PixVerse R2 - A real-time world model you can explore and change

PixVerse R2 is a real-time world model that generates continuously evolving audiovisual worlds instead of fixed video clips. It accepts text, images, audio and actions while generating, remembers what happened earlier in the session, and carries those changes forward in real time. R2 scales to longer, more coherent and controllable experiences — powering everything from interactive stories and characters to playable generative worlds.

Add a comment

Replies

Best

Oh boy! Imagine this paired with a Virtuix Omni One or a Kat Walk C2 Core. I love where innovation is heading.

 I’d love to try that too 😄 There’s something about taking an actual step and seeing the world respond that a keyboard can’t quite match.

Congrats on the launch, PixVerse team!

What interests me most about R2 is the chance to explore a generated world and change it as I go. That could be useful for interactive stories and early game prototypes. I’d be curious to see how well the world holds together as people move around and make changes. How long should a change, like a new character or a shift in the weather, stay consistent? A typical response time would also help set expectations.

Thank u Gene, I’m glad to hear you like it. Yes, it works for game prototypes. You’re spot on that consistency over time is one of the key challenges we’re working on with R2. The goal is for changes, whether it’s a new character, weather, or something in the environment, to persist naturally as you keep exploring the same world. Response time can vary depending on the scene and the change you’re asking for, and we’re actively working on making the interaction faster and more consistent. Would love to hear what you think if you get a chance to try it, especially how far you’re able to push the world before it starts to feel less coherent.

Nice launch, huge congrats 👏

Curious to see what creators build with this once they start pushing it beyond the demo worlds.

 Thank you Ristan, your support mean a lot to us!

the "remembers what happened earlier and carries it forward" part is the actual hard problem here, most generative video demos reset context every clip. curious how far back the memory actually reaches before a long session starts drifting or forgetting earlier changes, is there a practical session length where consistency breaks down

 Fair question. I don’t have a validated “X minutes before drift” number to share. R2 combines recent history with longer-term scene and object memory, but that doesn’t guarantee every earlier change survives indefinitely. Our explains the mechanism, but doesn’t establish that practical limit. “Remembers” needs that qualification—there’s still work to do on long-session consistency.