Seedance 2.5 Just Got a Load of Awesome New Features!

Author: Theoretically Media · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Seedance 2.5 Just Got a Load of Awesome New Features! · Channel: Theoretically Media · 19:44 · Watch on YouTube

Topics: Task and workflow automation, AI content generation, Multimodal models

Summary

Dreamina has updated its platform with an improved canvas workflow, a revamped AI assistant, and an Elements folder for reusable characters and voices. The video tests canvas-based planning and generation, finding that it helps organize assets and shots but that manual prompting can still produce better results. It also covers new Dreamina generation skills and a Seedance 2.5 preview mode that lets users test prompts at 480p before re-rendering selected clips at 1080p. The video includes a sponsored demonstration of GenSpark Design and closes with a brief look at BytePlus Dramagic and rumors of a ByteDance world model.

Key points

Dreamina has significantly improved its canvas and AI assistant, while retaining the option to use the platform without canvas workflows.
The canvas assistant can use models including GPT Image 2, Nano Banana Pro, Seed Audio 1.0, Seedance models, and MiniMax H3.
Canvas planning can organize assets, storyboards, and generation plans, but the host recommends taking manual control when writing prompts for shots.
A new Elements folder lets users save characters, add custom voices and alternate angles, and call them into prompts with an @ mention.
PJ Ace’s Skyline Heist canvas is presented as a useful example of character sheets, production notes, and generated-video workflows.
The sponsored GenSpark Design demonstration shows editable posters, social-media assets, short videos, and a web presentation created through conversation.
Dreamina’s generation skills include options for iconic film styles, cinematic long takes, film storyboards, and micro expressions.
Seedance 2.5 Preview offers lower-cost 480p generations for prompt testing, then can re-render a selected clip at 1080p rather than upscale it.
BytePlus Dramagic is described as a beta enterprise platform for script analysis, storyboarding, and video review; the host also mentions rumors of a ByteDance world model.

Chapters

Start	Chapter	Summary
00:00	Dreamina’s canvas and assistant updates	The host introduces Dreamina’s upgraded canvas workflow, improved AI assistant, and the models available within it.
02:02	Testing canvas planning and video generation	The host demonstrates a character scene, then tests the assistant’s asset preparation, shot planning, and storyboard-inspired generation.
06:55	Editing results and reusable Elements	The host reviews generation issues, recommends manual prompt control for shots, and demonstrates the new Elements folder.
09:12	PJ Ace’s canvas workflow	The host explores Skyline Heist’s production board as a source of ideas for character sheets, notes, and video-generation methods.
11:20	GenSpark Design demonstration	A sponsored segment shows how GenSpark Design can create and edit a poster, social assets, a short video, and a web presentation.
15:26	Dreamina skills and Seedance 2.5 Preview	The host reviews Dreamina’s new generation skills and tests low-cost 480p Seedance 2.5 previews with a 1080p re-render option.
18:40	Upcoming ByteDance tools and closing	The host discusses the expected Preview release, BytePlus Dramagic’s beta, and rumors of a ByteDance world model.

Quotes

they have finally introduced their own murder board. — 00:30

when it comes time to generating your shot, yeah, again, just go in and and manually take care of that — 07:57

we now have an elements folder. — 08:41

we now have SDNS 2.5 for preview. — 16:57

This is not upscaling it. This is actually a full re-render. — 18:09

Concepts and entities

Dreamina, Seedance 2.5, ByteDance, Genspark Design, PJ Ace, Skyline Heist, BytePlus, Dramagic, GPT Image 2, Nano Banana Pro, MiniMax H3

Full transcript (original language)

00:00 Big news from ByteDance and Dreamina on the SeaDance front. It is not SeaDance 3.0, just to be clear. But Dreamina has rolled out some long-awaited SeaDance features and some I actually really big quality of life features to their platform. Uh and today we’re going to take a look at those. Plus there’s a new feature that’s currently in preview, but it’s one that I and you have been waiting for. All right, let’s go out dancing. Kicking off, Dreamina have unveiled a uh well, there’s a pretty big update to their platform. Dreamina of course being

