Shengshu releases Vidu S2: real-time avatars and streaming video editing

On Sept 15, Tsinghua-affiliated Shengshu released Vidu S2, a real-time interactive video generation model, only 69 days after S1, with same-day full availability plus online demo and API. It pushes video generation from "generate then use" toward generating, interacting and editing simultaneously.

Confirmed

• S2-Avatar (real-time avatar): generates an interactive avatar from a single reference image, upgraded from S1's 540P to 720p at 25–42FPS; supports voice-driven dialogue and voice-controlled complex actions (e.g. dancing), and accepts new reference images mid-conversation for outfit changes, object pickup, or scene/background switching while keeping motion continuity — suited for livestreaming and virtual personas.
• S2-Editing (real-time editing): swaps outfits, characters, backgrounds and styles on a playing video stream in real time; QbitAI's hands-on test found character consistency fairly natural and garment deformation acceptable with occasional artifacts. Shengshu cites e-commerce virtual try-on, IP virtual-idol livestreams and cross-border e-commerce staging as target use cases.
• Spatial video: billed as the first real-time interactive spatial video generation and editing system, aimed at VR, supporting dynamic reference updates during generation.

Why it matters

Real-time avatars and streaming editing map directly onto livestream, e-commerce try-on, virtual idol and VR scenarios, and same-day open access with demo and API lowers adoption barriers. The 69-day iteration gap also signals how competitive the real-time video generation race has become.

2026-09-15 ~ 2026-09-15 · 9 related posts

• Vidu S2 launches: real-time interactive, editable, spatial video generation — Tsinghua · 2026-09-15
• Shengshu launches real-time video model S2: 720p avatars and live video editing at 25-42 FPS, plus spatial video for headsets — 赛博禅心 · 2026-09-15
• Vidu S2 launches as first real-time interactive spatial video generation system for VR — Scobleizer · 2026-09-15
• Vidu launches S2: real-time AI video generation you can talk to and edit live — Scobleizer · 2026-09-15
• [source] Shengshu's Vidu S2 edits live video streams in real time — swap outfits, faces and scenes — 量子位 · 2026-09-15
• [source] 69 days after ViduS1, Shengshu ships ViduS2 with day-one access — 生数科技 · 2026-09-15

3 near-duplicate retellings: 生数科技 · _akhaliq · iamaliveix