Meta's Horizon Create and Studio Bring Prompt-to-Game AI Tools to Phones

What Meta announced

Meta introduced two tools aimed at getting more people to make games for its Horizon platform. Horizon Create is a mobile app where you describe a game in plain language and an agentic engine builds it. Horizon Studio is a browser app that offers more granular controls. Per The Verge, both let you create games from AI prompts.

According to Meta's developer blog, creators can start from a blank canvas or a template, and the output can be a 2D or 3D game. Meta describes the loop as: describe an idea, play the generated build, ask for changes, repeat.

Two surfaces, one backend

The design choice worth noticing is the split. Meta says everything built in Horizon Create carries over into Horizon Studio and vice versa, including assets, systems and iteration history, because the two share a common backend. Studio is described as a browser workspace for visual editing: arranging scenes, swapping assets, adjusting script parameters and shaping layouts.

That is the shape most AI-assisted development tools are converging on. A prompt-first surface gets you to something playable fast. A precise surface lets you take back control when the generated result is almost right but not quite.

Why it matters to builders

• Prompt-to-product is becoming a distribution strategy. Meta says games published through the tools are eligible for distribution across Facebook, Instagram and Horizon, so the tooling is tied directly to the audience.
• Iteration history is a first-class asset. Keeping prompts, changes and assets portable between surfaces is what makes agentic workflows recoverable rather than one-shot.
• The review step does not go away. Generated games still need testing for behaviour, balance and safety. Meta's tools generate; someone still has to judge what ships.

What we do not know yet

Meta opened a waitlist on September 24, so independent, hands-on evidence is limited. We have not seen the underlying models, pricing, creator terms or content-moderation approach confirmed in the sources we could verify, and we are not going to guess. Claims about output quality should wait until people outside Meta have shipped something with these tools.

Takeaway for vibecoders

If you build with AI assistants, the lesson is portable: pair a fast conversational loop with an inspectable, editable layer, and keep the history of both. Whatever the domain, whether games, dashboards or trading tools, the teams that get reliable results are the ones who can see and correct what the agent produced.