How to Think Like a Filmmaker: AI Video With Seedance

Before we jump in today, I've got some real big news. We just announced our first speakers for AI Business World 2027, the 2-day AI conference running inside of Social Media Marketing World dedicated entirely to helping marketers implement AI. Nearly 2 dozen speakers were announced so far across both conferences with dedicated AI sessions designed to take you from experimenting to actually implementing. Every speaker I personally recruited. Every session is pitch-free. Go see who's speaking at AI Business World dot live. That's AI Business World dot live. Welcome to the AI Explored Podcast, helping you put AI to work.

And now, here's your host, Michael Stelzner.

Hello, hello, hello! Thank you so much for joining me for the AI Explored Podcast. Brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators, and business owners who wanna know how to put AI to work. Today, we'll explore a proven process for creating AI video. My special guest is an AI educator who specializes in AI images and video. He's the co-founder and chief creative officer of ZenRobot, a studio that helps marketers and creators create AI visuals and videos. He's the head educator at ZenRobot Academy. Ross Simmons, welcome the show. How you doing today?

Hey, good. And you? Thanks, Mike.

Thanks for having me. It's super awesome to have you. So let's start with your story. How'd you get into AI video? Because I know there's a story there.

There is a bit of a story. So I come from an advertising background. I worked as a web developer building websites, apps, anything that clients needed. But working in a corporate environment, I quickly worked out that I wasn't built for that at all. I loved working, or, and I do love working with people, but working for somebody was just not something that really resonated with me. I didn't know that at the time. I just thought I was being, you know, full of it, but realized that it was a better path for me to try and go do something by myself. So, in 2014, I set out— this is while I was still working— to do a project where I dedicated one little part of my day for an entire year to a single project. It was just to see if I could do something for an entire year or dedicate a little bit of time each day to something. That idea resonated with me since I heard about it years before that. And I thought, let's do this. And the only thing I really had as a pastime or a hobby at that stage was origami. I was getting into paper folding and doing that. So I set out for the duration of 2014 to fold something, take a photo of it, and post it onto social media. I was on Instagram at that time. And, you know, just to fast forward through that year in April, so that was sort of 4 months into the project, I quit my job. I became a freelancer. So I was doing freelance, I had freelance work coming in. So it wasn't like I just dived out and like, you know, they had nothing lined up. I had some work. But by the end of that year, just through a series of very fortunate and I guess timeous events, I was able to build up a bit of a following on Instagram. This is just when Instagram was kind of, you know, picking up as a social media platform, which was becoming popular. And I just caught it at the right time. There was no algorithm. You just had to post at the right time every day and you had to be consistent. And by the end of that year, I'd amassed a sort of pretty big following of over 55,000 1,000 followers. And I was making animations at the time too. I was making stop-motion animations and, you know, just telling little stories with these, you know, just trying to get better at origami, but also telling stories. By the time that kind of happened at the end of the year, I was like, well, you know, no one else is doing this. I enjoy it. I've always got my web development to fall back on, but let's just see how far we can push this thing. And I ended up doing that for over 10 years, up until 2024. But always through that process, or always through those years, I was always dabbling in anything technical, so learning new software, just trying to create content in as many ways as possible, particularly digital content. So whether it was short stories or making images or anything. So when AI kind of reared its head, this was in the beginning of 20— oh, halfway through 2022. The first application I ever used as an AI image generation tool was Midjourney, Midjourney being a very popular tool now. And something woke up in me when I, the first time I was able to essentially code a piece of code or a piece of text and turn it into an image. The conversion from that, you know, coming from a web development background and understanding how tech works, to me, that was like voodoo. I couldn't understand how, like, how can a piece of text create an image? So I was obsessed from that moment onwards. Later on that year, ChatGPT came out, the API got released. I was dabbling and trying to make applications and just anything I could get my hands on. And at that stage, the tech was moving quite slowly. That was, you would get one new tool every 4 to 5 months, maybe. So, you know, as the tools progressed, Stable Diffusion came out, and then video models started coming out as well. And to me, that was another shift. It was kind of like, you're not only making images with text, you can now make videos with text. And I mean, in 2022 and 2023, the video models were like dismal. It was actually, they were unusable. But there was still, I still knew that this was the worst. It was just something in me that understood this wasn't a sort of fly-by-night thing. There was a lot of money being pumped into it. And I just was guided by interest and curiosity and just kept up to date with the tools. As the tools got better, I got better at using them. I eventually started teaching people how to use them. All my agency and marketing connections would reach out to me after they saw I was posting content onto both Instagram and LinkedIn. And they said, cool, would you mind coming doing a little workshop and, you know, just showing us how this stuff works? Fast forward to the end of 2024, I bumped into my now business partner. He used to be my boss. And he said, listen, you clearly have an appetite and an understanding of how this stuff works. I've got some clients that might be interested in a bit of training. And ZenRobot was born. So we've been going for nearly 2 years. In November, end of October, it'll be 2 years. And we do content production. So it's both You know, creating the content, be it video, audio, and images, as well as teaching brands and creative teams and marketing teams how to apply those workflows and tools in their own business.

