Transfer over Sora 2, there’s a scorching new AI video mannequin on the town.
In early February, ByteDance unveiled Seedance 2.0, a significant improve to its flagship video mannequin, which had beforehand remained pretty obscure. Its highly effective capabilities instantly shocked the AI ecosystem in China, even amongst audiences who had as soon as been skeptical of AI-generated video and considered the expertise primarily as a solution to produce slop.
Feng Ji, the founding father of Sport Science, the studio that developed China’s world hit online game Black Fable: Wukong, wrote on-line that he was “deeply shocked” by the mannequin’s talents and believed Seedance 2.0 would pose important challenges to China’s present copyright laws and content material moderation techniques. Pan Tianhong, who leads a Chinese language skilled video manufacturing studio with over 15 million followers on social media, posted a video during which he mentioned Seedance 2.0 is considerably higher than any video-making fashions that got here earlier than it. “It thinks like a director,” Pan mentioned.
Nevertheless, most individuals can’t get their arms on the mannequin at this second as a result of entry stays pretty restricted. As of this week, ByteDance is barely permitting present customers of its consumer-facing AI apps in China—the most well-liked one is the chatbot app Doubao, however the firm additionally has a complicated constellation of lesser-known apps like Jimeng, Xiaoyunque, and Spark—to expertise Seedance 2.0. All these apps are for the Chinese language home market solely, stopping individuals outdoors the nation from testing the mannequin themselves. (The restrictions have prompted some savvy individuals in China to resell their ByteDance accounts to keen early AI adopters abroad.)
However there are indicators that the mannequin may develop into extra accessible quickly. This week, ByteDance up to date its API platform and disclosed the proposed pricing of Seedance 2.0: A 15-second video, the longest it might probably generate proper now, would price barely greater than $2 to make, Chinese language publication IT Home estimated. ByteDance nonetheless hasn’t opened up API entry to third-party builders, however that must be on the horizon.
Afra Wang, writer of the Substack publication Concurrent and an in depth observer of the US-China AI panorama, tells me that Seedance 2.0 is one other fascinating instance of how the 2 international locations have taken diverging paths. Even earlier than the discharge of Seedance 2.0, a few of the most established video-making AI instruments on the planet, resembling Kling AI, have been developed by Chinese language firms. “China hasn’t produced any first rate AI coding device, which is why Chinese language persons are all depending on Claude Code or Codex; however relating to video AI, China is miles forward of the US,” Wang says.
However all of the hype apart, Seedance is working into two severe issues. Weeks into its launch, ByteDance is dealing with a compute bottleneck that’s inflicting the mannequin to take hours to generate a single video. In the meantime, main film studios, together with Disney, Netflix, and Paramount, have all despatched ByteDance cease-and-desist letters alleging that Seedance 2.0’s outputs are infringing on their copyrighted works. ByteDance didn’t instantly return a request for remark.
The Bandwidth Downside
Even for those who get entry to a ByteDance AI app, it’s nonetheless removed from straightforward to generate a video with Seedance 2.0, as a result of too many individuals try to do the identical factor, and ByteDance has but to supply sufficient compute assets for everybody.
After I tried to create a clip with one among ByteDance’s apps this week, it advised me that I used to be quantity 90,985 within the queue, and it will take about 4 hours to generate a five-second video. After ready for 2 hours, the app advised me I now had six extra to go. At that time, I made a decision to simply go to mattress.
