Seedance 2.0 may be the next big hope for AGI video, but it’s still slow

When Irish director Ruairi Robinson began uploading a series of short clips created with Seedance 2.0 – TikTok developer ByteDance’s latest video generation model – It was hard to deny that the footage was more impressive than what we’ve seen from other AI sets. The star of the clips (a digital version of Tom Cruise) looked very much like the real thing as he battled Brad Pitt, humanoid robots and zombies. The characters moved with a complex, almost choreographic fluidity, amplified by the kinetic “camera.”

AI generation enthusiasts like to proclaim that the traditionally produced entertainment industry is cooked, and some of Hollywood’s biggest studios appear alarmed by Seedance’s latest capabilities as artificial-Cruise videos continue to rack up online views. Motion Picture Association, Disney, Paramountand Netflix Each company sent ByteDance cease and desist letters over allegations of copyright infringement. In response, ByteDance said It will take steps “to strengthen existing safeguards as we work to prevent unauthorized use of intellectual property and the like by users.” ByteDance has not yet officially released a version of Seedance that prevents users from creating snapshots that the company does not have the rights to create.

Everything about the Seedance 2.0 rollout is there It felt like a viral stuntespecially when the studios have already made it clear that it is Willing to sue when AI companies steal their intellectual property. It’s true that the videos created by Seedance look much better than a lot of the stuff we’ve seen made with Sora, Veo, Runway, and others. But the fact that producing highly polished copies is the new model’s main claim to fame makes the Seedance 2.0 just another generator of aberration – albeit a more fascinating one.

When we refer to AGI video as “ramp,” we’re usually commenting on the aesthetics and presentation. But the means by which AI shots are generated is an important part of the equation. Unlike traditionally produced movies, shows and online videos – which can be foully placed – Objects made using AI are considered “rough” because they are the product of a workflow devoid of any direct authorial or artistic intent. Unlike a team of human filmmakers, an AGI video model can’t always follow story beats or character motivations, but it can analyze simple inputs and generate outputs. It seems Informed by narrative (if you’re squinting) because the software has been trained on massive amounts of visual data.

Being able to mimic the real thing (read: human-made) is the primary goal of projects like Seedance 2.0, but models can’t do that unless they’re first given plenty of source material to iterate on programmatically. By allowing such blatant intellectual property infringement, ByteDance is telling us that — aside from more vivid action shots and stronger sound design — at its core, Seedance isn’t much different from its peers. It was easy to recognize Seedance 2.0 as a slob generator when you look at the most viral clips created with the software, which tend to feature A-list celebrities and clearly copyrighted fictional characters. But the trick to all of this is much more difficult when you watch Chinese director Jia Zhangke’s film Jia Zhangke dancea short film produced by Seedance 2.0 featuring Zhangke Have a discussion about the nature of creativity with the AI ​​version of it.

Jia Zhangke dance It transitions into the Meta where two characters debate whether films created with AI should be considered compilations of man-made works or a new kind of art form. After one member of the Jias reveals himself to be an AI clone of the other, the short follows them on a file MatrixA flight-like journey through different settings aims to showcase the ability of artificial intelligence to conjure any images a stimulus can think of. Jia Zhangke dance It unfolds with such fluidity and narrative cohesion that you’ll be hard pressed to find a scroll through Sora app from OpenAI. But when you look closely at what’s happening in the background of shorter, busier scenes involving background characters, it’s not hard to see Seedance 2.0 making some of the same continuity errors that plague all video generators.

Jia Zhangke dance This is a prime example of how filmmakers can make acceptable things with AGI provided they are skilled enough to know how to work around the limitations of the technology. Although the film footage is very short, like most AI-generated videos, it has been edited together in a way that creates the illusion that it is part of longer footage. Although characters in the distance are sometimes clipped in and out of view, you can see that Seedance 2.0 is trying to obscure those errors by covering them up with objects moving in the foreground.

If anything, Jia Zhangke dance It shows us how many AI enthusiasts don’t go out of their way to make their creations look like the kind of art that would put their butts into theaters or get people to subscribe to a streaming service. ByteDance engineers deserve at least some credit for building a model that can recreate real people’s faces so accurately. But it looks very much as if this power may be linked to illicit model training data, which has put ByteDance in such trouble that the company… It has paused its plans to release the Seedance 2.0 API to the public.

In addition to looking better than it currently is, the only way AI-generated video might be able to escape the association with regression is for the companies behind it to prove that their models can create things without having to steal other people’s work. Studios like Asteria and Companies including Adobe They try to address this second problem with “IP secure” models built on data that has been properly authorized. But until we start seeing good work coming out of this new wave of AI software, it’s going to be a slow ride along the way.