00:30 the western home to SeaDance both 2.0 and 2.5. Uh and like to be fair, this is a pretty long overdue update. Uh first and maybe biggest, they have finally introduced their own murder board. Now admittedly, this isn’t entirely new. They have had their canvas feature up for a while. They’ve just made some massive improvements to it. And look, I know that no base workflows are not everyone’s cup of tea. I I do want to know that you can still do it like the old way, but this is here if you want. Alongside that, they have revamped the AI assistant over here. Uh basically,

01:02 you know, the the chatbot in the canvas system. Uh to be fair, once again, the canvas system did have the old AI assistant in here as well. It’s just, you know, early on, uh in my opinion, it it just wasn’t very good. It was it was actually pretty stupid. To their credit, Dreamina does seem to have up their game on this front. I wouldn’t call it perfect, but it’s definitely drastically improved. And there actually there are some surprises here as well. Uh for example, if you want to generate images on the canvas, uh you’re not just limited to the seed uh dream models. Uh

01:32 you actually can use uh GPT image 2 and nano banana pro. No sign of image 2.5 yet. That that would be a nice addition. If you want to generate audio, of course, uh seed audio 1.0 is here. Um we’re going to take a look at some of the other stuff in here, too, cuz there’s some cool stuff in here. But uh on the video front, of course, there is, you know, uh all of our various uh seed dances. But uh surprisingly, actually, this one really surprised me. Uh, Mini Max H3 is available here as well. So, just taking it out for a quick spin, I

02:02 took our PI office along with our character of Tuesday Sinclair. Uh, and then usually we have uh, Malloy in here, but in this case I just recast him as James uh, from uh, Death Walks into a Bar. Um, again, you know, just all three nodes going into this and then using the chat system, uh, I essentially just gave it the transcript of the first version of this scene uh, and said like like let’s let’s redo this with these characters. Uh, this is what we got. >> Mr. James?

02:33 >> Can I help you? >> I hear you’re a detective and I am in need of one. >> Detective? No. Gumshoe, maybe. Got a pretty good nose. So, what’s the case? >> It’s a murder. But the thing is, it hasn’t happened yet. >> You don’t say. Who’s getting iced? >> Well, Mr. James, you are. >> Well, that’s certainly a way to get my attention. I didn’t catch your name. >> Tuesday. Tuesday Sinclair. >> If I hadn’t taken this case, maybe I would have lived to see Friday. >> Now, admittedly, nothing here that we couldn’t do in a non-canvas version. So,

03:05 uh, you know, pushing it forward and making it a little bit more intentional and challenging the board a bit, I gave it the material from our flamethrower girl / jungle girl uh, prompt reference stress test that we ran for C Dance 2.0. Uh, and it it actually ended up doing some pretty interesting things here. Now, interestingly, one of the first things that it did is it actually went through and created uh, basically like transparent PNGs for all of our elements and characters. Uh, I thought that was kind of interesting. Kind of just decided to do that on its own. Uh, from

03:36 there, after discussing what we what I wanted, uh, you know, to accomplish with this uh, this little test, it basically ended up coming up with its own little uh, you know, brief on it and uh, asset list uh, and it’s actually, you you a breakdown and plan for the generation. Very interesting side note here is that it decided to name flamethrower girl Lyra, which is not her name. That’s actually the name of our barbarian from the bridge. Now, I don’t know where it pulled that name from considering that

04:07 the bridge was not generated on Dreamina. That’s it. I mean, nothing conspiratorial. Somewhere in past generations, I probably used the name, but I do I did think that was pretty interesting. Anyhow, so what we’re looking at here is actually not the original plan that it came up with. We’ll talk about that a little bit later, but these are the initial three generations that it came up with because it’s initial plan was like, “Hey, instead of doing one 32nd generation, I want to do this as 10 separate C 2.5 generations.”

