PixVerse launches R1 real-time world model
PixVerse has released R1, which it calls the first real-time world model, generating continuous, interactive 1080p video that responds to user input rather than rendering fixed-length clips.
PixVerse, a Singapore-based AI video platform that says it has more than 100 million registered users, announced R1 on January 12, 2026. The company describes it as "the world's first real-time world model" for interactive AI video. Nobody has independently verified that claim, and the release includes no benchmarks.
Most video generators turn a prompt into a fixed clip, and the user waits for it to render. PixVerse says R1 instead runs as a persistent, stateful simulation. It produces continuous 1080p video, audio and world state together, and it changes in response to user input as it plays.
According to the company, three pieces make real-time output possible:
- An "Omni Native Multimodal Foundation Model" that treats text, image, audio and video as one token stream.
- A "Consistency-aware Autoregressive Framework" for video of unlimited length that stays coherent over time. PixVerse says memory-augmented attention keeps scenes physically consistent over long stretches.
- An "Instantaneous Response Engine" that cuts the number of sampling steps from dozens to between one and four.
PixVerse names AI-native games, in which environments and non-player characters react to the player, as one target use. Others are interactive entertainment and live streams that viewers steer by voice or gesture, plus enterprise training simulations and product visualization.
R1 is available at realtime.pixverse.ai, and qualified partners can get enterprise API access. The announcement does not give latency figures, frame rates, pricing, hardware requirements or model size.