ByteDance's Seedance 2.0 AI Video Model Surpasses Sora 2, Drives Chinese AI Stocks Up 20%

A New AI Video Model From ByteDance is Making Waves

It may have just lost control of TikTok U.S., but this weekend ByteDance released Seedance 2.0 -- an AI video generator that is turning heads. CNBC reports that Seedance's headline feature is something called "multi-lens storytelling". AI video is often short, but this model will create several scenes while keeping the same style and characters. The AI can reportedly generate 2K video 30% faster than other models. It can be prompted with an image, a short video, text, or audio. There have been some impressive examples shared online. See below. Beijing-friendly newspaper South China Morning Post reports that the release of the beta model and subsequent buzz has sent the stock prices of Chinese AI firms upwards. SCMP notes that Seedance V2 is currently only available to a few select users of Jimeng AI, which is ByteDance's AI video platform. The feedback has been overwhelmingly positive. Seedance 2.0 provides precise camera controls and editing tools. "With its reality enhancements, I feel it's very hard to tell whether a video is generated by AI," Wang Lei, a programmer from China's southern Guangdong province, tells SCMP. That AI video is getting good is no surprise. It was always obvious that video would eventually catch up with still images and text generators. But where AI video will lead to is an intriguing question. Last week, PetaPixel reported that famed director Darren Aronofsky has released a series of AI-generated short films called 'On This Day... 1776' that mark the semiquincentennial anniversary of the American Revolution. Aronofsky's project has been met with widespread derision. Yet the two videos released so far have racked up hundreds of thousands of views. Whether those numbers come from people who are morbidly curious about a respected filmmaker apparently damaging his own career, or they're genuinely popular remains to be seen. AI is undoubtedly being used to enhance productions in small, background ways. But a fully AI-generated production still seems unlikely to be loved. Despite countless AI influencers declaring "Hollywood is cooked", will audiences really ever connect with dead-eyed AI characters? Right now it's difficult to see that happening.

ByteDance's new Seedance 2.0 supposedly 'surpasses Sora 2'

Seedance 2.0 says it generates 'cinematic content' with 'seamless video extension and natural language control'. TikTok-parent ByteDance launched the pre-release version of its new AI video model, called Seedance 2.0, over the weekend. Positive response to the launch drove up shares in Chinese AI firms. Seedance 2.0 markets itself as a "true" multi-modal AI creator, allowing users to combine images, videos, audio, and text to generate "cinematic content" with "precise reference capabilities, seamless video extension, and natural language control". The model is currently available to select users of Jimeng AI, ByteDance's AI video platform. The new Seedance model allows exporting in 2k with 30pc faster generation that the previous version 1.5, the company's website read. Swiss-based consultancy CTOL called it the "most advanced AI video generation model available", "surpassing OpenAI's Sora 2 and Google's Veo 3.1 in practical testing". Data compiled by Bloomberg earlier today showed publishing company COL Group Co hit its 20pc daily price ceiling, while Shanghai Film Co and gaming and entertainment firm Perfect World Co rose by 10pc. Meanwhile, the Shanghai Shenzhen CSI 300 Index is up by 1.63pc at the time of publishing. Consumer AI video generators have made leaps in advances in just a short period of time. The usual tells - blurry fingers, overly smooth and unrealistic skin and inexplainable changes from frame to frame in AI videos are all becoming extremely hard to notice. While rivalling AI video generator Sora 2 by OpenAI produces results with watermarks (although, there's no shortage of tutorials on how to remove them), and Google's Veo 3.1 comes with a metadata watermark called Syth ID, Seedance boasts that its results are "completely watermark-free". The prevalence of advanced AI tools coupled with ease of access has opened the gates to a new wave of AI deepfakes, with the likes of xAI's Grok at the centre of the issue. Last month, the EU launched a new investigation into X to probe whether the Elon Musk-owned social media site properly assessed and mitigated risks stemming from its in-platform AI chatbot Grok after it was outfitted with the ability to edit images. Users on the social media site quickly prompted the tool to undress people - generally women and children - in images and videos. Millions of such pieces of content were generated on X, The New York Times found. Don't miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic's digest of need-to-know sci-tech news.