Baidu announced the launch of a new artificial intelligence (AI) video generation model on Wednesday. The MuseStreamer AI model has the capability to generate videos with integrated Chinese audio, making it the second model of its kind after Google’s Veo 3. The company claims MuseStreamer to be the first AI model globally to support native Chinese audio generation. In addition to the AI model, Baidu also introduced a new video content creation platform named HuiXiang. Currently, both MuseStreamer and HuiXiang are not accessible outside of China.
Baidu’s MuseStreamer Can Reportedly Generate Chinese Audio
The AI video generation landscape has made remarkable strides over the past two years. Earlier models struggled with tasks like accurately rendering people, while the latest large language models (LLMs) demonstrate an ability to depict realistic physics and movement. However, the integration of native audio support in video generation has remained largely uncharted territory for many AI developers.
At Google I/O 2025, Google became the first to introduce this functionality with its Veo 3 model, which gained immediate attention and surpassed its main competitor, OpenAI’s Sora. The company has since expanded Veo 3’s availability to 154 countries through the Gemini app, showcasing its commitment to advancing this technology.
According to a report from Tech in Asia report (via AI Base), Baidu has entered this competitive arena with the release of its MuseStream AI model, which uniquely generates videos with Chinese audio—an area where Veo 3 only offers English language support.
MuseStreamer is reported to generate synchronised dialogues alongside sound effects and ambient sounds in videos. Baidu claims the model has achieved a benchmark score of 89.38 percent on the VBench I2V test, placing it at the forefront of its category. The company is marketing this LLM as a tool for consumer content creation.
Along with the launch of MuseStreamer, Baidu has also introduced HuiXiang, a video content platform designed to work in conjunction with the AI model. HuiXiang allows users to input prompts and generate videos. The platform currently offers 10-second video generation at 1080p resolution, surpassing Veo 3, which is limited to eight-second videos. Details regarding the default aspect ratio of the videos or options for generating videos in different formats have not been specified.
