Seedance 2.5 is an ABSOLUTE MONSTER - Master it in 20 minutes

0:04 Dan! Dan, have you HEARD IT YET? WHERE THE HELL ARE YOU? he's space walking again. Okay. Let me just update his software instead. >> You got 30 SECONDS TO MOVE.

0:57 >> SORRY. OKAY, I UNDERSTOOD. SEEDANCE 2.5 IS INSANE! >> Seedance 2.5 is now available to

1:22 everyone. And in this video, I will teach you everything I know about it. Over the last week, I've spent over $2,000 together with my team to try and master Seedance. So, please watch this entire video to the end so you don't end up wasting credits like I did. If you want to follow along and if you want to

1:38 try this out yourself, I will leave a link to Higgsfield in the description down below. They were kind enough to sponsor this video, and they now have Seedance available on their platform alongside with Flux and Halo, which are also two other good models that just came out. Now, the first big change that

1:52 Seedance 2.5 has is that it can do 30-second generations. And this is huge, but it also becomes a lot more expensive because of it. So, you need to use your credits wisely. And that all comes with prompting responsibility. Now, before you start generating, let me first explain the new prompting method on how

2:11 to get the most out of your generation. So, I'm going to show you two different methods. One is the more simple one, and the other one is the more advanced one. Keep in mind, I will share all of the assets, everything that I share, all the prompts, all the assets that I used in this video in a detailed document in the

2:26 free school community that you can find in the link in the description down below. Now, the first format is very simple. I call this the simple prompt format. You let C-DANCE fill out the details. Basically, you want to add in a bit of context about your references. So, this is the reference window. Here,

2:42 you're going to say like, "Image one is, for example, a guy in his mid-20s. He is this, this, and this." "Image two is your other subject." Then, "Audio one is a voice reference for your character. You want to match that voice with the dialogue." Then, you want to give some scene context, where we are, who's

2:57 there, and is this a long continuous take? You can add in some technical details like the lens, the depth of field. You can add grain, etc. Then, following up, you have the dialogue language. For example, I have a slightly Dutch accent. If I want to include it, I can say like, "A male Dutch accent. He

3:12 is about 25 years old." Something like that. Following up, we're describing our scene. So, we're describing exactly what we want to happen at cut one, at cut two, at cut three. This is a very simplified timeline prompting not including any like, um, "At this second, this happens. At that second, that

3:30 happens." You just include what you want to have happen in that scene, no matter the length. And you're just going to describe it like you are a director. If you have one long continuous take, then you have no cuts, and then you need to describe what's happening. You describe the shot plus all of the dialogue. And

3:45 lastly, I have this part that is the keep the following things consistent. And that's just like a prompt to just dial it in. So, for example, if we put this into action, I have this example right here of my asset is this young Japanese woman and I want her to do a handheld vlog on a date in Japan. So, it

4:04 looks as follows. Our prompting reference is very simple. We just have add image one. It's the look reference, the Japanese woman. Then we have the scene context and I'm going to describe on a date in Japan early morning into night. Then here I'm adding in some technical details and basically here

4:19 we're going to give or idea. So, we're going to describe like we are the man behind the frame holding the camera. The woman is excited like we adding in all of our different context. You can pause the screen or you can read my prompt in the school community. Here we're going to add in all of these different

4:32 details. Then the more context about the scene is that I want to show the whole date like how the whole day went and then lastly I'm adding in just the whole idea of what I want [music] in here. It's a very simple prompt. So, we're going to take this prompt. We're going to go over to Hixol. Here you want to

4:46 switch over to Cdance 2.5. Then you want to drop in your prompt and then you want to add in your images as we don't have the image right now. Then you want to attach the image right here and you want to make sure that this one is linked. Now, you can go up to 30 seconds right here and you can only do 720p for

5:02 [music] now. I'm hoping and I'm expecting that maybe in a few weeks time we will also have 1080p. And then you want to click generate and then you get a result like this.

