Tsinghua University and Shengshu Technology Unveil ‘Vidu S2’ Real-Time Interactive, Editable, and Spatial Video Generation Model, Boosting 720p Generation and Instruction Following

September 20, 2026

arXiv (Tsinghua University, Shengshu Technology) China

Overview

Tsinghua University and Shengshu Technology researchers have announced ‘Vidu S2,’ a real-time interactive, editable, and spatial video generation model on arXiv. Comprising Vidu S2-Avatar (real-time interactive digital character model) and Vidu S2-Editing (real-time video editing model), Vidu S2 significantly expands the feasibility of real-time spatial video generation. Compared to its predecessor, Vidu S1, Vidu S2-Avatar supports real-time 720p video generation, dynamic reference-based generation, and enhanced instruction following, achieving substantial performance improvements.

In Depth

Key Findings

Researchers from Tsinghua University and Shengshu Technology have unveiled ‘Vidu S2’ on arXiv, an innovative spatial video generation model capable of real-time interaction and editing. Vidu S2 features real-time 720p video generation, dynamic reference-based generation, and significantly enhanced instruction following capabilities. It achieves substantial performance improvements over its predecessor, Vidu S1, particularly in real-time interactive digital character generation (Vidu S2-Avatar) and real-time video editing (Vidu S2-Editing).

Technical Details

Vidu S2 is primarily composed of two sub-models. First, ‘Vidu S2-Avatar’ enables the real-time generation and interactive manipulation of digital character videos based on user text prompts or other inputs. Generating at a high resolution of real-time 720p significantly enhances visual realism and immersion compared to previous models. Second, ‘Vidu S2-Editing’ provides the capability to apply real-time edits to existing video content, allowing users to quickly modify scenes or manipulate objects. A notable technical strength of this model is its ability to generate and edit videos while maintaining spatial consistency.

Background & Context

The field of AI-powered video generation is advancing rapidly, but simultaneously achieving real-time interactivity, editability, and high-resolution spatial consistency has been a long-standing challenge. Creative industries, including film production, game development, virtual reality (VR), and augmented reality (AR), have a strong demand for tools that allow users to freely manipulate generated video content and receive immediate feedback. While Vidu S1 demonstrated the potential of video generation, Vidu S2 represents a crucial milestone in addressing these requirements.

Strategic Significance & Outlook

The announcement of Vidu S2 heralds a new era in AI-driven video content creation. Its real-time interactive video generation and editing capabilities will empower not only professional content creators but also general users with advanced production tools, greatly expanding creative freedom. In the future, technologies like Vidu S2 are expected to play a central role in diverse application areas, such as natural interaction with avatars in metaverse environments, AI-driven live streaming, and the creation of personalized media experiences. The evolution of this model holds the potential to fundamentally transform how digital content is produced and consumed.

Source: https://arxiv.org/html/2609.11638v1

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.