Genie by DeepMind (not publicly released)

Genie is DeepMind's line of research into generative world models — systems that turn a text prompt or an image into an interactive environment you can move around in, rather than a video you can only watch. DeepMind publishes about it and demonstrates it. It has not released it. That distinction is the whole point of this page, and we previously erased it. This listing previously said Genie was a "completely free research release with model weights" and included "model checkpoints and inference code". None of that is true. We read DeepMind's Genie page in full on 7 September 2026 and it contains no download, no weights, no access programme and no waitlist. We previously claimed generated worlds run for "hours" and offer "infinite replayability"; both were ours, not DeepMind's, and neither is something we could test on a model we cannot obtain. What DeepMind does describe is the research direction: generating navigable environments from a prompt, holding them consistent while you interact with them, and using them to train agents — work that matters more for robotics and agent research than for making games. Judging how well any of it works from published demonstrations is not the same as using it, and we are not going to pretend otherwise. If you are looking for something you can actually use to build a game with AI assistance today, our games category lists tools that are released. This page stays as a record of a research programme, and it comes off our free tools page because nothing here is available to be free. Browse related picks.