SenseTime Unveils Fast Image Model Amid Sanctions

SenseTime, the Chinese artificial intelligence company widely recognized for its facial recognition systems, has introduced a new open source model that it says can generate and interpret images far more quickly than leading competitors in the United States. Released on Tuesday, the model, called SenseNova U1, could help the company regain traction after falling behind in China’s rapidly evolving AI sector.

What sets SenseNova U1 apart is its ability to process images without first converting them into text. This shortcut speeds up performance and reduces the computing power needed. “The model’s entire reasoning process is no longer limited to text. It can reason with images as well,” said Dahua Lin, cofounder and chief scientist at SenseTime, in an interview with WIRED. Lin, who also serves as a professor of information engineering at the Chinese University of Hong Kong, believes this direct image processing capability will one day allow robots to better interpret the physical world.

Like DeepSeek’s latest flagship model, SenseNova U1 can run on Chinese-made chips. “Several Chinese domestic chipmakers have finished optimizing compatibility with our new model,” Lin noted. On launch day, ten Chinese chip designers, including Cambricon and Biren Technology, announced their hardware now supports U1. This flexibility is crucial because US export controls restrict Chinese companies from accessing the most advanced AI chips, especially those used for training, which are primarily produced by Western firms like Nvidia. “We will continue to push for training on more different chips,” Lin said, though he acknowledged that SenseTime “may still need to use the best chips to ensure the speed of our iteration.”

SenseTime released U1 for free on Hugging Face and GitHub, signaling how Chinese companies are increasingly becoming major contributors to the open source AI community. Founded in 2014, SenseTime once dominated computer vision for applications like facial recognition and autonomous driving. But when ChatGPT and other natural language processing models surged in popularity, the company struggled to turn a profit and fell behind newer Chinese startups such as DeepSeek and MiniMax.

By making SenseNova-U1 publicly available, SenseTime hopes to catch up with both domestic and Western AI players. Lin said the company decided last year to embrace open source, largely because of the valuable feedback it receives from researchers, which allows faster iteration. “In this day and age, being open source or closed source is not the winning factor; the speed of iteration is,” he explained.

Going open source also helps SenseTime maintain international research collaborations despite geopolitical tensions. The company has been repeatedly sanctioned by the US government over allegations that its facial recognition technology powered surveillance systems used to monitor and detain Uyghurs and other minority groups in China’s Xinjiang region. As a result, US firms are restricted from investing in SenseTime or selling certain technologies to it without a license. SenseTime has denied the allegations.

In a technical report released alongside the model, SenseTime claims that SenseNova-U1 produces higher-quality images than any other open source model currently available. Its performance is comparable to leading Chinese closed source models like Alibaba’s Qwen and ByteDance’s Seedream, though it still falls short of industry leaders such as GPT-Image-2.0, which launched just a week ago. The model’s main advantage, however, is its speed. It relies on an innovative technical structure called NEO-Unify, which SenseTime previewed earlier this year.