Chinese AI Firms Lead US Counterparts in Video Generation

Chinese artificial intelligence companies are demonstrating a notable lead over their US counterparts in video generation. This position arises from extensive short video ecosystems, less stringent copyright regulations, and aggressive pricing strategies. Analysts suggest compute capacity alone does not guarantee success in this sector.

Credit: UNSPLASH

Models from Chinese firms, including Alibaba Group Holding, MiniMax, and ByteDance, presently hold dominant positions in benchmark rankings on the industry evaluation platform Artificial Analysis. These organisations exhibit a concerted effort in advancing video AI capabilities.

Alibaba's Wan 3.0 secured the top position on Artificial Analysis' text to video leaderboard, which includes audio. Google Gemini Omni Flash was second placed. But, MiniMax H3, a version post trained by US platform fal, ranked third, followed by the original open weight H3 and ByteDance's Seedance 2.0. Chinese models occupy eight of the top ten positions.

Chinese models also lead across several related categories. These include image to video and video editing, determined by blind user comparisons using identical prompts. Alibaba owns the South China Morning Post.

Industry analysts attribute China's momentum in video AI to a combination of proprietary training data, rapid iteration cycles, and strong domestic demand for short form digital content. Kyle Chan, a fellow at the Brookings Institution's John L Thornton China Centre, noted that Chinese companies like ByteDance have prioritised video generation.

And, ByteDance has established a positive feedback loop, developing AI generated short form video into a genuine industry, according to Chan. Early successes cultivated strong viewer demand. This prompted ByteDance to expand the technology into a profitable business line.

Domestic downstream commercial demand has provided further support. China's short drama and AI generated comic drama industries have seen rapid development, with Chinese production teams also supplying content for overseas markets, said Wang Zebin, an investment manager at private equity firm JG Investment.

Analysts indicate that diverging approaches to copyright risk have shaped a distinct operational environment. US AI developers face substantial potential risk and liability regarding intellectual property, Chan stated.

Conversely, China's more flexible copyright enforcement provided domestic firms greater opportunity to test and commercialise their models, Wang explained. So, Shenzhen based AI enthusiast Cecilia Huang noted Google's Gemini rejected her request to convert a photo into a Ratatouille character for a 15 second video on copyright grounds, whereas ByteDance's Doubao fulfilled the request.

ByteDance previously informed the South China Morning Post in Feb. that it "respects intellectual property rights". The organisation also reported it was "taking steps to strengthen current safeguards" to prevent the unauthorised use of intellectual property and likeness by users, in response to allegations concerning Seedance 2.0's training content.

Open weight Chinese devs also compete aggressively on cost. MiniMax's H3, which can generate 15 second 2K videos with stereo sound, is priced at one third of mainstream competing products, according to the company.

And, this model became available through ByteDance's Dreamina creative platform, granting it access to global creators. Artificial Analysis estimates, based on default API settings, suggest generating one minute of 1080p video costs around US$7.80 via MiniMax H3.

Google's Gemini Omni Flash costs US$6 for the same duration. ByteDance's Seedance 2.0 costs US$9.07. But, effective commercial prices for Chinese models are often discounted heavily through enterprise agreements, with Wan 3.0 and Seedance 2.0 trading at substantial reductions from their published rates.

This technical and pricing advantage contributes to top line revenue. Kuaishou Technology reported its Kling AI video generation platform generated over 850 million yuan (US$118 million) in second quarter revenue.

So, this represents an increase of more than 200 per cent year on year. By June, Kling AI's user base had topped 100 million globally, up about 67 per cent from late 2025.

Its enterprise client roster expanded by roughly 67 per cent to nearly 50,000 businesses, according to Kuaishou. Kling 3.0 1080p Pro version ranks ninth on Artificial Analysis' text to video leaderboard for audio.

• Chinese AI models dominate video generation benchmarks on Artificial Analysis.

• Alibaba, MiniMax, and ByteDance models occupy eight of the top ten positions.

• Factors include proprietary data, rapid iteration, and domestic demand for short form content.

• Divergent copyright approaches and competitive pricing contribute to this lead.

• Kuaishou's Kling AI platform reported over US$118 million in second quarter revenue.

Source: SCMP