The company behind TikTok has developed an artificial-intelligence model that can turn a single text prompt into a high-quality video with a story line, scene changes and distinctive characters.
The new AI video-creation model from Beijing-based ByteDance is generating buzz in China and a backlash in Hollywood over copyright issues. It shows how ByteDance, known for creating TikTok, is emerging as a rival to OpenAI and Google in the race to build tools for making AI movies and other video entertainment.
“China’s visual models have been very, very competitive,” said Steve Long, a videogame developer in Helsinki who has participated in beta-testing programs for Google and ByteDance.
ByteDance recently ceded control of the U.S. version of TikTok to an investor group to resolve U.S. national-security concerns.
Global users of another ByteDance app, CapCut—a popular video-creation and editing app—will soon have access to Seedance 2.0, its latest AI model for creating videos, the company said. The model is already available to users on CapCut’s Chinese version, Jianying.
Given a text prompt, Seedance 2.0 can arrange a coherent story line, create realistic voice-overs and background sounds, and generate complex character actions.
“Storyboarding has long been considered the one area where human thinking and innovation really shone,” said Liu Yiran, a Hangzhou-based film director who has tested the new Seedance model. “It’s now been proven that AI can replace it.”
The ByteDance video-creation tool competes with OpenAI’s Sora and Google’s Veo, which have also wowed users with their ability to create cinematic videos from text.
Seedance 2.0 has a limitation for now: It can only generate videos of up to 15 seconds. Sora’s clips are limited to 25 seconds for Pro subscribers.
Some said ByteDance’s tool wasn’t ready for prime time. They cited video-generation glitches and said people would need expertise in video editing and prompt writing to generate Hollywood-level videos.
Other critics have raised privacy and copyright concerns, with echoes of controversies over deepfake images generated by Elon Musk’s Grok.
Tim Pan, a filmmaker and founder of a popular Chinese tech channel called MediaStorm, said the ByteDance model produced audio nearly identical to his own voice based only on a photo of his face, even though he hadn’t given the tool a voice sample. Pan questioned whether ByteDance had fed videos of him into its model for training and raised concerns that someone could use ByteDance’s tool to forge the identity of public figures.
Shortly after Pan posted a video about the issues, ByteDance said it was making urgent changes after receiving user feedback. It suspended a feature that allowed the creation of digital avatars based on real people.
The Motion Picture Association, which represents major Hollywood studios, said the model used U.S. copyrighted works without authorization “on a massive scale.”
“ByteDance is disregarding well-established copyright law that protects the rights of creators and underpins millions of American jobs,” the association said Thursday.
On Friday, Disney sent a cease-and-desist letter to ByteDance, saying the Chinese company “is hijacking Disney’s characters,” including ones from “The Avengers” and “Star Wars.” Disney called the behavior “willful, pervasive, and totally unacceptable.”
ByteDance said it respected intellectual-property rights. “We are taking steps to strengthen current safeguards as we work to prevent the unauthorized use of intellectual property and likeness by users,” the company said.
ByteDance is among many Chinese AI developers introducing new models and features ahead of the Lunar New Year starting Tuesday. AI heavyweights DeepSeek and Alibaba are preparing to release their next-generation models soon.
While principally known outside China for TikTok, ByteDance is a wide-ranging AI powerhouse in its home country, with a chatbot and tools for image and video generation.
Daily usage of ByteDance’s flagship AI models is approaching that of Google and OpenAI based on the number of tokens used, a measure of AI workload, according to data released by the companies. ByteDance’s chatbot in China, called Doubao, has nearly 250 million monthly active users, according to data tracker QuestMobile.
Getting enough computing power to handle all that work—especially data-heavy video generation—is a challenge.
ByteDance is nearing a deal to use AI servers containing more than 7,000 Nvidia B200 chips at a data center in Indonesia, people familiar with the matter said. ByteDance plans to use the chips for AI research and development, some of the people said. The company is also seeking to expand data-center capacity in Southeast Asia, they said.
The chips—from Nvidia’s flagship Blackwell series—can’t be sent to China because of U.S. export controls.
The Seedance model was developed by a lab called Seed that ByteDance set up in 2023 with the goal of pursuing artificial general intelligence. The unit is now led by Wu Yonghui, who was a senior researcher at Google before joining ByteDance early last year and is now based in San Jose, Calif.
While Chinese-origin chatbots are unlikely to challenge U.S. rivals such as ChatGPT because they are expensive to train and run, “Chinese apps stand a solid chance to lead in select segments, such as photo and video editing, text-to-video generation, etc.,” said Laila Khawaja, director of Gavekal Technologies.
CapCut has 642 million monthly active users, according to market-intelligence firm Sensor Tower.
“Chinese companies have more proprietary data and solid experience” in these types of apps, Khawaja said in an email. “We could see a replay of [the] TikTok story in AI—Chinese-origin AI app making a global impact.”
Long, the Helsinki videogame developer, cautioned that Chinese developers were held back by U.S. controls on AI chips, which could make further advances more difficult. And, he said, U.S. consumers may hesitate to use a Chinese model over national-security and cybersecurity risks.
Comments