Seedance 2.5 Tutorial: My Full Cinematic AI Film Workflow transcript

SeeDance 2.5 just released, [music] and with it comes a lot of exciting new updates. Over the last week, we've been testing it and pushing it to cinema-grade realism. >> [music] >> And in today's video, I'm going to teach you how you can go from idea to final execution at the highest standard, [music] so that you can charge the highest rates in the industry for AI commercials or short films. [music] My name's Roe Keeth, and it's our mission to create 100,000 new jobs for creatives in the AI space.

[music] So, in today's video, we're going to also teach you all of the new updates that have come from SeeDance 2.5, as well as teaching you where is the most cost-effective [music] place that you can use it right now. So, without further ado, I want to show you a project that we've worked on, which has gone super viral on social media, and this is that project that we made with SeeDance 2.5. >> Apparently, you can now generate 30-second long videos with SeeDance 2.5. 30 seconds is a long time, right?

[ __ ] >> How many images can you use as a reference? >> 50. >> What? >> 50. >> You playing with me? >> 50. You now have full control over every generation, every character. >> [ __ ] I shouldn't be here. >> Okay, so that was super cool. Now, no, that wasn't 1 30-second long generation. We actually kind of separated it scene by scene. So, we went, okay, so let's say we had this scene right here, for example, we would go and we'd find like and design our own location for this.

We'd design all the characters that we wanted in it. We'd design all the objects that we wanted in it. And then we would create that short sequence, and then we would build and stitch it all together in our editing software of choice. Now, that's not to say that you can't do one single 30-second long generation and get an incredible output. I'm about to show you that very, very shortly. But, let's discuss all of the new updates that have come to SeeDance 2.5.

So, the first one is, as you learned from this video, is that you can now generate videos that are 30 seconds long. You can one-shot prompts, which is incredible. You can also upload 50 image references in the prompt bar now. Not only image references, but video references, audio references. So, if you want to have a certain specific voice that stays consistent throughout your generations, you can upload that. You can upload 50 locations, 50 images of certain products that you want to show up in the video.

You have a lot more flexibility and control with this new model. But, with more control, more flexibility, and more time, all of that equals more compute for these companies, which means more cost for the end user. So, let's talk about costs. SeeDance 2.5 right now is very, very expensive. To generate a 30-second long video can cost you anywhere between $40 to $60 right now, which is criminal. And we've spent thousands of dollars over the last week making three different projects for three different clients, and it has been a very expensive process.

However, I'm going to teach you right now where you can get unlimited SeeDance 2.5 for the next 33 days, so you can play, test around with it, and you can get that right now on Higgsfield. So, we've left a link in the top link in the description, so you can go and get that straight away if you're interested. And thank you, Higgsfield, for sponsoring today's video and giving everyone access to unlimited, so we can play to our heart's content.

So, without further ado, I'm going to show you what it looks like to create a video that is 30-seconds long. So, here is that video, and then I'm going to walk through the process of how we made it all the way from idea to execution, so you can replicate the same process for your brand or business. >> Do you mind doing my back? >> Not a chance, mate. >> Come on, it'll only take >> I'll do it for you, baby. >> [ __ ] Fancy doing both of us? >> Protects your skin.

Not your dignity. >> I love that. This is honestly one of my favorite videos that we've ever had the chance to make. It just felt so fun, so fresh to create, and the design for all of this was such a laugh. So, I'm going to walk you through how we came up with the idea, how we structure the ideas, because I think this is where the the biggest part where everyone kind of fails and crumbles is in this early ideation phase.

It's in that foundation period before you start your project. You need to have strong foundations so that you can build on top of it. If you don't have those, it's going to crumble, your project will crumble, and it won't end up looking exactly how you want. So, here is what we do. So, I will go to something like a simple as a Google Doc, and I will say, "Okay, for this idea, I came up with the core message." The core message in this video was that we're promoting a sun cream brand.

Now, in that, we then wrote out a list of things that could be funny about that. So, we were like, "Okay, uh you know, put rubbing sun cream into your mate's back. Uh maybe that would make some guys uncomfortable. Let's do something around that, and then let's have a character in there who is really up for putting sun cream on someone else's back. Uh what could we design that character? What could they look like?" So, we came up with this really fun concept.