04:37 And admittedly, I was not keeping an eye on it, so it decided to generate in 16:9. You know what? Overall, I wasn’t that happy with it. Actually, we’re just going to run these MOS because well, I mean, I’m not overly in love with them. I mean, from a technical standpoint, nothing super wrong with this. It just it does feel a bit a little bit on the bland side. The other problem is that at this 10-second mark, we can see that we’re supposed to cut then to the next shot, which would be this. So, there’s kind of like this massive jump cut. The size of the idol completely changes. And then as she runs, you can see the idol

05:09 just completely vanishes. Our third shot picks up. She’s kind of in mid-roll. She does have the idol back again. Our guardian is chasing her, obviously. And she actually not not too much wrong with the back end of that. But, you know, overall, it’s just not it’s not going to cut together, and I’m not in love with it. So, experimenting around a bit more, I decided to give it the original storyboard breakdown for this stress test. Now, this was generated up in GPT image one. So, yeah, it has been a minute. And specifically, I told the

05:40 agent like, “I don’t want you to use this as storyboards, but rather use it as inspiration for your version of this.” Actually, it’s kind [clears throat] of funny. The The agent was like, “This is one of the best previews shot breakdowns I’ve ever seen.” So, yeah. You know, the Dream Agent will gas you up. If you’re feeling a little bit low, go have a chat with it. It’ll make you feel better. Now, what I thought was pretty neat is that after providing it with the storyboards and kind of heading into this direction, that all of sort of our our shot planning and text stuff over here, that all auto updated. So,

06:11 actually I thought that was actually very smart. So, you know, it just helps again keep everything organized and streamlined. Now, all that said, what we actually ended up getting out as a generation, well, I mean I do have notes. Let’s take a quick look.

06:55 So, yeah. I mean, overall it I mean, it’s all there. Maybe not quite as kinetic again as like CDS 2.0, but that’s that’s just something that we I’ve noticed a lot with CDS 2.5. I will say that we’ve been running into like these problems with the glowing eyes. I mean, that was obviously the thing that stood out the most. And then we have this weird jump cut here. Now, that said, there is actually kind of a cool new edit segment feature over here in which we could scroll over to the glowing eyes. And then we can essentially prompt out of that. So,

07:27 that’s pretty handy. Now, that said, when I did run that edit, what CDS decided to do was just have her eyes closed the entire time. So, yeah. I mean, I got to admit that was that was pretty funny. It generally does just make me laugh when AI is like the ultimate snooty waiter. “Oh, you wanted her eyes open and not glowing? Well, sir, that is not what you ordered. So, ultimately, I mean, I do think that the canvas feature has significantly improved. Um but, when it comes time to actually, you know, prompting your shot, I think that that’s a spot that you should take over manual control. So, um

07:57 again, all of this stuff, I think it does a great job keeping all of that organized and pulling things together. Uh but, then ultimately, when it comes time to generating your shot, yeah, again, just go in and and manually take care of that or you know, use your LLM of choice. Uh for example, taking all of this and then just running it via my prompt.

08:41 >> [sighs] >> And look, I mean, I’m biased, but I do think that that looks better. Quick quality of life feature that they’ve thrown in, we now have an elements folder. And while this is like uh you know, I haven’t loaded this up with everything yet, but uh while this is very simple, uh like Kling has had this for ages. Um but, you know, essentially, I can put uh Tuesday Sinclair in here. Uh we can even give her a custom voice, which is pretty handy. Uh and then now in any prompt, anywhere across either the canvas or standard generation, I can just at Tuesday and then it’ll automatically add her in. Uh plus, we

09:12 have, you know, additional angles for her as well. Um again, nothing groundbreaking, but at the same time, um sorely overdue and uh very much appreciated. If you want to see how all of this can come together for a longer-form piece, um friend of the channel, PJ Ace, uh recently released uh Skyline Heist, I believe it’s called. It’s the uh it’s a chapter from his Nexus Project or Nexus Universe, I guess. Yeah, I mean, it’s very, you know, uh steampunk uh kind of vibe um heist film in the sky. No notes. So, as

09:45 always with one of PJ’s projects, um there is Oh, I mean, there’s a lot going on on this board. Um so, you know, you can see essentially all of the shots uh generated, but I think that where you might find some value in it as well is if you start punching it like you you’ll see like his version of uh character sheets here. Um so, I’ll have to dig into this. This is actually it’s a pretty good format for it. Now, this whole thing is definitely worth a gander um mostly because I mean, you know, you can sort of peek behind the scenes a little bit uh and see some stuff in here like uh

