14B autoregressive world model that generates interactive, camera-controlled video worlds in 3 denoising steps.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
EVOKE is a 14B-parameter autoregressive world model for persistent, interactive world generation. It runs with three denoising timesteps and no classifier-free guidance (CFG-free). World state is kept outside the denoiser and addressed through camera pose. A long-horizon interactive teacher is used so the few-step model stays coherent over long sessions.
Users navigate the generated world with camera controls, similar to playing a game. Prompts can change during a running session, so new objects or events can be introduced without restarting. Per-step computation is bounded, so sessions can continue for hours without a growing generation context. According to a YouTube creator summary, the model takes an image plus joystick input and generates video, and the final model is about 57 GB.
A YouTube creator reports an Apache 2.0 license and open-source release, but the provided pages do not confirm this, and no repository or weights link was found. The vendor is not identified in the provided material. Release date and hardware requirements are also not stated.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source 14B real-time interactive world video model taking an image plus joystick input, generating video in three steps; supports text-prompted events. Apache 2 licensed, final model ~57 GB.
Closest neighbours by meaning, not just by category.
Model
Interactive autoregressive world model generating minute-long playable video with real-time camera control and prompt events.
Model
Real-time interactive world model with compressed memory and hybrid action and semantic control, built on Wan 2.2.
Model · Robbyant
Real-time interactive world model from Robbyant that generates controllable, long-horizon environments.
Model
Interactive world model that generates action-conditioned video rollouts in real time on a single RTX 5090.
Model
Open framework and dataset for turning video diffusion models into real-time, long-horizon interactive world models.
Model · Adobe
Adobe Research video world model that builds navigable interactive worlds from an image or video at 16 FPS.