Once we had like a little like group discussion with our team about ideation, and we had this rough idea, it was then time to start blocking the idea into real acting performance. And this is key. This is where most people will mess up. So, the acting performance here, you can see. For example, we'll start with me describing the camera shot that we want to open the scene. So, we'll say like a medium shot, or a close-up shot, or a wide shot, establishing shot.

And we'll we'll kind of write out that camera at the start so we can picture it in our head. Then, it will we'll kind of just dictate it as though I'm a director, right? And I'll say, "It starts with my friend taking a sip of beer, me applying sunscreen onto my arms." So, we're describing the actions that we want. We can hear the beach ambience, some seagulls, some general conversation ambience, the beach bar behind us.

So, really close your eyes when you have these ideas and start imagining in your head taking yourself into that environment. What would you hear? What would you see? Start picturing that acting performance from your characters in that specific environment and location. So then, this is where the acting performance comes in. It cuts to a side shot of me with my friend blurred in the background. My friend is blurred here, you can see that the focus is on me.

So, we're picturing that camera shot where you've got the focus on me and we kind of got this shallow depth of field and we've got my friend blurred in the background. I'm looking down and I ask my friend, "Do you mind doing my back for me?" I turn my body slightly and I point towards my back over my shoulder. So, you see that we're describing I'm picturing in my head as I'm talking, "How would I say that if I was actually doing it in real life?" Like, "Do you mind doing my back for me?" It's something like that and you have to describe all these actions for best results out of Seedance 2.5.

So, let's dive in a little further here. So, for example here, when I got this really great acting performance that you can see right here from this character here, which was super realistic, I specifically asked for this. I said, "Um he spits out his beer um that's in his mouth and he laughs obnoxiously, okay?" So, we need to use key describing words. So, he laughs obnoxiously for half a second. So, we're being very specific with timing cuz if you just said he laughs obnoxiously, he might laugh for like 2 seconds and we don't want that.

We want very like good pacing of the video. We want it to, you know, continue to be enjoyable and we want like cuts. We want it to be, you know, highly engaging. So, he also has an Australian accent. He says, "Fuck off, I'm not doing that." And then it cuts to a medium shot where we can see both of us looking at each other and there are people moving in the beach bar behind us. This is key. We want to add moving elements into the background of the scene.

We want to make sure that it feels realistic. You've got that realistic, you know, it's not just like two people sat there and it's like you're in a ghost town. We need to add in these details, otherwise Seedance won't do it. So, you've got to picture as though you're on set, you're in that location and you're filming it real, okay? Take yourself into that environment. So, I look disappointed with my friend and I plead.

You know, it's using these key descriptive words. So, saying that I'm pleading by saying, "Come on. It'll only take" then I get interrupted from a voice in the distance and we hear a lady shout from behind us, okay? So, you're like you're describing it. The key part is that you really need to be descriptive. And my script for a 30-second long project looks something like this, about a page and a half of writing, okay?

What I would do is I would then copy all of this and I would go into Claude and I would say, um, "I have an idea for a video, uh, using Seedance 2.5. So, make the video 30 seconds. I want to give you, I've got like the idea mapped out here. I'm going to give you the script. I want you to then write it and put it all into a prompt that is 30 seconds long, okay? So, we'll paste that in here and then I'll also paste in my script.

So, now we've got the entire script in here and we've also got my instructions. Now, what's also really important is that we give Claude all of the context of what the environments look like, the character sheets, the product sheets. So, let me show you what that looked like for us right now. So, first thing we designed is our character sheets and this is what a typical character sheet looks like. You've got your front, side, back, front close-up, top down view, height of the character and then we've got all of the details in here.

We've got, you know, the tattoos, the skin textures, the eye colors, the color palette of this character. Now, he was good, but he didn't he didn't it wasn't right for me. It didn't feel right. So, what I then did was I tried him with a hat. Still didn't feel right. Then I tried him with a kind of surfy beach blondie hair, and I eventually got to a character that I was happy with. Now, here is an extra bonus tip for you.

If you are going to create a character sheet, I'll leave a prompt right here that you can go ahead and copy to create your own character sheet, and you want to upload a reference image of your character, a single reference photo, and then you want to use GPT-2 as your image model. GPT-2 creates some of the most realistic characters with very good skin texture and detail. Nanobanana Pro gives you that plasticky skin texture look, so we don't want to do that.