Yeah. And everything you're creating is, is fueled by AI images and AI video. Is that correct? Very cool. So we're here to talk about AI video today. And, and you've got a background, you've been doing video for a long time. And it's fascinating your journey into it. When it's done well, because we've seen plenty of it not done well. But when it's done well, and we're going to talk about how to do it well, but when it's done well, what is the benefit? What is the upside? What's waiting for creatively minded people on the other side of this if they pay attention to what we're going to talk about today?

You know, firstly, I think that there's the big misconception around AI itself is that it's, well, AI video particularly is that it's quite easy to make. And it is easy. You can type a prompt in and you can make a video. Is that video going to be good? That depends, you know, how good are you at prompting? How well do you understand the platform? So, you know, with that misconception aside, once you do get into what quality video content in the form of AI video content looks like, it's, I mean, the upside is anyone with a creative story or looking for a creative outlet where they're usually gatekept by the people that can make videos and animations and creative directors or artists, essentially, like if you're not an artist, then you can't create. Now you don't need to be an artist. You just need to be someone who's technically Well, you don't even have to be technical. You just have to be inquisitive enough to want to understand how the tools work. And you're able to create not just single clip videos, but you can start telling stories. I think that that's a massive upside. Creating social media content, as you know, video has and will be king for a very long time. So creating content for whether it's, you know, for a social media campaign, you're trying to again tell a story about pretty much anything. You're trying to keep a log of what's going on in your life. I think that It's, yeah, when done well, it's just, it's such a powerful medium, but to do it well is not an easy thing to do.

I'm glad you brought that up because I accidentally skipped one of my questions, which is the misconception question, and you brought it up anyways, and I just want to reiterate it. Hey, it's not easy to do quality video, but it is possible. That's the upside, right? And it's not just possible, it's ridiculously powerful when you do it really, really well. And I just want to start with like, what do we need to know at a base level or a foundation level before we start creating AI video? Because I think it is kind of a very new thing. I mean, it's like the last real frontier, I think, on the creative side of things. So talk to me a little bit about what we need to be thinking about.

There is a new way of communicating, I think, with machines and with AI. I use the term machines, but I think it is communicating with computers, you know, the language which you use and the format of that language is a little bit different to how we speak to each other. But also in, you know, we've been introduced to ChatGPT and Gemini and Claude and all of those tools speak English to you and they're talking to you like you're a human and it's very, you're riffing and it feels like you're speaking to a person that fully understands what you're saying. The reality is it doesn't, but it feels like it is. But when it comes to video, and it comes to even images. Like, I think that this new form of communication, the ability to articulate in a specific— I don't want to call it coded form, but it is almost in a structured, keyworded, sequenced format, which is something that we're having to learn. It's not a difficult thing to learn. This is something we teach in our masterclass and to a lot of corporate clients, but there's a switch that has to happen and you can try to do it by yourself and you we're going to hit, you know, roadblocks like, like you always do. But, um, it's that understanding of that new form of communication that needs to happen. There is also a, a level of understanding the difference between, uh, what a diffusion model does. So a diffusion model makes images and videos, and a large language model, which is an LLM, like, uh, it's essentially a chatbot. There's a difference in the way that you communicate with them. to get certain results. For example, if you go into Midjourney and say, hey, Midjourney, create me an image of a cat walking on the beach with a cowboy hat on, it's going to look at that as a sentence, but it's only going to pick out the visual keywords that it's after. Where with ChatGPT, you'll say to it— because ChatGPT can create images too— you can say to it, please create me an image of XYZ. I want it to look like this, and blah, blah, blah. It will then translate it because it has reasoning built into it, which Midjourney as a diffusion model doesn't. Much like some of the video models too. It'll then convert it into what it understands and then produce something. And the jury's out as to which tools are better. Is it better to make images with that or that? It really comes down to preference, but also what's starting to become very apparent is that certain tools are better for certain jobs. If you look at Midjourney, for example, as an image generation tool, it's a lot better. I keep bringing up image, just so you know, because image is the foundation for video. Which we'll get into shortly. And if you understand how to make images, but also how to guide the tool in a way that gets you the image that you want, you have a lot better success rate of creating decent quality video, as well as images in a sequence consistent series.

This is really fascinating to me because folks that have been around for a little while might remember when ChatGPT first came out with images, you would give it a prompt and it would rewrite the prompt. And then you could see how it rewrote the prompt before it generated the image. And the benefit here is this large language model, ChatGPT, Claude, Gemini, understands intent and it understands kind of what you mean when you say it in like the way you would text somebody, you know what I mean? Or in your short typing. But when you go to an AI image generator, it does not understand that because it's got its own method. It's looking for things, especially Midjourney. Midjourney is looking for all sorts of weird little symbols and cues. And I've had people come on the show to talk about Midjourney. So it is absolutely fascinating that the way these things work is evolving. And effectively what I'm hearing you say is we're moving towards an era where there's like this layer that sits on top of the tool that's interpreting what it is you mean, and then it's creating the final output on the newer tools. But on the older tools, which might be even better, like Midjourney, that's not there. So understanding how they work is absolutely essential because what I heard you say is you might need to communicate with these things differently than you would with just ChatGPT. And that's kind of important for everybody to understand. So did I get that right?