10:16 PJ’s got a a a note to the animator, which I guess would be essentially the board itself. Uh little outdated, I think he calls uh out uh 2.5 is limited to 720. It is 1080 now. Um but, you know, again, most of the information in here is uh is is is worth scoping out um to steal for yourself. Um he’s also got an acting system in here. This might be a little bit on the outdated side as we’ll see in just a minute because CD has now has kind of a micro acting skill that’s in there, but still, you know, again, worth uh worth cribbing this. Um

10:49 the other thing I think is like just, you know, valuable to look through. If you come down to generated videos down here, um it gives you an idea of essentially what he was using for uh you know, various elements. I know PJ is more of a um uh character reference sheet and like location or back plate um guy and then just providing very detailed prompts in terms of action as opposed to first frame, but um yeah, I think this gives Overall, there’s just like there’s a lot of really, really interesting stuff on this board that you can kind of like nick and and mine from.

11:20 So, uh and definitely end up giving you some ideas of your own. The link to PJ’s film is down below, and of course, uh the entire canvas workflow is over on Dream and uh Now, there are two other really interesting updates, uh one of which is currently in preview, so this is kind of like your first time looking at it, but again, I think you’re really going to be appreciative of it. We’re going to take a look at those, but first, a quick word from today’s sponsor, our friends at GenSpark. Today, we’re going to check in with our friends at GenSpark, who were kind enough to sponsor today’s video. Specifically,

11:51 GenSpark Design, which is a cool new feature that is worth showing you rather than just talking about. Overall, the idea is one prompt and it design, and it comes back editable, so you control it. So, as a quick test, I of course asked it for a poster for The Bridge. The process was actually pretty cool as well. I first had it like watch the film, and yeah, I mean, it proved that it definitely knew what the story was about, and then it followed up with a few questions. And that was followed up actually by another set of questions,

12:21 which oddly enough actually ended up leading me into a direction that I wasn’t even aware that I wanted when I first started this. So, we ended up in this cool spot here. Let me move into present mode over here, where we were kind of more in like an indie film kind of like, I guess, A24-ish style, which actually I mean, I thought kind of worked for this. Now, what’s what’s really kind of cool about this was that I can take any element here and just, you know, resize it or, you know, if I wanted to change the the font on this, I could. If I wanted to change the color

12:52 on this, I could. If I wanted to move things around, I could. All of this is fully editable, so like there’s no worry about having to go back and reroll the whole thing. But here’s the big time saver is like you can take this whole same look and say something like, “Give me a whole launch kit for it.” So, now we have assets in like 916 and 11 for various social media platforms. And I do want to point out that this was all done very conversationally over on this side. And on top of it, you can this was actually surprising to me, but you can’t actually browse local files right from

13:24 here, or you know, obviously you have uh GenSpark’s AI Drive where you can put your files as well. Obviously, we can see that we have 916 uh sort of like, you know, poster versions of it. Uh it actually came up with these like YouTube thumbnails, which actually are really good. I actually like this title treatment a lot. Um and then uh oh, just a 1:1 Instagram post as well. Once again, all of this fully editable as well. Does anyone remember actually having to do all of that manually in Photoshop? I mean, that was a day. And now, here is the big one is that all of this does video now, too.

13:56 Once again, we went through a series of questions as to what I was looking for. Um in this case, I went with kind of like a 10- to 15-second 9 by 16 like TikToky vertical thing. So, uh just kind of briefly chatting with it and giving it some direction, this is what it came up with. Now, admittedly not perfect, but also for a one-shot, I mean, not bad. I did

14:26 like the music that it picked out for this. Now, if we want to take another path of the edit, again, we just we just chat with GenSpark Design. Now, all of that did kind of get me thinking since it has all of this context, could it design a web app for me? And yeah, after just a few minutes, we end up kind of with this applet web presentation sort of thing. We’ve got our video streaming here. Uh we just kind of can scroll down and there’s our teaser again. We have all of our character images here. Uh obviously, down here, we can tweak anything that we want as well. Uh and then it provided us with uh a stills