And you can design all of your character sheets inside of Higgsfield. They've got all of the top image models, so you can do all of that right here inside of Higgsfield. So, once you've uploaded the reference image of your character, and you've got this prompt in here, it will give you a character sheet like this that you can now use inside of Seedance. Now, we replicated this process for all of the characters that we wanted to use inside of this scene.

So, we went and designed these characters, and we went and found references of similar characters that we like the look of, and you can use tools like Pinterest to go and find the design of certain characters that you like, and then go and restyle them in your own way. We obviously don't want to take anything because you don't own the copyright to that. So, you've got to go and remake your own, give it your own creative twist, your own creative spin on it, so that you can have the character that you like.

But, designing the characters and the locations is essential. So, here is my character sheet, for example. We then kind of took a reference image of my face, and we designed this character sheet here. We've got my height, we've got all the details. We've even got the details of the hat in here, and the shark print on the shorts. And you can see this was our original image that we created, which was the original idea behind this lady.

And we took this image as a reference, which we created with AI, and then we used that to create the character sheet and design the clothing that she was going to wear and everything all in our text prompt. So, now we have our character sheets locked in. It's now time for one of the most important parts, which is the location reference. Now, why is the location reference so important? It's because it locks in the color grade throughout the rest of your project.

And you want to make sure you've got a really nice color grade for your location. You don't just want to take a normal, overcast, boring, plain color graded image. You want to find something that looks good, that gives you the vibe. So, we created this image here. But how do we get to a nice location with a good color grade? Well, it all comes from fantastic references. So, you can use platforms any image platform like Pinterest, Cosmos, loads of different image platforms out there where you can find really cinematic frames.

Okay? So, this is an example. Beautiful color grading. You know, you might find a shot like this or like this for example, like this video here uh looks absolutely beautiful. You know, you might find a shot that you just absolutely love the color grading of and you want to design your location inside that environment. So, let's say I love the color grading on this one right here. What you can do is you can download that image and we can upload that inside of Hexfield in here uh right in here and we can say I like to design a nice sandy beach in the same environment with the same color grade, but I'd like there to be a wooden beach hut with a straw roof and a little sign that says love shack and a surfboard leaning up against a palm tree and there's got to be nice like deck chairs where people can sit on, but let's have no people in the scene, okay?

So, now we can go ahead and generate that. You can either use Nano Banana Pro or you can test out some of the other image models like SeaDream, which is also a super realistic one. Or we could even try Let's try GPT-2 as well. So, we're going to run a couple of generations in here, and then we're going to see which gives us the best result. So, while that's loading, what we can do is we can upload all of the image references that we have that we're going to use inside of this prompt.

So, for example, let's say we wanted to use this reference here, we can go ahead and upload that as well as all the characters that we're going to use inside of our video. So, let's go ahead and upload all three of our characters in here. So, we've got myself, my friend, and also the lady behind the bar. Now, we can go ahead and go back and have a look at our locations. So, now we've got some beautiful color-graded shots in here.

Some of them are looking a little AI, but they're not too bad. So, you can go ahead and do some refinement, I'd say. NanoBanana Pro has given us some pretty good ones. Uh Like this is really, really nice. I actually quite like this. Could do with maybe like a bit more of a bar kind of scene where there's, you know, some alcohol and like a bit more life in it. This feels pretty lifeless, but the color grade is there, right?

So, you can go ahead and refine this prompt until you get like a nice bar that you're happy with, for example. So, let's do that together. So, let's say we wanted to reuse our prompt in here with NanoBanana Pro. We can also add in Add in a some alcohol and a shelf of alcohol in the the love shack. We also would like a little bit more life and soul, so maybe add some like fairy lights, some party lights, a lot more seating.

We need more deck chairs, maybe like more like a couple of tables and chairs for people to sit at. We want more of like a community vibe, okay? So, we can add that into our prompt in here, and then we can go ahead and generate a couple variations of those. So, those are going to start loading in. And then, once you've got your location reference, so let's say for example, we just took the one that we just made, you can upload your location that you've got with the color grading that you're happy with locked in.

With all of this here, now you can go ahead and run it. And I know we haven't got our new location designed yet, but let's just say we ran this as a test subject. Now, Claude is going to go through, read through our original script, our idea, and it's also going to listen to our instructions that we want to create a 30-second prompt. Now, here is where the key thing kicks in, is that I have a C Dance skill that creates a prompt with perfect structure.

