Video generators give you a clip to watch. World models go one step further: they generate a scene you can step into and control, frame by frame, in response to your inputs. That's why people call them "playable worlds."
I kept losing track of which of these systems were actually usable, so I built PlayableWorld AI, a small independent hub that tracks them with links to official sources. Here's the short version of what I've learned.
Google DeepMind Genie 4 and Project Genie
Genie 4 is DeepMind's latest world model. Right now it's a research preview with limited, invite-only access. Project Genie is the Google Labs experience built on top of it, tied to a paid subscription tier. There's no public API yet, so if you're a developer the honest answer is "watch and wait." I keep the current status and eligibility notes on the site and update them when Google changes anything.
Alternatives you can use now
While Genie gets the headlines, several teams ship world-generation tools you can try today:
- World Labs Marble generates explorable 3D worlds from images or text.
- Tencent HunyuanWorld is an open model for generating 3D scenes.
- Skywork Matrix-Game focuses on interactive, game-like world generation.
- Odyssey works on interactive video you can steer in real time.
Each has different trade-offs in latency, control and output format (interactive video vs. exportable 3D). Prices change often, so I check them against the providers' own pages rather than repeating numbers from social media.
Why this matters for devs
If world models get fast and cheap enough, they could change prototyping for games, robotics simulation and training data. Today they're best for experiments and demos — but it's a good time to learn the vocabulary.
Follow along
Everything is collected at playableworldai.com, including official DeepMind demos and an optional one-email notice when a Genie API opens. Corrections and tool suggestions are very welcome in the comments.
