IN THIS VIDEO: In this video I'm walking you through Dreamina Seedance 2.0, ByteDance's brand new AI video model that just launched inside the Dreamina platform. I'll cover what's actually new under the hood, including the new multimodal reference system (you can now stack up to 9 images, 3 videos, and 3 audio files in a single prompt), the unified audio-video architecture that handles lip sync and sound effects natively, and the big jumps in motion stability and physics. From there, I'll run through a few real demos: the Omni Reference feature merging multiple asset types, a no-reference hyperrealistic text-to-video test, an AI product commercial workflow, and using a video clip as a reference input. By the end, you'll have a clear sense of whether Seedance 2.0 is the right model for your commercial or creative projects.
NOTE: Dreamina Seedance 2.0 is now live on Dreamina. No queueing.
CHAPTERS: 0:00 What Just Launched 0:16 In This Video 0:38 What's New in Dreamina Seedance 2.0 0:46 Seedance & ByteDance Background 0:54 Multimodal Reference Inputs Explained (9 Images, 3 Videos, 3 Audio) 1:39 Unified Multimodal Audio-Video Architecture 2:04 Native Audio Generation, Lip Sync & Sound Effects 2:17 Improved Motion Stability & Physics Understanding 2:23 The Shift Toward World Models in AI Video 2:56 Seedance 2.0 Benchmarks (Seed Video Bench 2.0) 3:07 What is Dreamina Seedance 2.0? 3:38 Dreamina Seedance 2.0 Dashboard Walkthrough 4:24 Other Dreamina Seedance 2.0 Features (AI Avatars, Mimic Motions) 4:27 Demo 1: Omni Reference Multimodal Generation Setup 5:04 Combining Image, Character & Audio References in One Prompt 6:04 Demo 1 Result: Merged Multimodal Output 6:24 Demo 2: Hyperrealistic Text-to-Video Test 7:00 Rainy City Street Prompt 7:19 Demo 2 Result: Realism, Reflections & Layered Audio 7:43 More Realism Examples: FPV Drone Skateboarder 8:24 Johnny Harris Style Map Animation 8:46 Demo 3: AI Product Commercial Workflow 9:35 Premium Brand Commercial Prompt with First Frame 10:15 Demo 3 Result: AI-Generated Headphone Ad 10:47 Demo 4: Omni Reference with Video Input 11:29 Demo 4 Result: Multi-Modal Asset Fusion 11:39 Final Thoughts on the Future of AI Video
------------------- Clips channel @jonlawclips
I do NOT sell any products of my own. Watch out for scams.
<html>
<head><title>404 Not Found</title></head>
<body>
<center><h1>404 Not Found</h1></center>
<hr><center>nginx</center>
</body>
</html>
What you're seeing right now is completely AI generated from a single prompt in about 60
seconds. And this is the frontier of AI video. It is not the Google models anymore. It is this model,
if you haven't heard of it, that is DreamInner Seedance 2.0. And that in mind today, I'm going
to give you everything you need to know to best use the DreamInner Seedance 2.0 model.
So exactly what the model does under the hood, and then commercial ready workflows,
all the features, et cetera, so that you can save time, make more money, have more fun,
be more creative, do whatever you want to do with the cutting edge of AI video.
And that in mind, let's get into the video. So first I want to jump into just technically
speaking the differences or the updates on that before we'll get into all the demos and visuals.
So like I said, this is a Seedance model from the company ByteDance. Yes, they make TikTok,
but honestly, these guys are really an AI powerhouse at this point. And one of the biggest
technical updates is that you can now have multimodal reference inputs. And by that,
I mean, most AI video platforms or models to date let you maybe prompt it with a text box
or do image to video, like upload some images and say, I want these things,
these images to be in the finished video. But Seedance 2.0 lets you add nine images,
three videos and three audio files simultaneously. So that is up to 12 reference assets for a single
generation. And that basically just gives you more spot control over the different elements
that you want to be in this video. And then you can tie that together with your text instructions
really cleanly by referencing those assets individually. Again, you'll see what that
looks like in a second. And then going back to the architecture of the model as enabling
that step forward, they use a unified multimodal audio video joint generation system. That's a
little bit of a mouthful, but basically what it means is that they're not generating the
video and the audio separately. It's all a single architecture or a single model that generates the
audio and the video in sync. And so that makes that connection or tie in between the video and
the audio that much stronger. And then that really matters when you want things like lip syncs that
actually look accurate. You want layered sound effects to match what's going on in the video.