It time codes everything. And not only that, it also uses really nice camera presets, so that all of our generations come out looking like a professional grade camera. And it also uses the same camera lens, same camera body, same everything, same focal length. A lot of the shots remain the same. So, between all of the cuts of our scenes, every shot feels like it's been shot on the same camera. We have the same color grading, same characters, same details, same camera lens, body, everything remains consistent.

And that is so important when you're creating high-quality grade AI work. So, if you want access to our own custom C Dance skill, then you can get that in the description as well. So, here is our new environment. I would say the text up here looks pretty AI, but we're we're getting closer with the decking and the seating, and this is starting to look a little bit more realistic. It's all about creating that initial vibe that you want.

And this was the vibe that we chose for our actual project, but this one here also looks pretty good. I would probably take this into Photoshop and remove this and add my own text in that doesn't feel so like AI and perfect, but overall, I'd actually just remove this sign, and it would look really, really nice. I wouldn't remove this sign inside of Nano Banana, because each time you regenerate the image it's just going to reduce that quality and suck all that quality away.

But this is pretty good. So you could download that and go back into Claude. Now you can see in here what we've got here is my perfect prompt structure. So you can see we've got the global look and this is our prompt that we're designing for SeaDance 2.5, okay? Now you can see at the start of all my prompts I have the camera, right? I've got the lens, I've got all of the key terminology in here, the depth of field, the halation of the lighting.

I've got all of these like fine organic grain. I've got all of these key details that I can then use at the start of all of my prompts. And I have this custom baked into every single prompt that I do with SeaDance. It gives me the same high-quality results every single time. So you can also see that we've got it broken down into the cast. So SeaDance knows who the cast is. And then when you paste your prompt into any kind of platform that you want to create with SeaDance, you can actually tag those characters in that area.

So let me show you. So if we were to do SeaDance 2.5 here for example, in the prompt we can go ahead and remove that in here. We can paste this prompt in. Now you can see if we scroll down in here where we get to the characters, you can see here it says "Rogue, mid-30s, shoulder-length hair." So we can now go ahead and start uploading our elements in here. So we can upload our files and upload all of our character sheets in here.

Now while these are uploading, I'm going to show you why this is important because if you use, for example, SeaDance on other platforms like Dreena, you only get 5,000 characters that you can use. Whereas on Higgsfield, you can actually use a lot more characters. Why is this important? Because it allows you to get super descriptive and detailed with your prompts. So you can get all the custom acting performances that you want, so all the nuances, all the imperfections, you can get all of those with SeaDance.

Because if you're trying to do a 30-second long generation, you need as many characters as you can get to be as detailed as you possibly can be. Now here, check this out. So, if I click on this, we can upload that as a reference in here. We can upload my friend as a reference in here. And we can then go in and tag them. So, for example, Rogue, we can go in and tag so we can give that a reference. My mate in here, we can go in and tag and give that a reference.

And you can do that so forth with all of your prompts. So, you can go in and you can detail out the the lady as well. You can also detail the location and tag that as a as a reference as well. Um and you can see by giving Claude context of the characters and also the location, it can use those descriptive terms inside of the scenes. It's really really important that you give these as references to Claude to give you the best results.

Now you can see in here, it's going to detail out the style of shot. Is it a close-up? Is it a medium shot? Where is the camera positioned? Behind them? Is it a reveal? Okay, so we have all of these like nice camera terminology. You can also see we've got everything broken down into time codes. So, we know exactly how long each shot is going to be on scene for. So, you can also see in the prompt in here, we've still got that key terminology, "obnoxious half laugh for a second, then flat and certain." So, uh you know, we've got the detail of how we want the acting nuances to go ahead.

So, what we can do is just copy this, go straight into SeaDance, and then we can go ahead and start generating. So, down here you can see that you can adjust the duration of your video all the way to 30 seconds. You can adjust the aspect ratio. I like to do 16 by 9. And you can see here the quality we have, maximum is 720p. Now you can also switch on unlimited mode if you purchase that plan, and then you can start generating.

Now, while that's loading, I'm going to show you something else which is super cool. So, what you can do with Seed Ar 2.5 is you can create 3D movement, right, with custom tools like Blender, for example, where we create the camera motion and we dictate exactly the style of the shot that we want, and you can upload that as a reference. And Seed Ar 2.0 can actually adhere to that really, really well. And you can upload your own reference image of the location, for example.

