ShengShu Unveils Vidu S1, Pushing AI Video From Fixed Clips to Live Conversation

On July 3, 2026, ShengShu Technology unveiled Vidu S1, its next-generation video foundation model, at the 2026 Global Digital Economy Conference in Singapore. ShengShu was founded by Tsinghua professor Zhu Jun, and the Vidu S1 project was led by his PhD student Zhang Jintao. The release marks the company's first model built entirely around live, two-way conversation rather than the offline clip workflow that has defined AI video since 2023.

Traditional video generators take a text prompt and, after a wait, return a finished clip of fixed length. Vidu S1 instead accepts voice input and, when the camera is on, the scene itself, then generates frames on the fly as the conversation continues. The default pipeline runs at 540p resolution and 25 frames per second, climbing to 42 frames per second when extra compute is available. A user can upload a single image, pick a voice, and start talking to a character that listens, responds, and changes expression within a few hundred milliseconds.

Inside the model, Vidu S1 fuses an autoregressive design with a diffusion renderer: instead of producing a full video in one pass, it predicts the next frames from the history of frames already produced, the current voice signal, and the live dialogue context. The serving layer leans on TurboDiffusion for fast per-frame inference and on TurboServe for streaming deployment. Together, those two pieces keep single-frame cost low enough that the conversation can run for hours on consumer-grade GPUs without the character drifting or losing its identity.

ShengShu is positioning Vidu S1 as the moment AI video shifts from content generation to interaction generation. The first use cases it names are video calls, live streaming, virtual companions, and game characters. A public beta is already open on the Vidu website and an early API is live on the Vidu developer platform, though pricing and a wider enterprise rollout have not yet been disclosed.