Yeah, absolutely. I think there's also another very basic understanding, which is if you don't know how to communicate with Midjourney or any new tool or old tool that exists, if you know how to communicate and if you understand the power of a large language model, it gives you the ability to convert what you're actually trying to say. This is when I say it, I mean a large language model like ChatGPT. If you tell ChatGPT, please write me a prompt, or I want to create this image, in Midjourney instead of telling it to create the image, because I personally, I prefer from an art direction and creativity and ideation perspective, Midjourney is the boss like that. And you speak to most creatives that are using it, they'll agree. But Midjourney, like you said, has a specific format that it needs to adhere to in order to produce better results. But you don't need to know what that structure is. You can literally ask ChatGPT and say, I have no clue of how to do this. Can you produce an image that makes, you know, cat walking on a beach with a cowboy hat, and it'll break it down into a structured format that is better accepted by Midjourney, uh, to produce much better results.

Excellent. Okay. We're going to break it down now, starting now, uh, on how to actually go about doing this. And the first thing that you and I agreed to talk about is the idea of a concept. So talk to everybody about what the heck a concept is, because not everybody thinks like you think. We have a lot of great people here, but maybe they don't have like that art director hat on. So let's talk about what the heck is a concept and why it's so important.

Well, look, I mean, a concept for anything is really an idea around what story you're telling. Let's call it that. So, and when, you know, story is a loaded word, it's kind of like not everyone's gonna be creating a, you know, a full sort of cinematic experience, like a Christopher Nolan-style film every time you sit down in front of your machine. Sometimes the story can literally be you want to teach, for example, like me, I teach something, you know, I'm teaching AI all the time, but the concept is how do I frame what I'm teaching in a format or a story or a narrative that makes it more palatable and cut through the AI slop, let's say, you know, which there's a lot of it out there. And it's just because of the ability for everyone to be able to create anything quite easily. So I think the concept is serious. Well, really just the idea behind what it is you're trying to say in the message. It doesn't have to be elaborate. It can be something that just pops into your head. It's literally just like, cool, What am I starting with? What is my idea? What am I trying to say here? You might not be saying anything, you know, moving or emotional or deeply conceptual, but it can just be like, well, cool, what am I trying to say here?

You got an example?

One of the animations I did years ago was I used a Red Bull can. So the types of animations I used to make were stop-motion animations. And I'd, you know, I just love the Red Bull brand and I wanted to do something. It wasn't a paid concept, it was almost like a spec ad, let's call it that. And the idea was simply the can sliding in, piece of paper sliding in, piece of paper folds into an origami bull, the bull runs into the can, opens the can, the bull drinks some of the Red Bull liquid, wings pop out of the bull's sort of, you know, out of the side of the bull, the bull flies up and off it flies. It's the whole play on Red Bull gives you wings. So it was a concept within a concept. And what I did recently is I did a side-by-side comparison of exactly the same ideas, but one was generated with SeedDance, which is one of the most popular video models out at the moment, definitely the most powerful. I kind of reverse-engineered what I'd created as a stop-motion animation using Gemini, broke it down into what the sequence of events is, fed that into Seedance to create a new animation using a couple of image references. And it needed to be origami. This is what the Red Bull can looks like. And the concept, regardless of what it was I was saying, the idea was translated via the— AI animation as well as the stop-motion animation. And that could have been hand-drawn animation. It could have been a cinematic, you know, like story shot with cameras and real bulls and real cans. And so the concept, regardless of what medium is being used to create that, the concept would translate and be as good an idea regardless of what you're using.

So the idea of a concept really is you've got this vision in your mind of what it is you want to create, and you're effectively describing it in words. In your case, it was a Red Bull can. sitting on a table with a red piece of paper that turns into origami bull, and then it jams the thing and some Red Bull liquid comes out of it. I mean, that's the concept, right? So anyone who's listening, you might have kind of a vision in your head of something you want to create. It could be an ad, it could be YouTube Shorts or Reels or something like that, but you got to start with the concept. So once you have the concept, what is the next thing we need to be thinking about?