You can also upload a character sheet of, you know, who you want to transfer onto that mannequin, and you can then create shots like this. I mean, just look how impressive this is. This is super, super impressive, and you have so much control now with this model. So, you can really lock in this exact camera motion you want. You can lock in the exact movement you want, all in 3D in pre-production, you know, so that you can then see maybe how it's going to look on set in the real thing.

So, the whole pre-production process can really, really be nailed down with this new tool. So, while that generation's loading, I just want to walk you through my actual editing timeline. So, you can see in here, we've got our, you know, core videos in here, and we've got our sound underneath, and you can see that we really don't have too many layers here. The color grading is great out of the model, so we don't really need to do any tweaks, but you can also add your own Lumetri color effects on it if you wanted to to adjust the color grading.

But, let's take a look, right? So, if I just mute all of this audio, so we can just have a little look-see together. You can see in here, um I've just added my own subtitles at the bottom, and we've kind of edited the 30-second generation down a little bit to cut out any of those like awkward silences, so that we can really lock in all of the right timing that we want. Cuz a lot of the time, when you do the generation, there might be like an extra awkward second of silence that you might just want to cut out.

So, we'll go in and make those tweaks manually, but what I really like doing is running three generations of 30 seconds and then mining out the best parts of each of those and taking the best parts of each and then clipping that all together until you have a really nice 30 second sequence. So, not all of the time you're going to get perfection straight out the model. Obviously, if you're really really talented, you might get super lucky on your first go, but that's a bit of gambling.

So, what we like to do is do one generation, test to see how it how it looks, how is the acting performance looking? Are we getting the results that we want? And then if it's not there, then we go back, we refine the prompt with Claude. We say, "Okay, this is cool. This is, you know, the acting performance here is good. Maybe here it's not so good. Let's change this. Let's change that." And we'll refine. Then you go back, you paste that prompt in and then you redo it again.

You see the acting performance. Is it good? Yes, I'm happy. Okay, now we're going to use this three times. We're going to take the best parts of each of the prompts and we're going to put that together to build something beautiful. So, that's really like how we like to work. So, you might take like the best parts of of each part and kind of stitch it all together. It's probably important to mention as well, the only real editing that we actually did in here was adding in our own music track to this scene right here to really emphasize all of the hat like a comedic value.

So, that's really one of the only editing parts that we did in this video. So, here is our generation right here. Do you mind doing my back? [ __ ] off. I'm not doing that, mate. Come on. It'll only take >> Go on then. Can you do both of us? Protects your skin, not your dignity. >> Now, it's not bad to begin with. Now, what you can do is go back with Claude or refine and constantly edit until you get the exact results you want.

And you can just talk with the AI and just say, "I want to change this. I want to change this." And you can keep massaging it until you get an output that you're happy with. But, make sure once you've got to a position that you're happy, you then run three generations and take the best parts from each one, stitch it together. You still got to do a little bit of manual tweaking in there to elevate it. Now, to make it even better, I would say you want to add in your own subtle color grading with Lumetri color effects, maybe a bit of film grain in there as well, to really take it to that next level.

So, it doesn't feel like something that's super polished that's come straight out of an AI video model. So, honestly, creating really high-quality work has never been easier. And right now, behind the scenes, we're actually working on one of the most incredible automations that can automate the entire process of creating cinematic ads, which sounds ridiculous to say, but we're going to be releasing something on that in the next week.

So, make sure you stick around for that one. Also, if you're interested in learning more about creative AI and actually turning it into your full-time career, then you can come and join Gen HQ. We've generated over a million dollars for our students so far, and I'm hoping that you can be the next one to find a full-time job or even start getting paid by your first AI client. And we teach you all of those skills in Gen HQ, not just the tools, but actually how to build a career in this space.

So, if you're interested, there's a link in the description as well. You can come and join us over there. We're one of the most successful communities in the world for creative AI. And that's everything you need to know about Seedorn's 2.5. So, if you want to try out unlimited for the next 33 days, you can click the first link in the description. And if you're interested in learning more about creative AI and turning it into your full-time career, then you can check out Gen HQ.

Thank you very much for watching, guys, and I'll see you in the [music] next one.