On the same day, Liu Xing published a post directly labeling the recent claim—“Hengdian has partnered with ByteDance to feed the portraits and voices of 150,000 extras into ByteDance’s large AI model”—as “seriously inaccurate.”
Liu Xing clarified that the true purpose of the “Hengdian Film & TV AI Service Platform” project is to establish a public digital infrastructure: transforming real-world assets like film studio settings into high-quality datasets, thereby helping production teams more efficiently select locations, plan pre-production, and shoot scenes. It also lowers the threshold for small and medium-sized film and TV enterprises to leverage AI, empowering more creators to deliver outstanding works. In other words, the platform serves the core process of filmmaking itself, rather than simply feeding facial images and voice recordings into AI models.
Data does not belong to ByteDance; portraits and voices require authorization.
He further drew clear boundaries: Volcanic Engine acts only as a technical support partner in this project, with all data assets remaining owned by Hengdian or the authorized parties. Whenever the use of anyone’s portrait or voice is involved, explicit consent must first be obtained from the individual concerned, and these datasets will never be used to train ByteDance’s large AI models.
This clarification came against the backdrop of the 2026 Hengdian Film & TV Cultural Industry Development Conference held on October 23 in Hengdian, Dongyang, Zhejiang, themed “Digital Intelligence Empowers Light and Shadow, Craftsmanship Deepens High-Quality Creations.” At the conference, the Dongyang Municipal People’s Government and Volcanic Engine announced a partnership to launch the “Hengdian π” AI platform, targeting the entire workflow of film and TV creation—from ideation and shooting to production and distribution—exploring how AI can empower diverse creative formats such as live-action dramas and short-form series. As “AI + Film & TV” rapidly takes shape, concerns about whether extras’ portraits and voices might secretly be used for training models have become particularly sensitive among industry professionals and the general public—and Liu Xing’s response was squarely aimed at addressing these very concerns.