China’s Tsinghua University-backed Shengshu Technology has showcased a subject consistency feature for its AI video generation model Vidu, just over a month after the product became available for global users.

The AI company said the updated free feature ensures that elements from users’ uploads “remain consistent” in the video generation process, regardless of whether the model is processing human or non-human forms.

Why it matters: Shengshu appears to be positioning itself as a major player in the global AI video generation market, an industry that has seen the emergence of at least a dozen similar models in China following OpenAI’s sudden release of multiple demos produced by Sora. Sora is not yet widely available.

Details: As part of the technology updates, Vidu can now generate videos of up to 32 seconds, double its previous capacity. 

• Tang Jiayu, chief executive of Shengshu, told media attending Wednesday’s launch event that text-to-video models have uncontrollable limitations and are prone to producing videos that lack continuity.
• The technology that allows subjects to maintain coherence has had a “significant impact” on professional creative fields like animation, according to Shengshu. The Beijing-based startup has helped to bring down video production costs by approximately 97.5% in a real-world case, Tang claimed. 
• Tang also outlined the company’s business models at the event, noting that SaaS-based subscriptions and API portals are its two main revenue streams. Shengshu has received “tens of thousands of API access requests globally” since Vidu launched with support for both Chinese and English prompts on July 30. He added that revenue from the business sector surpasses that of the consumer-focused market.

Context: Like many of its Chinese counterparts, Shengshu Technology is backed by government-support funds and local tech giants. In June, it completed a Pre-A financing round worth hundreds of millions of yuan, with Beijing’s AI-centered fund, Baidu, and Ant Group among the investors.