5:42 Okay, so that was the simple prompting style and that already looked really really good. Like I would have a few small remarks, but that could be fixed in editing. But now I'm going to show you the second advanced prompting technique. And this is the one I use mostly throughout some of the advanced

5:57 prompts that I did. So here we are going to use the following format. Here we have the goal. We're basically describing what we want to have happening in this video, like when this video is kind of like a success. I'm including the duration, but you don't have to do that because you can already

6:10 toggle that in the generator. Then we have the references. Here we are describing once again what each reference does. And this is very important because you can add up to so many different references in there. You can add up up to 50 references, 30 images. So that is quite a lot. So you

6:26 want to describe them. And you want to group them together if you have references of the same people or of the same thing, but maybe different angles. More on that later. Then I'm adding in the continuity. So character A, what they look like and what they wear the whole way through. Character B, same

6:41 thing. Then the scene. I'm describing what the location is like. Then the stages. And here we have basically the different shots that are inside. So stage one happens at, let's say, second three. Then I'm going to give that segment a name. And then I'm going to describe what's happening and how it is

6:57 framed. Then I do the same thing for the second stage, the third stage. Like depending on how many stages you have, you can add that in. If it still is confusing, I will show you an example in a second. You do that for each of your different shots that you include in there. And then you have described in

7:11 pure detail what you want to see. And here you need to think once again as a director. The process here is a lot more complex and a lot more time consuming than just using this like a slot machine. You don't want to do that anymore. You just want to be aware because you're spending good amount of

7:25 money on each generation. So you want to think it through and you want to be mindful about how you prompt. Then we have the look. Here we're describing the sharpness and the format. For example, the format clarity on a handheld phone, the color and the grain, the lighting, the shadow, and the mood. Then we're

7:40 including the camera and the performance. So I want to have fast and punchy cutting or if you want to have slow cutting, then you're describing how the character should carry themselves. Um you can also include something about the audio, what you want to have happen in there. Uh so, do you want to have

7:54 music? What kind of style? Do you have want to have sound? Do you want to have the ambience, subtitles? All of that you can include that in your prompt. Lastly, I'm going to include two rules that is keep consistent and I'm going to say like same character A, same character B. You want to customize this according to

8:10 your needs so that these things don't mess up. Then you can also say do not include brand logos, extra people, text on screen, or anything else that you don't want to have in the frame. What that prompt looks like is as follows. Now, here is a quick example of what a prompt like that could look like. I have

8:25 the goal of a cinematic chase sequence shot with large format IMAX clarity. Then I want to have continuity. I'm going to describe what my subject is like, the creature, all of those different things. Then the stage, what's going to happen in the first 4 seconds, then stage two, that is the approaching

8:41 the edge, then the camera stays close. You want to think like a director here. You have full control over what you want to see. But you got to be mindful again that what you're putting in there needs to make sense. So, you're doing that for all of the stages. Then you can include

8:54 your visual style. Again, you can pause this if you want to read all of this. You can do the camera performance, the audio, the exclusions, and then we're also going to say maintain consistency. So, that is one example of what you can make. Now, with that prompt I made this shot.

9:23 Moving on, you got to learn how to use references. With 50 references, it has become more important to be detailed and correct about them than it was before. Because up until before you could maybe add like 14 or 16 references, but now it is as follows. You can do 30 images, 10 videos, and 10 audio clips. If that

9:43 sounds confused you, let me explain to you. So, regardless of anything, you can always add 30 images. For the videos, you can have 10 videos that are 30 seconds combined in terms of length. So, that means that you could have one video that is 21 seconds long, while the other nine videos are 1 second long. Like

10:00 together, that is 30 seconds. Then for the audio, the same thing is applied. You can only do audio until 30 seconds long. And keep in mind, this is per prompt or per request. Would I recommend using that many references? No. Would C that recommend it? No. Like this is what I found on C that's pitch. They want you

10:18 to put one to eight distinct subjects across subject reference images. If that sounds vague to you, that is because it is. The best way to describe this is instead of doing one image of your character, which is a character sheet, where you have like three different angles, you can now, because you have so

10:35 many different references that you can put in there. So, for example, I am a character, I can have four different angles of my face, then I have four different angles of what my body looks like, and then we're adding that in as one distinct subject. So, I am one distinct subject. By doing that, you can

10:52 be super detailed because you're giving a lot of context on what your subject looks like. And that is the thing that we need to change to what we were doing. We were using a lot of character sheets. I would not recommend using character sheets now. And that's like you will see a few character sheets still throughout

11:07 that video, because it still works, but because you have so much context to give now, you can be more expressive with that. So, you can include more details. For the videos, you can apply the same thing. You can add one to five distinct subjects and five to 10 seconds per subject. So, it all depends on how many

