PixVerse R2
A real-time world model you can explore and change
780 followers
A real-time world model you can explore and change
780 followers
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.
Hey Product Hunt! 👋
We’re the PixVerse team. Meet PixVerse R2: a real-time world model you can explore and change while it’s being generated.
Explore. Change. Keep going.
Explore with WASD. Move through the world and choose where to go next.
Change it with a prompt. Introduce a new element or change the scene as it unfolds.
Keep going. Continue interacting in the same evolving world after each change.
A collection of R2-powered worlds is now open to explore. Try R2 →
How R2 works
R2 builds its world-model capabilities through continuous pretraining with Omni Causal AR, then distills that same foundation for real-time interaction. The aim is to make worlds more persistent, controllable, and scalable.
Read the technical report →
Whether you’re experimenting with interactive stories, game prototypes, or AI worlds, we’d love your feedback. What would you explore or change first?
The gallery combines AI concept scenes with an overview of R2’s architecture.
@jadenx Shoutout to this new launch, Jaden! Keep up the great work!
@jadenx Can R2 remember previous interactions across longer sessions ?
@jadenx @dipanshu_kushwaha5 Within an ongoing session, yes. R2 keeps recent interaction history alongside longer-term context, such as characters and the scene. It won’t remember every detail perfectly, though, and keeping things consistent over longer sessions is something we’re still improving.
I can see this being pretty fun for interactive stories. You could probably take the same world in completely different directions. How much control does the user have?
@hehe6z Thanks, glad you had fun! R2 uses both recent interactions and longer-term context, like character identity and scene details, to guide what comes next. We also train it on examples where things drift, so it can learn to recover. It’s not perfect yet, and consistency over longer sessions is still a big focus for us.
@hehe6z Thanks Helena, I know fish audio, a star startup that's doing exceptionally well in Silicon Vally, as a former Chinese female entrepreneur, I'm also proud of you all!
This is just crazy where we're heading 😮 This looks really great! Congrats on the launch! The landing hero kinda made me feel a bit dizzy because some are really moving fast visuals so it was hard to focus on the text (but this just might be me). Otherwise you do a really good job at explaining the product upfront. The one other minor detail I noticed is the "muted" icon when I go to the gallery, that stripped musical note is not very common and I honestly though it looked a broken icon until I clicked it and saw the complete icon. Maybe use the same icons as you find in YouTube etc.? This is my 2c, wish you all the best with the launch!
Really fun demo. I spent way longer than expected just testing random prompts 😅
@cheng_ju1 Glad to hear you already have fun with R2🫶