14:56 gallery as well taken from uh some of the uh screenshots from the film. Um and then we we have a like a download section. Again, all of this is editable and customizable. Now, while all of this is pretty cool for me, I mean, I think the where the real power of all of this comes is uh when you upload your logos, colors, and fonts as essentially your brand in and, you know, lock in a visual lookbook that is your own. Now, if you’re looking for inspo or just a quick start, uh GenSpark Design, of course, does have a number of templates here

15:26 available for you. So, uh great place to get started with your project. Uh also, all of these templates are totally free. Speaking of free, GenSpark does give you free compute credits every day you log in. So, you can try all of this today. Hey, a big thanks to GenSpark for sponsoring today’s video. Link is in the description. Sliding back over to Dreamina and specifically on the generate side as opposed to the canvas side that we were looking at earlier. So, this is kind of like our regular usual like prompt goes in here, image or

15:56 video references go here. Um there are some new things that they’ve added in. Um skills, namely. So, uh we have like an iconic film style over here. Um we have a skill for cinematic long takes. This is the thing that allows for longer than 30-second generations. To be honest, I don’t really mess with that that much only because I mean well I mean that gets expensive. Um although, something pretty great about that coming up in just a minute. We still have a film storyboard uh skill and then the micro expression prompt, which is uh we

16:27 kind of talked about that a little bit earlier with the acting thing. So, this allows for I guess like multiple uh emotional ranges within one prompt. Um so, there’s a whole bunch of other skills that you can grab now as well. I mean I kind of read them as like pre-built markdown files. Um but, you know, definitely again something worth kind of going in and taking a look at uh or just, you know, utilizing just by, you know, hitting using chat. Now, all that said, what I think is super exciting for everyone is uh and again, this is in preview right now. So, um

16:57 expect it to release at some point soon is uh we now have SDNS 2.5 for preview. So, what this will do is actually whatever you prompt in here will generate in SDNS 2.5 um at 480 for obviously a significantly lower price than it would cost you to do it in like 720 or 1080. So, giving it a quick whirl with uh channel fan favorite sniper girl. Again, we’re running this in C Dance 2.5 preview. So, this is a 480 generation.

17:38 >> Got you, Clanker. >> Admittedly, kind of a weird generation there uh with the cross dissolve in the sniper scope. Um at the same time, I think it looks kind of cool. That said, again, 480, so there’s not a lot that we can do with this in terms of upscaling. In fact, if you were to try to bring it up to 1080 or 4K, it probably look a bit of a mess. But, what we can do now is actually generate in full quality. And if we do so, C Dance will then re-render that clip at 1080.

18:09 So, this is not upscaling it. This is actually a full re-render. >> Got you, Clanker. >> So, really this allows us to actually really hone in on a prompt as opposed to using one of the fast models. And obviously, if you try to use that exact same prompt in one of the higher level

18:40 models, you’re not guaranteed to get the same output. This time, you are. Now, I’m not exactly sure when they’re going to be releasing this, but it is probably pretty soon. I mean, it’s definitely past beta. It is in preview. So, I expect you to probably have access to this hopefully like within a week or so. I do think that we’re on the verge of some pretty big ByteDance and C Dance updates as well. BytePlus has just introduced Dramagic, which is for it’s actually more on the enterprise side, but it does is script analysis assistant

19:12 setup, storyboarding, and video review, uh essentially all uh in, you know, one platform. Uh this one is uh in beta right now. There’s actually a link um to request access down below um if you want to give it a shot. I should say that’s a link to request access, not get access. Um so, I mean, I signed up for it, so I have requested access. Doesn’t mean I’m going to get it. Uh the other kind of interesting thing is that there have been rumors uh going around lately that ByteDance is nearing the release of a its own world model. So, that should be

19:44 very interesting as well. Definitely be keeping an eye out for that, and I will let you know as soon as I hear more. In the meantime, I thank you for watching. My name is Tim.