11:23 subjects you have in your video. Now, for the audio, keep only dialogue, voice characteristics, ambiance, or music direct relevant to the task. So, you want to separate it. You can do separate audio tracks. Don't include like music and voice in the same track. Those are the things you want to keep in mind when

11:39 using so many different references. So, now let me explain to you the best way to make references to make videos like this one, for example. So, this one has 30 different images that have been used to create all of these different characters. Now, that video was quite good already, but before I show you more

11:56 examples, let me first show you how to make your own references. Because making good references equals good output. For this reference, I'm giving you this template. You want to do your image type, your subject, your design, your pose, your environment, the camera, and things to exclude, and lastly the image

12:15 type. Now, the best way to make your subject is simply through GPT Image 2 or Nano Banana Pro. So, let's use Nano Banana Pro for this example. We're just going to include or prompt, and I want to make a trench coat. So, the subject is a extremely long oversized trench coat. I'm going to include the colors of

12:31 it, the pose of it, the environment, which is a pure white seamless studio background. That's how I would recommend using this. Camera and exclusions. For the size, you don't need to have this in 16x9. So, I'm just using 4x5. So, I can capture the complete thing. Then you're going to hit generate. And then we have

12:48 our first asset like this. You want to do this for all of your different assets. So, for example, I have done that for all of these different objects, and then you can also do this for your main character. Then for main subject, I'm this case using a character sheet. This is the character sheet format that

13:05 we're all used to. You make an image of your main character, and you ask Claude or ChatGPT to put this in a three-sided or like two-sided view to get a character sheet, and then you're left with something like this. This can be done in both Nano Banana Pro and GPT Image 2. For characters, I might

13:21 actually recommend GPT Image 2 because it might look a little bit more realistic here. So, that is how you can then generate your references. And now, we have to write a prompt putting everything together. Again, here I'm using the advanced method. You want to describe in your goal basically my idea

13:38 was that two guys show some like old object in a case and we have the reference material. Here you can see how I do this. So, I have a bunch of different references and I'm describing what each is. So, image two is headphones, image three is a golden necklace, image four is a tennis ball,

13:54 etc., etc. It's very simple. The other thing I did, which I wanted to find out, is I wanted to include an audio reference. So, I'm going to include a funny audio for you guys. So, stay tuned for that. Uh then for the continuity, I'm basically describing like what's going on. So, we have the vendor and we

14:09 have the customer. Then up until he removes them in stage 13, he's bare-eyed after. Then I have the setting, location, all of that I included. And then we're going to describe our prompt. Now, this one was quite complicated because I wanted to have a lot happen with a lot of different zooms and a lot

14:25 of different cuts. Um so, I had to describe it really well in what's happening. Again, I would recommend pausing this to analyze this so you can go through the prompt yourself and you fully understand what I do. But, once you do something like that and you hit generate, then this is a shot that you

14:41 get from that. >> I have exactly what you need.

15:08 >> That's a lot of stuff right here. >> So, yeah. You can generate some weird scenes with all of these different references that you can include. I'm hoping you have fun with it. I sure had when adding this voiceover in there. Now, one of the things Cidens 2.5 excels at is realism, and I want to point this

15:27 out by showing you this clip. >> [screaming] [cheering] >> Thank you. >> [screaming] >> I made that generation using the simple prompting formula. Here you can take a look at it, and it's not that complicated. It's basically a character sheet of me as some kind of like singer.

15:52 Then I have the context added. There's a massive crowd. It's being handheld recorded. A realistic lens bleeding, natural camera shake. We're using realistic live action. It's a concert footage, and we're adding like a lot of different details on what I want this to look and feel like. And that is how, in

16:10 my opinion, you can translate best to the AI what you want. And keep in mind, you can just use MD file for that to help you write it and get the exact look and feel that you want. Now, when I saw that shot, it made me so curious that I had to compare that to Cidens 2 and Kling. And here are the results, so you

16:30 can have a look for yourself. >> [screaming] [cheering] >> Thank you. >> [screaming] [music]

16:55 [cheering] >> Thank you. Thank you. >> [cheering] >> Now, if on all three we paused a frame on the girl that we're panning to, then you can see that the quality of C-DANCE 2.5 is the best. Even though it's 720p,

17:22 it just looks the most realistic. The other ones [music] looks a bit plasticky. It just doesn't look as good as the 720p version of C-DANCE 2.5. And the other thing to point out is the details on my skin, like the freckles that I have, and all the detail in my face is quite good on C-DANCE 2.5. So,