Once you've got the idea, and which again, you could use a large language model, even if your idea is quite simple, using a large language model to help you further extrapolate and develop it, I think is essential. And I'm sure some writers will think like, oh, are you being lazy? Maybe. But in this day and age, I don't think we're all— again, we're not all making Tarantino films. Sometimes you just have a basic idea and a basic concept, and you want to get that out as quickly as possible. So starting with, I guess, once you've got the concept, then it's moving into, well, What are the key visuals that you need? What is the look and feel? What is the art direction? And again, these are all things, as I'm saying these now, these are all things I would watch from afar as a web developer. I'd watch the art directors and the illustrators and whatever in the team of marketing teams that I'd work in, and I was just building the website and watching this stuff come up. But I now know that I can do all of that myself if I know what questions and what phrases and terms to use to give to ChatGPT to help me develop that. So So when we're starting with, you've got the concept, you move to images, and the images can be, like I said, the look and feel, the art direction, what you want the paper to look like, what you want the bull to look like in its final folded form, just to work on the Red Bull analogy. Once I've got those ideas, then it's like, okay, cool, we've got our key visuals. What is the sequence step? So this is where a bit of understanding of animation comes in. It's like you have a 6 or 10 or 12-frame storyboard, which is like frame 1, this is what happens. Then frame 2, this is what happens. through to frame 12. Not to say that there's 12 frames in the final shot. Maybe it could be a 30-second ad, it could be a minute ad. But these keyframes give you the angle, the lighting, the look and feel of what each of those frames is. It just gives you a high-level idea, which if you're selling the idea to the client or you're just trying to, you know, get people to buy into what's going on, you have these sequence steps. And those frames essentially become the starting points for your video. So once you've got those frames—

Real quick, let's back up the train just a little bit here. So, okay, I just wanna like I don't want to skim too far past the key visual stuff because I don't think a lot of people even understand how to do this. So what are some of the things that we need to be thinking about for the images or the key visuals? Because you do this all the time, you might take it for granted, but a lot of our audience has no idea what in the world. And I think maybe you have an example with a fragrance brand, I think, that you could share a little bit. Maybe you could kind of walk us through what those key visuals were so people can imagine them inside their their head.

To use that example, you know, I've got a fragrance bottle that I want to make an advert for. I want to help, you know, put it out on social media. This is just an example. I've got the product already and there's been a couple of photos taken of it. So one way of going about it is I could actually create a fake or a mock product using ChatGPT or Midjourney to create the product. So let's say it's created this fragrance brand bottle. Now, once I had that bottle, which becomes my subject or my hero. Yeah, your hero subject or hero object or character, let's call it that. Then you create the environment. So you'd like, what is the character? What is the focal point of— what is the point of the story? Where does this live? In the case that I used, I had this jungle environment. I just had this jungle. So I developed this arena, this, let's call it that, this environment in Midjourney because I know how to use Midjourney better. Not to say that that's the only place to start. You could create images in in ChatGPT, you could create the images in Gemini, and you could use reference images that you found on Pinterest or on Shutterstock, or it really doesn't matter where you draw the inspiration from. It's kind of like, cool, this is the look and feel, in the same way that you would, you would say to an art director, look, these— this is the mood board, this is kind of the look and feel of what I want the, the overall piece to look like and feel like. So then I've got these, um, concepts, or I've got my character, which is the bottle, I've got the environment, which is essentially the the area that it's going to live in, and then bringing in a character, another character, almost like a— let's call it a— yeah, just another character in the whole thing. And this concept that I did have was bringing this black panther kind of walking in. So it was— so I thought, cool, this was playing, and this was really just me riffing with an idea. So, you know, it was a still shot zooming into the fragrance bottle. You have the fragrance bottle in there with the black panther walking into the shot, and the black panther kind of finally staring at the camera and then jumping towards the camera. It's sort of a 30-second ad, but All of those parts and all of those pieces are developed outside, and then through video concepts using keyframes, you then piece everything together, add some music, add some sound effects, and you've got something that could be quite a compelling piece of content.

I love this, first of all, because lots of people listening right now have products, or maybe they are the character, right? So I would imagine if you are the character, then you could put a picture of yourself in here, could you not? And you can put yourself in a different setting. Have you done that for anybody before?

Only myself. But for— I mean, we—

Okay. Any tips on do— on doing that with, with people? I mean, is there any special things we need to be thinking about?

I think that, you know, getting a good photo, like a real photograph taken— it could, it could just be you, you know, standing like I am now up against the wall with a selfie, uh, having a selfie taken. I think that it's important to try and isolate the character that you're trying to— don't have it with a lot of people around them or a busy environment. That's just one thing. It just helps the AI tool see, cool, this is the subject. And the more detail you've got of it, multiple photos from different angles wearing the same clothing and using that to feed the model to say, cool, take these shots and turn this into a, let's say, like a corporate-looking photograph or a corporate identity portrait shot or series of photos. Place this character in the jungle, place this character on the beach, place this character— walking through a mall, whatever the case is, which you can do with all of this. So, and then again, this is just developing the idea and the concept. Then you see, okay, cool, this is looking kind of like what I want it to be. I mean, what we have to do for a lot of stuff is, you know, obviously I'm the face of ZenRobot, so my face is in a lot of places, but we have one photo that we use all the time, and it's just taking that one photo, placing it into multiple places. Like, this is Ross teaching at a seminar, or this is Ross, you know, against a kind of digital background, just depending on what it is we're trying to sell and what it is we're trying to— what story we're trying to tell. What's the narrative around that?

