Real-Time Interactive Video Model X2.0 Released

量子位 · wechat · 2026-07-15

量子位 covered XmaxAI's global debut of the real-time interactive video model, X2.0. The core update isn't just “generating video,” but transforming video into a real-time interactive medium: users can control characters, clothing, backgrounds, and objects via camera, touchscreen, text, and spatial movements.

Key capabilities highlighted include:

Technologically, the article emphasizes its use of streaming frame-by-frame autoregressive generation rather than traditional joint generation of entire video segments. It balances real-time performance, quality, and cost through multi-stage distillation, context persistence, sparsification, and FP8 quantification. The model can run on consumer-grade GPUs and iPhones locally, with an open API.

Application examples include virtual try-ons, e-commerce, virtual companionship, IP interaction, live streaming, smart hardware, and cultural tourism.

Related event: XMAX launches real-time interactive video model X2(4 posts)→

Original post →

More from Multimodal

Multimodal channel →