ByteDance New Releases AI Video Model – Goodbye Sora, Your Time Has Passed.

Just now, the launch of ByteDance’s Volcano Engine is basically over.
I’m a bit over-excited right now.
Even though the launch is over, I feel that a brand new beginning to disrupt the industry has, at this moment, officially arrived.


ByteDance has officially released their two new AI video models:


Doubao Video Generation – PixelDance model and Seaweed model.
I’ll talk more about the Seaweed model next time. This time, I want to talk about this Doubao PixelDance model because it’s so dope, so dope, that I literally watched it in awe the entire time.

The moment they officially announced this thing, there was so much applause that I felt like I was going to blow the roof off the house from the screen.
Really, if I had to summarize this Doubao PixelDance model, it would be in three words:
Complex continuous movement of characters, multi-camera combination video, and extreme camera control.
Sounds a little hard to understand, doesn’t it? No hurry, I will explain in detail.

I first put a few cases, to feel the shock of this thing:

Really, the film and television industry before, almost can not use AI, is because, the character performance is too garbage, and the scene and the character consistency is too poor, the operation of the mirror to be honest is not good.

Now, ByteDance has stepped in and taken AI video to a whole new level.


The singularity of industry disruption has officially arrived today, at this very launch.
And I, after holding my breath for 4 full days, can finally send out this article.
Yes, 4 days ago, I was invited by ByteDance, measured this Doubao PixelDance model in advance, at that time, I was shocked beyond words, you know, as a blogger, after measuring such a cocky thing, naturally want to be the first time to share it out, but because of the confidentiality agreement, I can only not say a word about it.
So you just know how hard it was for me to hold it in these 4 days.
And now it’s all coming together. I can finally fucking talk.
Back to those three most important features:
Complex continuous movements of the characters, multi-camera combination videos, and extreme camera control.

Characters can do continuous action


In the past, AI videos have a very fatal point, that is, they look like PPT animation.


Whether it is Sora’s video, or runway, or Keling, etc., the movement amplitude, but only the lens amplitude is large, there is never a complex movement of people.
Top of the day, turn around, or take a quick run, or wave, or hug. Honestly, just the hug alone, not many AI videos can do that.
And what if you had the girl in the picture, take off her sunglasses, stand up, and walk towards the statue?


All AI videos, all dead in action.
And this time the Doubao PixelDance, did it, literally.


Aside from some minor flickering of the watch on the hand, the character proportions, movements, limbs, lighting, etc., were almost flawless.
A play looks good, people’s action performance, is the most important ah.
For example, in The King of Comedy, in the last scene, Stephen Chow’s Yin Tian Xiu, after shouting the classic “I’ll support you” line to Liu Piao Piao, Liu Piao Piao sits in the departing cab and cries very sadly, looks at the money and the watch in her hand for a while, then puts them into her bag, and pulls out the book “Self-Cultivation of Actors”, which she regards as her faith, and cries very sadly. Self-Cultivation of the Actor, and hugged it sadly to his chest.
This performance, it’s continuous. It’s what’s continuous that has tension. It’s only when you can feel it, that aching emotion.

And now, with AI, generating character performances that can do continuous actions is no longer empty talk.


Look at another case where a man takes a sip of coffee, then puts it down, and a woman comes up from behind.

Also, the character expressions are dope, the old man smiles and laughs, then cries.


I want to cry too, really.
When I did the trailer for Wandering Earth 3 last August, I fantasized about a million possibilities for AI doing character acting.
Now, just one year later, Doubao has helped me fulfill this biggest dream.

Multi-camera combination video


The ability to generate a multi-camera video with consistent style, scene, and characters from a single image + Prompt is something I’ve only seen inside Sora’s promo.
It’s that famous video of a wolf howling at the moon.


Actually, to be honest, this video was, at the time, very shocking to watch, but it’s actually okay to watch it now; the style, characters and scenes are so simple that consistency is well maintained, and there’s no complicated story or subplots.
But that’s it, now, there still isn’t any AI video that can do multiple shots in a single video and still have perfect consistency.
Don’t even get me started on the LTX studio stuff, that’s fine for storyboards, but a feature film? Wash your ass, don’t even talk about the scenes, it’s hard to keep the characters in panoramic, medium, and close-ups uniform. And it’s really ugly.
But now, Doubao PixelDance made it, and the consistency is simply unbeatable, really.
And it only takes one image + Prompt.
For example, this one.

Prompt: death with a scythe approaches the woman. Close-up of the woman's face as she screams in terror.


Extreme camera control


Doubao PixelDance modeling is the most outrageous and awesome I’ve ever seen.
Now the AI video lens control, still basically focused on the camera + motion brush combination of two functions, but to be honest, the upper limit is really limited, a lot of large lens and zoom, simply can not be done.
And Doubao PixelDance, the effect is really fucking outrageous.
What bird’s eye view zoom up and rotate this kind of base manipulation I do not say, the key is, in a word, a variety of 360 degrees around the subject of surround, front and rear view zoom, panning, target following, lifting and lowering the lens of whatever thing can be.
The effect is surprisingly good, I saw for the first time, in the AI video, transport mirror can be so awesome, so cool.
Directly look at the case.

Prompt: the woman smiles and lowers her head, the camera pulls away, and a white man gazes at the woman.


The zoom is extremely natural and smooth, invincible, too invincible.
And then there’s this one, a 360-degree drastic wrap-around dribbler.
Prompt: black and white style, the camera shoots around the woman wearing sunglasses, moving from her side to the front, and finally focusing on a close-up of the woman’s face.

This is a picture, and then a Prompt, can you believe it? This range of motion, this stability, than the fucking modeling out of the outrageous, I’m really convinced.
How can you let the photographers still play, crazy ah…

Write in the end


Sora a giant futures, from the 2.16th to nowadays, late to see any trace.
And then, 6.6, can Ling silent, officially online, on behalf of the output of China Sora.
And today, 9.24, ByteDances again AI video, pushed to a whole new level, is a in Sora’s promotional video, can not see the height.
So far, China does not need Sora, Doubao model is the sky.
Doubao PixelDance also does not need any Chinese version of Sora’s nickname, Doubao PixelDance is Doubao PixelDance, he is now the days of AI video.
Also to this point, AI video is no longer a toy, but a real, can enter the film and television, advertising, animation workflow, bring some new imagination.
This shot was fired by us.
Today this Doubao PixelDance model, will give priority to the enterprise to open the invitation to test, in a few days on the volcano ark, as for when on the line that dream to the C-user full open, may have to wait for a period of time, after all, is too new, they said that they still want to optimize optimize the model ability, stable, then directly on the line that dream, to the full open.
Really, there has never been any miracle, everything is the accumulation of many years of precipitation, everything is as promised.
Today, I can also shout that line:

Other Video Generated by PixelDance:

At Last ： How to Apply for PixelDance NOW?

https://console.volcengine.com/ark/region:ark+cn-beijing/experience/vision?type=GenVideo

First Register your account :

账号登录-火山引擎 (volcengine.com)

Login with your mobile phone.

Apply access here:

Now you have done , plz waiting for reply