When Irish filmmaker Ruairi Robinson started uploading a series of short clips made with SeaDance 2.0, TikTok developer ByteDance’s latest video generation model, it was hard to deny that the footage was far more impressive than anything we’ve seen from other generations of AI products. The star of the clip (a digital replica of Tom Cruise) looked a lot like the real thing as he battled Brad Pitt, humanoid robots, and zombies. The characters moved with a complex fluidity that could be called choreography, amplified by the dynamic camera work.
Gen AI enthusiasts like to proclaim that the traditionally produced entertainment industry is cooked, but some of Hollywood’s biggest studios appear wary of SeaDance’s recent capabilities, as its Ersatz Cruise videos continue to rack up views online. The Motion Picture Association of Japan, Disney, Paramount, and Netflix each sent cease-and-desist letters to ByteDance over alleged copyright infringement. And in response, ByteDance said it would take steps to “strengthen our current safeguards to prevent unauthorized use of our intellectual property and likeness by our users.” ByteDance has not yet officially released a version of Seedance that prevents users from producing footage that the company does not have the rights to create.
Everything about Seedance 2.0’s rollout felt like a viral stunt, especially when the studio has already made it clear that it’s willing to sue if AI companies steal its IP. Indeed, the videos created by Seedance are much better than many of the videos created by Sora, Veo, Runway, etc. But the fact that churning out extremely sophisticated rip-offs is this new model’s main claim to fame makes Seedance 2.0 just another slop generator, albeit a flashier one.
When we call a Gen AI video “bad” we are usually commenting on the aesthetics and presentation. But the means to create AI footage is an important part of the equation. In contrast to traditionally produced movies, shows, and online videos, these sloppily crafted — Things made with AI are “sloppy.” Because they are the product of a workflow with no direct authorship or artistic intent. Unlike a team of human filmmakers, Gen AI video models can’t always follow story beats or character motivations, but they can parse simple inputs and produce outputs like this: seems to be The program is trained on vast amounts of visual data, so (if you squint) it is informed by the story.
At its heart, Seedance is not that different from other companies.
Being able to imitate the real thing (something created by humans) is the whole point of projects like Seedance 2.0, but a model can’t do that unless it’s first given enough source material to iterate programmatically. And by allowing such blatant intellectual property infringement, ByteDance has essentially told us that Seedance isn’t all that different from its competitors, save for more dynamic action shots and stronger sound design. It’s easy to see that Seedance 2.0 is a slop generator by looking at some of the most viral clips created with the program. The clips tend to feature A-list celebrities or apparently copyrighted fictional characters. But when you watch the movie directed by China’s Jia Jianku, the trick becomes even more difficult to understand. Jia Zhangke’s dancea short film generated by Seedance 2.0, depicting Zhangke discussing the nature of creativity with an AI version of himself.
Jia Zhangke’s dance The film takes a meta turn as two characters debate whether to consider AI-made films to be copies of human creations or a new kind of art form. After one of the Jias reveals himself to be an AI copy of the other, the short follows the duo. matrixIt’s a journey through different settings aimed at demonstrating the AI’s ability to conjure up any image a teleprompter can come up with. Jia Zhangke’s dance OpenAI’s Sora app unfolds with such fluidity and narrative coherence that it’s hard to scroll through. But if you look closely at what’s going on in the background of short, busy scenes involving background characters, you’ll see that Seedance 2.0 makes the same continuity mistakes that plague all video generators.
Jia Zhangke’s dance This is a shining example of how Gen AI can be used to create something passable if filmmakers are skilled enough to understand how to work around the limitations of the technology. The shots in this film, like most AI-generated videos, are very short, but they are edited together in a way that creates the illusion that they are part of a longer take. You’ll also notice that distant characters sometimes appear in the reflection or disappear, and Seedance 2.0 tries to cover up those mistakes with objects moving in the foreground.
Filmmakers can make something passable with Gen AI if they know how to work around the technology’s limitations
If you need anything, Jia Zhangke’s dance This suggests that many AI enthusiasts aren’t trying particularly hard to make their work look like the kind of art that will get them shown in theaters or subscribed to streaming services. ByteDance’s engineers deserve at least some credit for building a model that can reproduce real people’s faces so accurately. However, it seems very likely that its strength has something to do with the model’s fraudulently obtained training data. This data put ByteDance in so much trouble that the company paused plans to make Seedance 2.0’s API publicly available.
In addition to making AI-generated videos look better than they currently do, the only way AI-generated videos can shake off their association with slop is if the companies developing them prove that their models can create something without having to steal the work of others. Studios like Asteria and companies like Adobe are trying to tackle this second problem with “IP-safe” models built using properly licensed data. But until we start seeing quality results from this new wave of AI programs, it will continue to be a downward spiral.
