How to Create AI Anime with Seedance 2.0 (You're in Full Control)

Seance 2.0 just came out and it's the first AI tool that can make anime that actually looks good. I've been testing it for the past week and I've been able to create full anime scenes in less than 15 minutes. Just take a look at this. In this video, I'll walk you through the exact workflow I'm using so you can copy it and start creating your own anime scenes today, even if you've never touched an AI tool before. Now, there are two popular methods you can use with any AI video generator. text to video and image to video. The first one is as simple as it sounds. You write a prompt and get a video back. However, image to video takes an image as input and builds an entire video on top of it. Now, Cedense 2.0 supports both of them. And with text to video, you can get decent results really fast. But if you want results like the ones you just saw in the intro, you need to use image to video instead. And the reason for that comes down to three things. First, it saves you a lot of wasted credits. with text to video. You describe your character in the prompt, generate, and if it doesn't look quite right, you adjust the description and regenerate. Then you repeat over and over again. So, you're burning through your credits trying to get it right. Instead, you can build one good image with all the details you want up front and then generate the perfect video on top of it from the first try. Second, it keeps every detail consistent across all your scenes. And that's only possible because Seance is always looking at the same reference images right before every new generation. So you don't have to worry about weird changes between clips anymore. And third, it gives you full control before you even hit generate. If something about your character or location doesn't look right, you can simply edit the reference image first and the final video will match exactly that. So this is exactly why we always want to use references for our anime generation. But before we can actually use them, we first need to build them. For this, I found one simple method that does wonders every time. So let me show you the only two steps you actually need to make this work. First, you want to describe your character or location in plain natural language, exactly like you'd explain it to a friend. This is where you include all the details you want the AI to add to the video. How the character looks, what the environment feels like, or anything else that matters for the final result. But you do not need to worry about making it sound like a polished AI prompt. Just write it the way it comes to you. This is exactly what I wrote for my character. Now, once you have that rough description, the second step is to take it into Claude and ask it to rewrite everything into a proper prompt for the exact model you're using. And in our case, that's going to be Nano Banana Pro because right now, it's one of the best image models. Claude is really good for this task because it understands how these models work and how to structure the prompt to get the best possible result. So, here's the prompt it gave me back for my character. Now, let's actually test it inside Nano Banana Pro. For this, I'm going to use Higsfield because it gives me access to all the models we need for this workflow, including Cedense 2.0. So, if you want to follow the same process, I'll leave a link in the description below. Once you're inside, go to the image section and click on create image. From here, select Nano Banana Pro as the model and paste in the prompt Claude gave you. Then, just click generate and wait. And here's the result we get back. Honestly, there's nothing I can really complain about. The character has all the details I asked for and it genuinely looks like anime. The Giblly style feels spot-on and it even has that soft hand painted look. Because this is the image Sea Dance will use as the reference every single time, it's worth getting it exactly how you want before moving on. Now, let's do the exact same thing for the locations. The first one is going to be a mountain lake at dawn. So again, I'll write out all the details I want in normal language. Paste that into Claude and let it turn everything into a polished prompt. Once I have that, I go back to Higsfield, select Nano Banana Pro, and generate the image. Look at this. It looks beautiful. The transparency of the water came out so well that you can actually see the stones underneath it. I really like how this turned out. So now I'll move on to create the second location for our anime. This one is going to be a lakeside cabin. And again, I use the exact same process as before. And here's the image we get back. It looks warm and cozy, which is exactly the feeling I want this anime to have. And with that, we now have all three reference images ready. However, if you don't pay close attention to the next step, you can very easily waste all the work we've done so far. And that's because the reference images alone won't get you amazing results. To get those, you need to know how to properly write prompts inside Cedance. And because this model is quite different from any AI we've seen so far, I'm going to show you my personal prompting framework. It's called Macall, and it's fully customized for the new Seedance 2.0. So, if you want to follow along with the tutorial and test it yourself, I'll leave a link to Higsfield in the description below. All right. So, the MCL framework is made up of five different elements. The M stands for multi-shot, and this is easily the most important one. That's because seance 2.0 only takes a single prompt per generation. So, if you describe the whole scene like one continuous flow, the model has no idea where one shot ends and the next one begins. And what you get back feels more like an anime screen saver, than a real sequence. Think about it. Any real anime scene you've ever seen, even a slow one, is never just one continuous shot. It's always a sequence of moments pieced together. So instead, you want to break those 15 seconds into three to six separate shots inside the prompt. Each shot becomes its own beat, almost like a storyboard the model can follow frame by frame. For our first scene, I want the boy waking up on the lake. So I'll split that into three shots. An aerial view of the lake, a close-up beside the boat as he stirs, and a medium shot as he sits up and smiles. So the prompt for our first scene will look like this. Now, inside every shot, you also need to say exactly what the character is doing. And this is where some people really mess it up. They describe the setting in perfect detail but barely mention