Hey, I want to tell you something that's coming that I think you're going to be really excited about. We just announced our first speakers for AI Business World 2027. That's our 2-day AI conference running inside of Social Media Marketing World, focused entirely on helping marketers and business owners put AI to work. Every speaker I personally recruited because they're getting real results with AI right now, and they're going to show you exactly how they're doing it. Quote, the AI speakers were amazing. Every bit of information they shared was completely new and valuable to me, said Sarah Heinbrotner. The beginning lineup is now available at aibusinessworld.live. Tickets are on sale at discounted prices. And here's something worth knowing. If you grab a virtual ticket, You get access to the content from AI Business World and Social Media Marketing World. Two conferences, one ticket. Go check the speakers out at aibusinessworld.live. That's aibusinessworld.live. You mentioned ChatGPT and Gemini. Gemini's Nano Banana and GPT, I think Image 2.0, are as of this recording two of the leading image models. Do you have any tips on how to work with those either to create environments or like, do you recommend having a bunch of them mocked up? Because not everybody is a Midjourney expert here, you know. What's your thoughts on that?

I think that, you know, firstly, it obviously depends on the concept, uh, you know, where do you want this person to be? And for like really knowing where to get information like that— when I say information, like where do you find these images— I always start with, if you don't know how to use ChatGPT, I know how to use Midjourney, which is most people don't, um, it's doing a Pinterest search, back to the old school way of doing things. And or going through a library of photos that you have taken already. Maybe you're a photographer and you like taking photos of the desert or whatever the case is. You have this bank of images which you can draw inspiration from. You like the look and feel of that. But just really just finding reference images, I think that that's the best place. Then saying to, you know, collecting reference images of, you know, maybe not too many. I don't think it's the more images you create, the better the result is gonna be. You really don't need that many. But if you can articulate to ChatGPT or Gemini, I like this environment, or not I like this environment, What I like about this environment is X, Y, and Z. I like the time of day. I like how this light falls on the sand. I like how X, Y, Z. The more context you're giving it about what it is you like and what it is you want to see in the final result, the better the actual result's going to be. You say, okay, cool, this is the environment. Take this character, which is you as the person that you want to put into the shot, drop that person in and say, place this character wearing a certain Hat, shoes, shirt, whatever the case is, and this is the pose they should be in. And it's actually, you know, it seems a lot more— sounds a lot more technical, but once you get into it, you really— and you understand that these models really do understand and they adhere to what you're asking it to do, it becomes so fun because you're kind of like, whoa, I can do anything with this thing. And it's like, I can, you know, change this and do that, and I can move them in this environment. I can put them on a boat, on a horse, in the sky, doing pretty much anything.

Interesting. And do you have any recommendations on perspective, for lack of better words? Because generally speaking, images created by a lot of these AI models have typically a direct-on perspective, you know what I mean? And everything is, it's not isometric, but balanced, you know what I mean? Perfectly balanced. And do you ever recommend saying, all right, change the camera from up above or down below? I mean, do you understand? I don't know what the phrase is, but do you have any tips on how to change the view, if you will, of the camera?

So, I actually taught someone how to do this recently, and I think it just comes natively to— naturally to me because I love film. Music is probably my first love, film is my second. And just understanding a little bit about film, I think, helps. And also paying very close attention. This is— I'm just speaking about how I see these things. Like, what makes an emotional shot? What makes a character look more powerful? It's like a low-angle shot. What makes a character look more submissive or weak? You know, it's a high-angle shot. Even if you don't know what all of those are, again, asking the model to say, look, I love these type of films. This shot to me shows 2 characters. One of them seems submissive and the other one seems very powerful. Help me extract from this image what it is that's making that. Is it the camera angle? Is it the dramatic light? Is it the contrast? Is it the background? Is it the soft bokeh lighting? You don't need to know all of that. All you need to know how to do is— or what to do and what to say is to explain to the model that this is what you're trying to achieve using these new characters. Give me 5 examples of what these 2 characters in a— be it an avant-garde or a moody setting of a— like, if you do know a bit about film and you know what certain directors sort of resonate with you. I mean, everyone knows a little bit about film, right? They know what films they like. So it's like, well, I like this film and I like that shot. You don't need to know anything more than that to say, make this character look like Aladdin in this scene, for example.

Oh, okay. And it will know.

Obviously, having screen or visual references is definitely going to help. But that's— yeah. And then say, cool, well, give me— I did it the other day. I said, in a single image, create 6 of Guy Ritchie's most used shots. So it's like, it was like very close up. There was a high angle shot. And there were terms in there that I'd never heard of before. Which you don't need to know. That's the thing. But this is also a process of learning. So as you see these things pop up, you know, like a close-up is really over there, high angles that, low angles there. And just again, being curious about it and testing and seeing, like, cool, now let's go back to me as the character in my story. It's like, show me a Guy Ritchie-style shot of this character in this environment. And then you're like, okay, cool. This is starting to feel a bit more— again, it depends. I mean, you're not going to do that for every social media media post or, you know, promo post, but that is how you start expressing these different and getting different camera angles and feelings in your images and videos.

