Right? Exactly. It seems to be C-Dance. It sounds like what you're saying is the daily driver is going to be C-Dance generation and editing. However, there are some use cases that certain models have a different look, visual look, like you were saying, Enhancer V4. If I'm trying to go for that look, I might just use that for a specific use case. And also it depends on what the, again, what the user is used to. There's things that the user really likes or the creative in this, in this case, really likes a particular model and they just want to stick with it because it's good enough for what they're doing.

28:37And maybe it's cheap, right? Maybe it's faster. Maybe they're just producing low fidelity video for social media and they don't want to spend like$3 on a five second video like Google VO3 or at least back in the days when it was about a three video per, what is it, five or eight second clip. And again, people that are using generative tools are not using them just for fun. They are normally using them or monetizing through them. They either have some sort of a business, maybe they're a service provider, maybe they're creative, or maybe they're building their own app. It's not that they're coming to these platforms and spending all the money without making something back.