17:39 I'm very curious to what we will have when we finally get 1080p and 4K because now you still have to upscale this to get this in a better version. One of the most common ways to use AI and tools like C-DANCE 2.5 is UGC type videos. And with 2.5, it's a lot better if you know how to prompt it correctly. Let's say I

18:00 want to do a ad for climbing shoes. This is the shoe that I have in mind. We can generate multiple different angles of that shoe with a easy prompting format. So, with one simple reference, I created these six different angles of my shoe. Now, we need to do the same thing over UGC character. In this case, that will

18:20 be me. So, I've added in this selfie of me, and then I prompted it to get four different angles of my upper body, basically my face. So, I have like two front shots, I have one side shot, and I have one shot of what I look like from the back. After you generate all of the assets of the product, in this case the

18:38 shoes, of your character, in this case me, you can also include your audio. So, I recorded a short segment that is unrelated. I basically used a previous recording of me as a reference of what my voice sounds like. And it can be a very short recording. Then you add that also into Claude using the MD file, and

18:58 then Claude will use that MD file, and it will help you write a prompt based on your idea, and it will tag all of [music] the different references. You can then copy that idea. You can put this into Higgsfield in Journeyman 2.5, upload all of your references and then you hit generate and then you get a

19:16 video like this. So, these red bouldering shoes have been going viral and you guys know me, I'm serious when it comes to these things, so I'm going to test these and give you my honest opinion.

19:41 >> [screaming] >> Okay, these red shoes are pretty legit. In terms of the voice, I was a bit skeptical because I only added in like a 16-second reference of what my voice sounds like and I think it has done a decent job on getting it similar to what I sound like. Another thing that I want

19:58 to point out, which you want to avoid if you want to save your credits, is creating long videos. And by long videos, I mean videos that are 180 seconds long. On Cidence 2.5, you have the option to generate longer generations, but basically what it is, it is six 30-second videos stitched

20:17 together and that's all based on one prompt. And one prompt is only like 5,000 characters, so there's not a lot of detail you can add into like a 3-minute video. So, I would recommend not using that. The best thing that you can do if you want to have one continuous scene or if you want to have

20:31 everything matched together is you keep using your previous video as a reference and then you describe what you want to have happen next. This also allows you to introduce new references, so new images or new subjects or new objects [music] that you want to add in there and then it connects them seamlessly together.

20:48 You can also include video one, video two and you can prompt it that these videos need to be combined in a seamless manner and that makes Cidence 2.5 understand what you want and it will stitch them together quite good. Another feature that you want to check out is the editing capabilities of Cidence 2.5

21:06 because right now I would say Cidens 2.5 has the best editor I've seen across all of the different models that I've tried so far. If I have a shot like this one, for example, me sitting in the cinema. I look quite lonely there. So, if I were to prompt Cidens 2.0, so not 2.5, to add

21:25 in people in there, I get something like this. It looks okay, but it's not quite there. [music] If I do the same prompt, but I do this into Cidens 2.5, then I get something like this. With 2.5, everything looks a lot better. Like the people fit better, the lighting looks good, it all matches. It's not like it's

21:43 edited in there. The other thing that I want to point out is that you now no longer need to have video that is perfect because you can edit it. Basically, if you want to change a bag of chips into a bag of popcorn, you can just prompt it and it will keep everything else the same, it will just

21:59 change that object. If you want to change the color of something like a door, you can prompt it and now that thing has been changed. You can add in or remove a background music track just by asking it. So, these are all kind of different things that you got to explore, that you got to try out, and

22:15 you don't got to give up on your first generation because now you can actually make them good. The last technique that you can apply to Cidens is to use your previous generations, cut out scenes of the parts that you like, and then repurpose that. This is a tip I found on X. I will do my best to link this as

22:32 well in the document so you can go through it yourself because it is super useful because what you do is you cut out the perfect scene or the frame that you want to use, and you regenerate it using that new frame that already has the good look and feel to it, and then you can just prompt it to do it better

22:50 than what you did originally. So, by applying all these different techniques, you're hopefully not only saving credits, but you're also having a lot better generations. Now, if you want to watch my full filmmaking guide, then either click the video that's on the screen, or if it's not out yet, then

23:05 make sure that you subscribe, and I will see you in the next video.