TikTok took the world by storm. Now, Chinese companies are taking videos further with AI
Chinese companies are capitalizing on a trove of video and gaming experience to ramp up money-making AI tools for generating videos and visual effects
BEIJING — China's video-heavy entertainment world has yielded a trove of data for companies — and they're now ramping up money-making artificial intelligence tools for generating ads and film clips.
TikTok parent ByteDance holds the first and third spots in research firm Artificial Analysis' top-ranked text-to-video generative AI models, which were launched in the last two months. Google holds the second and fourth spots, while Beijing-based short video app Kuaishou's Kling AI ranks fifth.
Despite some consolidation in other parts of the AI industry, "competition in AI video generation models is at an earlier stage, and some Chinese companies have emerged as early leaders in this space," said Wei Xiong, China internet analyst at UBS Securities.
"We believe AI video generation has the potential to reshape the content industry," she said, "by enhancing production efficiency, lowering barriers to creation and unlocking new monetization models."
With such AI tools, users can upload a single image or multiple ones, and direct the AI to generate a video clip based on them. Other tools allow users to enter text, from which the AI will generate the video clip.
More than 20,000 businesses from advertisers to movie animators already use Kling AI for generating video, the Beijing-based company claimed this week during the World AI Conference in Shanghai. The latest version, Kling 2.1, can automatically add relevant sound effects to match the AI-generated video.
It's not just for users in China.
"Whether it's user scale or commercial revenue, overseas accounts for the majority," Zeng Yushen, head of operations at Kling AI, told CNBC in Mandarin, translated by CNBC. She said the company plans to enhance its support for the tool in places such as Japan, South Korea and Europe.
"This is something we've observed, AI big models are increasingly globalized," she said. "People don't seem to care which country's product it is."
Kuaishou claimed Kling AI made over 150 million yuan ($20.83 million) in revenue in the first three months of the year, and that daily advertising spend on generative AI tools was 30 million yuan during that time. The company has yet to announce when it will release second-quarter results. Zeng declined to share Kling AI's model training costs.
While the reduced production cost implies a "sizeable" market, UBS' Xiong said, "current model capabilities remain constrained by clip length, motion consistency and controllability."
Chinese video AI companies also face competition from the U.S., beyond the Trump administration's restrictions on China's access to advanced semiconductors needed for training AI models.
Amazon and Google have launched tools for generating video from images or text. The releases come as Microsoft-backed OpenAI launched its video generation model Sora to ChatGPT subscribers in December — nearly a year after it had revealed its capabilities in February 2024.
However, Kling AI had already launched to the public in June 2024. Users subscribe and buy credits to generate videos.
Vidu, a rival tool from Beijing-based startup Shengshu, launched to global users roughly 12 months ago, and around March this year said it expected annual revenue of $20 million based on user subscription fees.
"Chinese firms tend to attempt to first identify a commercial 'pain point' ..., areas where companies will pay for services, which has been a challenge for AI applications," said Paul Triolo, partner and senior vice president for China at advisory firm DGA-Albright Stonebridge Group.
He pointed to how Chinese startup 3DStyle uses generative AI to design new clothing styles and integrate them with internet-connected, automated manufacturing.
U.S. companies have also been applying AI to specific industries, Triolo said, but Chinese businesses are often able to integrate AI more quickly because they face a very competitive environment and can recruit from a "very qualified" local base of software engineers.
'AI as filmmaker'
Chinese e-commerce giant Alibaba has also stayed on top of the trend by releasing the latest version of its video generation AI model this week called Wan2.2. The company claimed that with the open-source model, users can control lighting, time of day, color tone, camera angle, frame size, composition and focal length.
Open source allows users to download a model for free, and customize, if not commercialize, products with it. Alibaba claimed that since open sourcing the "Wan" model series in February, the models have been downloaded more than 5.4 million times from the Hugging Face platform and a similar one in China called ModelScope.
Niko found that more than half of game development studios in China already use AI for content generation and reducing development time and costs.
But game development reflects broader challenges in using AI at scale for generating videos and graphics.
"While interest in AI is high," Ahmad said, "we've already seen some backlash to games that have poorly implemented the technology."
This story originally appeared on: CNBC - Author:Evelyn Cheng