And so the fact that it's native and not this kind of post-processed thing that's stuck on
makes that a lot better as well. Now, motion stability and kind of an understanding of
physics is also a lot better with this second version. That's a big one. And I think generally
a trend we're seeing in AI is just shifting toward, I don't want to say world models,
but models that have more of a first principled understanding of the world through physics,
through math. And so I think the better these models understand that, and maybe that's driven
by advances in training, more compute, for whatever reason, the more these models understand
the world at a fundamental level in terms of video output, the better they can then construct
these virtual videos, worlds, motion, movement, et cetera. And that all translates into really good
benchmarks for the model. DreamIn a Seedense 2.0 is leading across text-to-video, image-to-video,
and multimodal tasks on Seed Video Bench 2.0. And that in mind, if you guys have never heard
of the DreamIn a platform before, I made a prior video about it. We'll put a card up here somewhere.
So you can go watch that. I won't go into it too much. We're just going to jump into the demos now,
but basically again, DreamIn a is kind of the all-in-one home for the ByteDance models.
So Seedance on the video side, and then Seedream on the image side. Now, if you want to follow
along with the demos or just try to understand the platform while you're watching this video,
the link is in the description for that, and you help support the channel by doing so.
And without further ado, let's jump into my computer. Okay. So this right here is the dashboard
for DreamIn a at the urldreamina.catput.com. If you saw the last video, you're familiar with how
this works. It's a very intuitive system right here, where we can talk to the agent. The agent
can route us to the different individual models, or you can have point control over some models by
selecting AI image, AI video, et cetera. For example, if we go to AI video, then we can select
the individual models here. We can also see some reference visuals and creations with the Seedream
models out of ByteDance right here. If we go over to AI shorts, we can see videos created through
the models that we're about to take a look into. Now, there's also a lot that we're not going to
get into in this video, whether that's the AI avatars or something really cool that's coming
soon, which is mimic motions. We're going to start with the same setup. So I'll go under AI agent,
AI video, but then for first and last frames, I'm going to go ahead and change this to Omni
reference. And this is where we can add up to 12 references that are multimodal, so image, video,
and audio in nature. And under five seconds, I'm going to go ahead and change this to a monster
behemoth 15 second generation. I'll also go ahead and make this vertical. And to showcase this,
I want to merge three different reference elements. Right here, we have a visual of a
dock, then we have a image of a person or a character, and then we have some happy music,
so a 90 second mp3 track. So I'm going to go ahead, I'm going to upload all of these as references.
And then what I want to do is I want to merge these into the prompt that I'm going to give
the model. So I have image one, image two, and image three. What I want to say is create a scene
with image one sitting on a dock that looks like the environment and dock of, and then I'll reference
image two and realize that I uploaded the wrong file. I have corrected that so we can see the
audio track here. And so I'll say play audio one in the background. So again, you can get very
granular control over the shot that you want, whether it's an ad, whether it's a scene like
visual for a video, by just compiling all these different elements could be the texture of a wall,
it could be the material, could be multiple people or characters or animals or products,
whatever it is, and you just merge them into your scene like so by just talking to the agent. So we
have this all set up. Again, this will be a 15 second generation. Let's let it. Okay, here's the
final render. Let's give it a watch. So we can see the deck from the image just from a different
perspective. That is the exact same audio track that was provided and we have the Asian woman.
So that is how you can go ahead and construct these shots for use in whatever kind of media
that you would like. And that wraps up our final look into on the reference. We're going to start
with the basics here. No reference images, no frame control, just the straight realism that
we want to try and generate. We want to look at the physics, the consistency and see overall
how well the model forms. So I'm going to purposely leave this prompt fairly open ended
under AI agent. We're going to jump into AI video for the model. We're going to select
We're going to leave this on first and last frames versus multi-frame or omni reference,
but I'm just not going to actually insert images for those frames. We'll leave it at 16 by nine.