And you look different than what everyone else is generating because you're asking for these things, right? Because out of the box, it's just gonna—

Exactly.

Out of the box, it's just gonna go for the basic direct shot kind of stuff, right?

And the thing is, I think that's why a lot of people think that, oh, AI sucks, but it's doing exactly what you've asked it to do. It's like, show me a cool photo of this person in a cool environment, it's gonna be like, okay, what is your version of— what is your definition of cool? I think that also what you learn, and again, something we teach all the time, is being able to articulate what exactly you want to see in the image and in the video, and less about the feeling. I mean, feeling, I think, or vibe is important, but it's if you can direct exactly, because you know that to create something more intense, it's gonna be a close-up shot. If you want something— if you want someone to look vulnerable, you do like a zoomed-out shot of them them in this sort of desolate arena where there's nothing around them, or make them look slightly smaller than the people around them. You know, it's, it's these little visual storytelling techniques that help you get that. But only over time do you learn those techniques. But if you're just starting out and you just want to tell like pretty basic stories and get a pretty standard— not standard, but tell a slightly different story to everyone else— ask the model What is gonna make my content stand out? What is gonna make the story more compelling? And test those out and see. And then it's A/B testing. You run one series of ads, it's like, cool, that didn't really work. This one did. Was it the angle? Was it the clothing they were wearing? Was it the pace of the video? Was it, you know, it's all these things that like, it's all metrics, right? Like, and that I think just comes with time and iteration.

It's really intriguing because I have now a newfound respect for people that do storyboards, right? Because these storyboards aren't just the drawing of the character, they're showing them in a scene. And in a setting and in an action environment typically, which transitions into keyframes. You started there a little while ago and I want to come back to that. I'm anticipating that you could have a bunch of generic looking keyframes like, okay, I know that I've got this, this can of energy drink, which is Red Bull. And now I have this red piece of paper and then it turns into origami. You know, I know kind of the basic steps. I could just start out with these images. I could start to imagine and get creative about how could I more creatively use these images. So are the keyframes almost like the storyboard, or are the keyframes the foundation for a storyboard? Talk to me a little bit, because as I think about this, I'm like, okay, well, we could change up the perspective of the Red Bull, you know what I mean? So it can have Red Bull, so it's scared of the bull, right? Instead of just sitting there in the middle, right? So like, talk to me a little bit about these keyframe things and kind of how all this works.

I think that, you know, the example you've just brought up there to make the— if you want the bull to look more powerful and dominant, uh, you want it to be slightly bigger, maybe not bigger than the can, but bigger than what would be expected of it to be. So how do you do that? I think that with a keyframe, essentially a keyframe is just an image that you're using as a reference. If you're speaking about traditional sort of keyframing or in the AI sort of space, it's a start and end frame. So there's one technique, which is the start frame is gonna be where the video starts and the end frame is where the video ends. When I say the video, I mean, that could be anything from a 5 to a 30-second clip. I think if you're going into 30-second That goes a bit much. But if you're trying to create a short, let's say between 4 and 7-second-long clip, if you think about somebody filming, a cinematographer filming, if they're running the camera for 20 seconds, of that 20 seconds, they might only get 3 seconds that's usable content. Maybe the actor does something strange, an airplane flies past, but they capture that one little piece and they're like, cool, they've got it. Whether you know much about film or not, it doesn't matter. It's more about you're So you're creating these clips, and these clips are broken down into— you can do it as a start and end frame. You could use those images as references. So you could say, this is the basic angle that I want it to start at. It doesn't have to be the exact frame. And this is kind of where I want it to end up as a shot. And then the model, what's amazing about these models is it fills in what you don't know how to articulate. So if you do know what to say, you can, you can break it down into like into each second being like this, from 0 to 1 second, this is what happens. This is what happens from second 2 to second 5. This is what happens from second 5 to second 10. Over a 10-second clip, it can actually hit those and it hits them wonderfully. That's getting very technical, but for the most part, you can say, okay, first shot, I want the Red Bull can. You just want a clean plate. There's nothing on there. So empty shot, that's shot one as your keyframe. The second keyframe will be the Red Bull can in the middle of the shot with nothing around it. So that's your start frame. And between those 2 shots, it knows if you say, and then using the prompt, you're guiding what needs to happen in the scene. So you have your start frame, which is an empty shot. The second frame or your end frame is where the can is. And you're telling it using the prompt to say, Red Bull can slides in from the right at a slow pace and stops directly in the center.

Done.

Then you've It's almost like all 3 of these things, the start frame, the prompt, plus the end frame, are guiding what's happening in the screen. And you rinse and repeat that throughout. That is one way of going about it. The other way is just dropping them in.

Do you recommend going about that? I mean, that seems like the logical way to get started, doesn't it?

