01Model
A World Built to Remember, Respond, and Last
EVOKE is a 14B, 3-step CFG-free autoregressive world model for persistent, interactive world generation. It decouples world state from generation: persistent state lives beyond the denoiser and is addressed through camera pose, while a long-horizon interactive teacher gives the few-step model the ability to stay coherent and respond to changing instructions over extended sessions. The result is a world model that can remember, respond, and keep going—for hours.
- 14B
- Parameters
- 3-step
- CFG-free sampling
- Hours
- Continuous session
02Capabilities
What the model does
03Showcase
All cases are generated in 3 timesteps
InteractiveTake control and steer the world as it unfolds.
60 s
5–30 s
Evoke on DemandChange what happens next without stopping the world.
Non-interactiveStep back and watch the world evolve on its own.
EmbodiedThe same model generating in embodied scenes.
