R2 is an upgrade to PixVerse's real-time world model. The focus of this release is scaling: the model is capable of taking on more data, tasks, and input types while still running in real time.

PixVerse, a global AI video generation platform, today introduced R2, the latest upgrade to its real-time world model. R2's central advance is scale: a real-time world model that keeps getting more capable, more persistent, and more responsive to what a user does, without giving up the speed to run live.

This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260922037237/en/

PixVerse introduces R2, its real-time world model that keeps a generated world running, persistent, and responsive as a user moves through it.

Most AI video is generated in one shot. A model receives a prompt, produces a fixed clip, and stops. A real-time world model works differently. Instead of returning a clip, it generates a world that keeps running while a user is in it, navigable in real time and holding its state as the session continues. PixVerse introduced the world's first real-time world model, R1, in January 2026. R2 is the next step, staying coherent over far longer sessions, remembering what happens inside a session and carrying it forward, and taking text, references, audio, and action controls into the same running world.

A World That Remembers and Responds

The core change in R2 is what a user's input can do. In earlier real-time generation, input affected the current moment and then passed. In R2, input persists: an earlier action remains true later in the same session, so the world responds to what a user does rather than resetting between moments.

Explore a living world. A player enters a generated world and moves through it in real time, controlling a character with WASD keys and shifting the camera with the arrow keys. Prompts change the world as the player goes, adding elements, altering the environment, and the character interacts with the surroundings with a sense of physical logic. Each input builds on the last rather than starting from scratch.

Shape the story. The world can develop a storyline that bends to what the player does. In Zero Mark, an interactive film-game built on R2 by creator Xiaolongbao, a creature blocks the path and asks for a gift, offering it a dragon or offering it a leaf leads to genuinely different responses. The story branches from the input rather than following a pre-written path, and the model generates each outcome live.

Characters who respond in context. Characters in the world can understand a player and respond in step with the story. In one interactive story built on R2, players talk with a character who keeps a consistent identity and memory across the exchange, draws out clues, and responds in the moment with voice, expression, and movement. It reads less like a scripted non-player character and more like someone who knows you and follows what is happening. A demo of this can be found in the official PixVerse announcement.

The Scaling Problem and R2's Answer

Scaling, the practice of feeding a model more data and compute to make it more capable, has driven much of the progress in large language models. It has been far harder to apply to real-time world models, because two of their demands pull against each other. Making a model fast enough to respond live tends to make it simpler; making it more capable tends to make it slower. Historically, teams have had to choose.

PixVerse designed R2's architecture so that speed and capability no longer have to be traded against each other. Rather than forcing that choice, R2 splits the work into two stages built on the same foundation. Omni Causal AR grows a single model's capability, learning across more data, more input types, more tasks, and longer sessions. A real-time acceleration layer then compresses that same model to run live, so gains in capability no longer come at the cost of speed. The approach also improves stability over long sessions: in PixVerse's internal evaluations, R2 reduced visual drift across a long session by 35.8%. The full technical report, including architecture and evaluation details, is available here.

"Large language models showed that scaling is a reliable path to capability," said Changhu Wang, Co-Founder and CEO of PixVerse. "R2 is our case that real-time world models can follow a similar path, that with the right architecture, a model can keep getting more capable without giving up real-time interaction. Over time, this is how a creation tool becomes an environment people can enter and shape."

From World Model to Playable Games

R2 has crossed from research demonstration into something real products can be built on.

PixVerse's game engine, first launched in July 2026, now runs on R2, letting creators build worlds that players shape as they play rather than worlds built entirely in advance.

"What's exciting is what creators have been building on our world model since we opened the beta in July," said Jaden Xie, Co-Founder and President of PixVerse. "We believe our video models will empower more creators in bringing their ideas to life."

R2 is open to explore now at https://world.pixverse.video/.

About PixVerse

PixVerse is a global AI video generation platform trusted by over 150 million users across 177+ countries. With a suite of proprietary models developed entirely in-house, PixVerse empowers anyone to create cinematic-quality video from a prompt, photo, or clip. In January 2026, PixVerse launched R1, the world's first real-time world model, transforming video into an infinite, continuous, and interactive stream. R2 builds on that foundation, scaling the real-time world model to run longer, more coherent sessions. With teams distributed across Asia and the US, PixVerse was founded in 2023 with a commitment to making video the universal language of human expression. In March 2026, PixVerse closed its Series C round, achieving unicorn status. For more information, visit pixverse.ai.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260922037237/en/

In R2, input persists: an earlier action remains true later in the same session, so the world responds to what a user does rather than resetting between moments.