SHELFCRITTER AI Digest

MODEL

Vidu S2

modeltopic-notevideo-generation

Overview

Vidu S2 is the successor to Vidu S1 (arXiv:2607.03118, real-time 540p @ 42 FPS via TurboDiffusion + TurboServe). Announced as a two-part release: S2-Avatar streams real-time 720p interactive digital-character video with hot-swappable references and dance-level instruction following, and S2-Editing performs live style / clothing / character / background replacement on video streams. Structural claim: real-time editable 720p generative video moves diffusion video from batch renders to interactive apps — the pipeline latency floor practitioners have been waiting on for streaming-video use cases.

Timeline

  • 2026-09-15-AI-Digest — Vidu S2 paper “Real-Time Interactive, Editable, and Spatial Video Generation” surfaces on arXiv (2609.11638, ▲76). Two-part release: S2-Avatar streams real-time 720p interactive digital-character video with hot-swappable references and dance-level instruction following; S2-Editing does live style / clothing / character / background replacement on video streams. Extends the Vidu S1 real-time 540p / 42 FPS interactive generation thread from July 2026 by one resolution tier and by adding a distinct editing head to the same substrate. Practitioner framing: real-time, editable 720p generative video moves diffusion video from batch renders to interactive apps — the pipeline latency floor practitioners have been waiting on for streaming-video use cases. Log against MOC - Open Source Models.

Key Developments

  1. Real-Time Interactive Editable 720p Generative Video — S2-Avatar + S2-Editing (September 15, 2026): Vidu S1 shipped real-time interactive video generation at 540p at up to 42 FPS on consumer GPUs; Vidu S2 (arXiv:2609.11638) extends the substrate to real-time interactive 720p with a two-part release. S2-Avatar streams 720p interactive digital-character video with hot-swappable references and dance-level instruction following. S2-Editing performs live style / clothing / character / background replacement on video streams. Load-bearing framing: the pipeline latency floor practitioners have been waiting on for streaming-video use cases is now legible — real-time, editable 720p generative video moves diffusion video from batch renders to interactive apps. Covered in 2026-09-15-AI-Digest.

See also: Vidu S1, Kling AI, Veo, MOC - Open Source Models.