To get started, and if you've never done something like this before, absolutely, because it gives you a sense of what can be accomplished in 5 seconds. Because sometimes, I know people have done this where They have a start and end frame and a 5-second clip is selected. You've got a dial that you can select. You can select anything between 5 and 30 seconds with SeaDance at least. And you'll put a start and end frame in and they type this gigantic novel of what needs to happen in 5 seconds. And the model attempts to do all of it. And then there's hands and fingers and spaghetti and arms and limbs all over the place. And they're kind of like, oh, this sucks. Why is it not listening? But it's kind of like you're telling it between 0 and 5 seconds, that much must happen. Logically, that cannot happen. So in following through with that process, you realize, oh, okay, so the first 5 seconds, I can only tell it to do a certain amount. You cannot say, you know, Red Bull can comes in, piece of paper flies in as a jet, the bull turns into, or the paper turns into a bull and things explode. Like, there's just too much going on there. But if you break that down into a storyboard, beforehand and you have your keyframes, then at least you are guided.

It's—

the storyboard is less for the model and more for you to understand what the sequence should be. And then once you've got that sequence, you're like, cool, that's working. Then you start with the start and end frame. But yes, to your question, I think that starting with start and end frame is, is a really great way of going about it. You don't have to have start and end frame. You can just have just a start frame, which is also handy. So start frame plus prompt creates the video. So it's Like you could say empty shot and you just have your start frame, which is your empty plate. And you could say a Red Bull can slides in from the left. You could maybe make it a 10-second clip and say a Red Bull can slides in from the left, a paper jet flies in from the top, lands next to the can, and unfolds into a flat piece of paper. Then you've got your first 10 seconds. Cool. Then you're done. Then there's steps and sequences, you know, depending on how technical you wanna get, of grabbing the final frame from the previous video, using that as your next start frame.

Okay.

So now you're building, now you've got, okay, cool. So you've got the first 10 seconds, you're extracting a frame from the end frame as the end frame, which now becomes your new start frame for the second 5 or 10 seconds on the video.

You said there was another way. What is the other way? Just having a start frame as a reference?

Yes.

Okay, got it. All right, you've already mentioned SeaDance a little bit. So let's talk about SeaDance and any other models that you recommend. There's a really good chance that my audience is not familiar with SeaDance. They are owned by ByteDance, which is the company behind TikTok. Um, why don't you tell us a little bit about what it is about SeaDance which is amazing and why you love it?

From a prompt adherence and a reference image adherence perspective, it is, it is near flawless. It is insane. The latest model, SeaDance 2.5, is like nothing I've ever seen before, whether you have been tracking, you know, the state of AI video or not. 2 years ago, there was a very popular trend going around, let's say 2 and a half years ago, of Will Smith eating spaghetti, which did the rounds. It was basically Will Smith just shoving spaghetti into his— it kind of looked like Will Smith, but because it could have also been Quasimodo, could have also been anyone else, anyone else. And it was just him stuffing spaghetti into his face. And people are like, AI video, cool, bro. Like, good luck with that. And if you see the videos now that Seedance is able to produce, it's like flawless cinematic. You cannot tell the difference. Like, I saw one a recent Will Smith eating spaghetti video where they're doing a mock thing, like, at Sea Dance, like, almost mocking those old videos. And the real Will Smith walks in eating, like, a plate of spaghetti, and it's fully cinematic. He's in a beautiful suit, and he's commenting on— he's like, what's this guy doing? Kind of thing, you know? Eats the spaghetti, puts it down, they have a fight, and it's just insane. And you're really just—

Except it's not the real Will Smith.

No, no, no, it's not. That's the thing. And, you know, I think that it's a very advanced model. Just a note, a technical note on that. If you go and Google Seedance now and look for a website to just use Seedance, you're gonna get a list of about 10 websites. So understanding that, you know, regardless of which video model you're using, pick a platform that has that built into it. So a couple I can mention, one is Luma AI. Luma is something I prefer to use. There's another platform which I'll note code-based applications. There's one called Flora and there's another one called Figma Weave. But Artlist, Imagine Art, OpenArt, Krea, these are essentially platforms that allow you to generate everything inside it. So you would pull, you would just have an account on these platforms and you have access to Seedance. You have access to Kling and Veo and Grok and all these other models. So just in case you are—

Just for the So just for the record, lumalabs.ai, is that the—

That's Luma Labs, yes.

Okay, so they are kind of an aggregator is what I'm hearing you say, right? Like they will work with different tools?

Pretty much, yeah.

And this is the key thing, like you can't just download Seedance like you can ChatGPT is what I'm hearing you say. It's a model that is powered on the backend by some other tech stack, right?

Exactly, exactly. And it's connected via what is called an API, which is just a little connector which connects the capabilities of what that tool can do into this aggregator or platform.

What's the typical cost for something like Sea Dance? And you were mentioning like the duration. I'm assuming you were talking about Sea Dance, right? Like you said, you can do up to 5 seconds to 30 seconds or something like that. So what's the— what on average, what is it going to cost to produce something like this?

