Project Genie free? Access, Genie 3 and Genie 4 explained

Is Project Genie free?

Google’s January 2026 Project Genie announcement described an experimental application based on Genie 3 for Google AI Ultra subscribers in the US, aged 18 or over. That was not a general free launch. On 27 October DeepMind announced Genie 4 with a research preview and wider access coming soon. Neither announcement establishes a free public Genie 4 tier.

Searches for “Project Genie free” can also surface unrelated products. Check that the page belongs to Google or is linked from DeepMind’s official Genie page. A browser game carrying the word Genie is not evidence of access to this world model.

Project Genie is the application; Genie is the model

The distinction explains why screenshots and older tutorials can disagree. Project Genie provides the creation and exploration workflow. Genie 3 powered its January prototype. Genie 4 is the new model announced for a later rollout into that application. A tutorial showing World Sketching does not prove that every account has been upgraded.

The initial workflow included sketching a world, exploring, remixing and downloading video. Its preview image used Nano Banana Pro, and generation was limited to 60 seconds. These are historical details from the Genie 3 prototype, not confirmed specifications for Genie 4.

How to prepare an original scene

Start with a world description that has one clear subject and an environment that can be explored. Here is an untested editorial example: “An empty seaside rehearsal hall at dawn. A stagehand walks from the doorway to one red chair. Eye-level camera advances slowly, then stops. Rain is audible outside; footsteps remain inside.”

Add camera constraints separately: keep the stagehand in view, avoid cuts and preserve the doorway position. These are instructions to evaluate later, not promises of obedience. DeepMind’s examples show detailed camera and sound direction, while its limitations explicitly acknowledge drift after complex edits.

If you want to explore an archival photograph, record what is visible in the source image and what the model would have to invent. A view behind a wall is a generated hypothesis. In a documentary pitch, label that distinction before the image becomes a believable moving reconstruction.

Which upgrades belong to Genie 4

Google highlights native real-time audio, more natural people and physics, live text direction and an agent loop using Gemini 3.8 Flash. A new character can respond in dialogue while the scene continues. This goes beyond treating the output as a fixed clip, but it does not make the world a dependable game engine.

The announced limitations remain practical: only a few minutes of continuous interaction, uncertain object permanence, imperfect real places and ongoing work on audio. We have not tested either the new audio or live direction. Do not present a prompt recipe as an independently verified result.

The sensible next step

Check the official waitlist and your account’s current access information. A Google AI Ultra price is a subscription quote, not confirmation of a free trial or an immediate Genie 4 session. The announcement does not supply a country rollout calendar or API tariff.

For an experiment, keep one baseline scene and change one instruction at a time: first camera, then lighting, then an action. Save the initial description and the observed failures. This is our proposed evaluation method for future access; it has not been run on Genie 4.

Related guide

Genie 4: world model, access and what Google has shown

Sources

Official documentation and demonstrations; checked on 28 October 2026.

• Google DeepMind: Genie 4 - 2026-10-28
• Genie: capabilities, limitations - 2026-10-28
• Google: introducing Genie 4 - 2026-10-28
• Project Genie (2026-01-29) - 2026-10-28
• Project Genie: prompt guide - 2026-10-28
• Genie 3 (2025) - 2026-10-28
• Google One - Google AI Ultra - 2026-10-28