what's actually happening in the frame. So sea dance either keeps the character completely still or invents something random. And you don't want either of those. That's why the A in McCall stands for action. For shot two, that's the boy lying asleep inside the boat. Then he slowly stirs, opens his eyes, and squints into the soft morning light. Every verb there is a cue for the model. And the more specific you are, the less room seance has to guess. Now take a look at this line in shot one. The camera slowly descends toward the boat, sunbeams sparkling on the water. This is exactly the sea from our framework. It's the camera movement, and this is what makes good-looking anime actually become cinematic. So, for our first shot, I want that aerial view slowly descending toward the rowboat. For the second, a low angle close shot beside the boat. And for the third, a medium shot that holds on the character as he sits up. So, I'll get three different angles across 15 seconds. And that alone already makes the entire sequence feel more dynamic. But now we get to one of the most underused parts of the entire framework, audio. You see, even some so-called experts complicate this a lot with sound overs. But Cance 2.0 can actually generate audio directly inside the clip. That's what the second A stands for, audio. And for anime, sound is just as important as the visuals. If you don't believe me, try to watch an anime on mute, and you'll quickly realize it's not the same anymore. So, at the bottom of our prompt, I'll add a separate audio section and list exactly what I want to hear. For our first scene, that's gentle water lapping, distant bird calls, and a few more. The key is to tie the sound to specific moments in the shot, not just describe vague background noise. Now, the last letter in the framework is L, which stands for lighting. Giblly style scenes depend heavily on lighting because that's where so much of the mood comes from. But when you describe it properly, the model locks into that exact palette and holds it across every frame. So, for our opening shot, I'm going to layer it in clear with a golden dawn sky and sunbeams sparkling on the water. And if you want to get the best results, you want to add this lighting note to every shot from your scene. Now, here's the part that makes the whole framework click. When you stack all five letters together, you end up with a prompt that tells seed dance exactly how to stage the scene, move the camera, direct the character, paint the frame, and score this sound. Once you put together all five elements of the Macall framework, you'll get a prompt that looks like this. So, now I'll go to Higsfield inside the video sections and select the seedance 2.0 model. Next, I'll click on upload media to add the reference images I've created previously. But before you do that, make sure your reference is eligible. This is a safety feature from Seedance that makes sure you don't upload the wrong kind of content. So, let's upload the character image and the location one. Then, I'll set the duration to 15 seconds, paste in the prompt, and hit generate. Honestly, this turned out way better than I expected. The camera movement flows naturally from the aerial view into the close-up, and the Giblly lighting stays consistent through the entire sequence. But what impressed me the most is the audio. You can hear the water lapping in perfect sync with the movement, and it adds so much to the realism. So, with the Macall framework, you can now create a highquality scene for your anime. However, we just generated the first scene. There are a couple more to go and then it will come to the most important part of the entire video. So, let's first create the remaining scenes. The second scene is where the boy catches a fish with a net. And here's the good news. You can use the exact same process we just went through because sea dance works the same way for every scene. The only thing that changes is the content of the prompt itself. So again, I'll go through each letter of McCall and build out the shots, the action, the camera movement, the audio, and the lighting. And here's the full prompt I came up with for this one. The reference images stay the same. the boy and the mountain lake because this scene still takes place on the water. So, I'll paste the prompt in, set the duration to 15 seconds, and hit generate. In my opinion, the result is really solid. And what I like best is his expression when he caught the fish. But now, let's get to the third scene. This is where the boy goes back to shore and walks up to his cabin. And here's the one key change from before. Because this scene now includes the cabin, I'll also add the second location image as reference. So, we go from two reference images to three. And Cance will use all of them to keep everything consistent with what we've already built. Then, it's the same exact process. Write out the McCall prompt, paste it in, and hit generate. Here's the prompt for this scene. And here's what we get back. Honestly, this one turned out even better than the first scene we generated. The motion feels completely natural, and the Giblly style still holds up beautifully across every shot. And the last scene is the grilling scene where the boy cooks the fish over the fire. For this one, there's another small change. Since the action now happens back at the cabin and the lake isn't in frame anymore, I'll remove the first location image from the reference list. So, the final setup uses just the boy and the cabin references. And here's the prompt I built for this scene. Same process again. Set the duration, paste in the prompt, and hit generate. Take a look at it. Each scene looks amazing, but right now there are four separate clips. We need to somehow stitch all of them together into one single video. But this step is quite simple compared to what we've done so far. Let's jump inside Cap Cut and drag all four scenes straight onto the timeline in order. From here, you can trim anything that you don't like or add quick transitions between the clips. But for my video, everything already flows pretty well, so I'll just hit export. Now, let's see the final result. In my opinion, this is genuinely one of the most impressive AI anime videos I've seen in the last period. The style holds throughout the scenes, the motion feels natural, and it actually looks like a Studio Giblly short. But all of that is thanks to the new Sea Dance 2.0. So, if you want to start creating your own AI anime, sign into Higsfield with the link in the description and get full access to Seedance Nano Banana Pro and all the features we've used today. Thanks for watching and see you in the next one.