You can go for up to a 15 second mammoth generation. We're going to leave it at five
seconds and this will be the prompt. So a person walking through a rainy city street with reflections
and interaction with objects, in this case, the umbrella passing traffic. And we want all of that
to be hyper-realistic with audio. Let's go ahead and see how well the model performs here. Okay.
And here's the result. So if we just pause and look at a few frames here, I mean, you can see
the reflections. You can see the realism of the umbrella there. That is exactly what you would
get. If not something worse with a real camera, we can see the person looking hyper-realistic.
And then also I want you to listen one more time to the layered audio here. We have the noise of
car, we have the water, we have the city ambience all working together. Now, before we jump into the
second demo, I want to run through a few more examples of the kind of realism that you can get
with the C-DANCE model or at least C-DANCE 2.0 in these different environments. Here's the first one.
This will be a 10 second ultra cinematic FPV drone shot following a skateboarder. So here you can
start to see as well how the prompts kind of interplays with the quality of the output. So in
this prompt, we can see a little bit, you know, more specific detailing of the environment. We want
these hyper-detailed textures for the road, subtle V-shape, just things that you can do to make it
feel more realistic. And that is a pretty stunning output as far as the visuals and the realism goes.
And then here we can see something which is a very different visual style. So less about the
realism, but more so a real workflow that you may use as a content creator. So this is Johnny Harris
style map where we just described the camera movement. Let's see what that looks like. So pretty
amazing video there. And that is something you could directly drag and drop into the timeline. That is
just the base quality of the model. And that in mind, I want to jump into a little more sophisticated
of a workflow that's probably more representative of what you're going to use on a day-to-day basis
for your business, for your personal work, for your media generally. And that's going to be, let's say
we want to create some kind of a commercial. So we have a product, a service, or a person, or some
focal element that we want to build this video around. And then we want to create some kind of
visual for taking that element. And for example, making it appear in a studio, in a real world
environment, in some way, shape, or form that fits into, for example, an ad, a YouTube video, a short
form piece of content, whatever that final product looks like for you. So we're going to start with
the same workflow. I'm going to go under AI video. We're still on first and last frames. We'll get away
from that in just a second here. And this is the prompt we're going to run with. So create a premium
brand commercial and then use the provided product image as the base. And then we're describing a
little bit of the camera motion. So we want some more control over the shots. Now this right here,
this image of headphones is what we're going to use as the reference. I'll go ahead and save that.
Now here where it says first frame, I'll upload the image. So all this means is that this will be
the first frame of the video. And then the agent will take it from there and essentially create
this implicit 3D model that is used in the simulated environment. A lot of stuff happening
behind the scenes that we just get to enjoy the spoils of. And that in mind, let's let this thing
run. Okay, here's the result. Let's give this one a watch. So we can see the elements are hit on the
nose here with the lighting and the camera orbit. And again, this is starting with the first frame
as this image, and then it creates that environment with the audio and the lighting on top. Now as
far as the UI UX goes as well, if we press on the headphones right here, there's a few other tools.
So interpolate, upscale, and then generate a soundtrack. If you want to change something
about this render as well, you can go ahead and just edit the prompt or regenerate it for another.
Now, one quick feature note here as well, we didn't cover the omni reference video
reference ability. So right here, you can see that I uploaded one image, one audio file,
one video as a reference, and then I'll describe what I want with text. So that is four different
types of media that you can pull together to prompt the model to create what you want.
So I can say, I want you to recreate the reference video, except with the color of the green in the
image and using the audio I provided. And so this is a very apt descriptor of the kind of
multimedia creations that you can get out of the omni reference feature. Let's go ahead and see
what this creates. And here's the result. So we can see the green, the correct audio is used,
and the reference video. And that right there is the omni reference including video. And there you
go. That is Dream in a Seedance 2.0 and some pretty incredible AI video. I think this stuff
is going to change the world so much. And we're still, despite all these advances really early,
there is a really crazy future in front of us. And it pays to, I think, stay on top of these tools.
But anyway, I appreciate your time of being here. I hope you find this model useful for you,
whether commercially or creatively, or hopefully some fusion of the two.
And I will see you all in the next one. Cheers.