PixVerse pitches R2 world model for games
PixVerse is positioning R2, its new real-time world model, as the basis for AI-native games, where choices made early in a session shape what happens later.
PixVerse has published a recap of a talk by Qiushi (Chow) Li, Head of PixVerse Games, at the Tech in Asia Conference in September. The talk came the same week the company launched R2, the next generation of its real-time world model.
The post makes three claims about how AI changes games: how they are made, how they reach players, and what it is like to play one. On creation, the PixVerse Game Engine lets a creator set the rules and structure of a game while the model generates the world, so modeling, animation and scripting are no longer prerequisites. On distribution, PixVerse cites R1, where several users sent prompts into a shared live feed that shaped one world. On experience, the company says R2 keeps earlier events in a session in effect later, so player choices affect how characters behave and how the environment changes.
In PixVerse Game, players pick a genre and describe a world and characters in plain language, and the engine builds the world from that description. In one example, a player offers a non-player character either a dragon or a leaf, and the storyline branches based on the choice.
The post also gives a timeline. R1 launched in January 2026 and added shared worlds and personalized avatars three months later. The Game Engine followed in July. PixVerse says R2 is about scaling: more data, more input types and longer sessions, all still running in real time.
The post includes no benchmarks or technical details. It refers readers to the separate R2 announcement for the architecture behind coherence over long sessions. PixVerse Game now runs on R2 at game.pixverse.ai.