So one of the platforms I use actually gives the dollar amount of a 30-second. So you get tokens and those tokens can be broken down into, yeah, time or actual dollars. And I think a 30-second SeaDance clip at 720p resolution was $14.

Okay.

If I'm not mistaken. So as soon as you bump that 720 up to 180, it doubles. So that would go to sort of $28, between $28 and $32, which, you know, it sounds, I don't know if it sounds cheap or expensive, but from where we came from, where 5-second clips would used to cost less than 20 cents. It is relatively expensive. But what you're able to create, it's worth the money. If you know the steps, like honestly, if you've never used AI video before, I would not start by trying to create a 30-second 4K resolution sea dance clip because you're going to waste $50 on—

Can you upscale it? Is there upscalers? Like if you start with—

talk to us about Most of these platforms, I know Luma does, I know Flora and FigmaWeave definitely do too, is they have upscalers built into it. And I actually did a costing today. I created a 480p clip, which is reasonably small, and I upscaled it twice, I think. So it came out to roughly between 3K and 4K. And the clip itself cost me, what was it, $6, I think. And then the upscale cost me $3. So that's cheaper than what I would have paid the $15 or $14 for. So that does—

Was the resolution good? I mean, was it quality?

Yeah.

Upscalers do good, John?

Yes, definitely. So the 2 upscalers that— the 2 that are available in terms of, well, of image and video, uh, one is called Topaz Labs, uh, which you would see built into all of these platforms, and the other one is Magnific.

Okay, some people are gonna want me to ask this question. Gemini does video. Does it make sense to ever start with Gemini and then go to SeaDance and have SeaDance enhance it just because maybe Maybe some people get some basic video for free included with their Gemini, or what's your thoughts on that?

So Gemini's video model, the current one, which is VEO 3, V-E-O 3, you do get free generations there. And VEO 3.1, when it came out, was state-of-the-art. It was the best. This was before SeedDance and before CLINK 3.0. You know, it's such a tricky thing. It's not like you can practice with VEO to become better at SeedDance, if that makes sense. It's not like it's a It is a more superior model, but I think you would do yourself a lot more favors by understanding the prompt structures for the different models. So I know that SeaDance, from a prompting perspective, you can break it down. You can say for a 15-second clip, you can tell it between 0 and 4 seconds, I want this to happen. Between 4 and 8, I want this to happen. Between 8 and 12, so on all the way through to 15.

And you include all your keyframes in there, Right?

You don't even really need keyframes. You can, you can tell it in the prompt itself. You either use those as keyframes or you use them as reference images. The key difference there is I think references work better with the model because it's not the model— like, the keyframes are like, it must look exactly like that at that point. And if you, which most people can't, can't visualize that is where it's going to end at second 4 and this is the first frame, what is happening between those frames, the model's trying its best to get there and you're going to get weird stuff going on. The camera angles are going to be weird. There's going to be little floaty bits. So using it as a reference rather than say, this is kind of what I want it to feel like. That's essentially what you're saying with reference images. This is basically what I want the feeling to be like, but I kind of want the person to end at second 15 holding the stick up in their hand or whatever. And then it does a much better job because it's drawing from references that make sense to it as opposed to it being dictated by the keyframes that you're using, if that makes sense. Yeah.

It totally does. Ross Symons, we could keep going for a very, very long time. I know you have a class that you're teaching on this. So why don't you tell everybody, first of all, if they want to connect with you on the socials, where should they go? And secondly, if they want to potentially join or work with you or whatever, where should they go for that?

So zenrobot.ai, you can, like, everything is there. All our contact details are there. I am very active on LinkedIn. So my LinkedIn name is Ross Symons, S-Y-M-O-N-S. And Zen Robots pretty much anywhere. So if you just Google Zen Robots, or I think you'll remember that maybe more than my name, but we do have a masterclass that runs every month. We have 3 left this year. The last one will be in November. And that is a, it's a seat-based 4-week program where we take people from the basics of learning how to communicate with AI to make images, move from images into consistency, from that consistency into motion, and then kind of putting it all together. in what we refer to as a portfolio-ready piece.

Awesome. Ross, this has been so amazing. I know there's a lot of people that are going to like go literally start messing around with creating model images and go and messing with SeedDance as a result of it. So my creative juices are flowing. Thank you so much for coming on and sharing your insights with us today.

For sure. Thanks. Thanks for having me, Michael.

Hey, if you missed anything, we took all the notes for you over at socialmediaexaminer.com/ASMR. 123. Be sure to follow the show on your favorite podcasting app. And if you've been a listener for a while, we would love a review. Also, do let your friends know about this show. You can tag me on Facebook, LinkedIn, Instagram, X, and even on TikTok now. I'm on TikTok @MikeStelzner. Also, do check out my other show, the Social Media Marketing Podcast. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored podcast is a production of Social Media Examiner. One more thing. We just announced the first speakers for AI Business World 2027. Dedicated AI sessions, personally recruited speakers, pitch-free training designed for implementation. And virtual tickets cover both AI Business World and Social Media Marketing World. See the lineup at aibusinessworld.live.