Genie 4 · access status
Genie 4: Access Status and a Free World Prompt Builder
Genie 4 is Google DeepMind's world model, announced on October 27, 2026. It is not publicly available: it runs as a research preview for trusted testers, with Google AI Ultra subscribers in Project Genie and early access partners through the Gemini API and Google AI Studio planned next. There is no public Genie 4 endpoint, download or published price. This page tells you exactly where access stands, and builds the four-block prompt you will paste in when you get it — starting today.
- Access checked
- 2026-10-29
- Builder
- Four blocks, no model call
- First-frame check
- Runs in your browser
Free, no account · Nothing is uploaded · No model is called
What Genie 4 actually is
Genie 4 generates an interactive world rather than a video clip. You move through it with normal keyboard controls and change it while it runs: switch the weather, move the camera, add a person or an object. Audio — music, ambience and interactive dialogue — is generated with the world instead of added later, and a reference image can be used as the exact first frame.
Two consequences matter for anyone arriving here from a search for “Genie 4 online”:
- The output is a running world, not a file. Nothing is downloaded, so "Genie 4 download" is not a thing that exists.
- Access is gated by Google, not by prompt quality. No third-party site can give you Genie 4 because a site cannot host a model it does not run.
Google's own launch post is the primary source for all of the above; the official prompting guide covers the environment/character structure used below.
Access status (checked 2026-10-29)
| Question | Status today |
|---|---|
| Who can use Genie 4 right now? | Trusted testers in a research preview |
| How does everyone else get in? | The official waitlist; Google says AI Ultra subscribers get access in Project Genie "in the coming weeks", and early access partners through the Gemini API and Google AI Studio |
| What does it cost? | No standalone Genie 4 price is published. The announced consumer route is a Google AI Ultra subscription, and a subscription is not a guarantee of immediate access |
| Is there a public API? | Not generally available — early access partners only |
| Is there an app or a download? | No. There is no Genie 4 app, installer or local build |
| Can this site run Genie 4 for you? | No. We do not have Genie 4 access, and we will not pretend to |
We re-check this table each week; the date above is the last check, not a promise about tomorrow. A Google AI Ultra subscription is the announced consumer route and is unrelated to this site’s pricing, which covers the image tools only.
What this page does today
It builds the part of the job you control: the brief. Google's prompting structure splits a world into an environment prompt and a character prompt, and Genie 4 adds audio and mid-session steering on top. The builder below turns one photo or one sentence into those blocks, keeps them short, and checks whether your first frame will survive camera movement.
- · Input: a reference image, or a scene in one line, plus an optional subject, camera mode and audio layer.
- · Output: an environment block, a character block, a camera line and an audio line — short declarative sentences, in the order the model expects.
- · Usable now: paste the result into Project Genie, which already builds worlds from text or image prompts.
If you would rather start from the picture, image to prompt reads a frame back as text you can split into these blocks; if you want several candidate briefs from one idea, the image prompt generator writes the variations.
The Run in Genie 4 button stays disabled until we have API access. That is deliberate: a button that silently does nothing, or quietly calls a different model and labels the output “Genie 4”, is a lie we are not going to ship. When access opens, the button turns on and the changelog below records the date.
Build your first world brief
Start with an image or a one-line scene. You'll get an environment block, a character block, a camera line and an audio line.
JPG, PNG or WebP, up to 10 MB. Centre the subject; keep text out of frame. The file is read in your browser and never uploaded.
0/200
0/120
Genie 4 is a research preview; no public API yet. Your prompt still works in Project Genie today.
Your four blocks appear here
Nothing is generated and nothing is sent anywhere. The builder assembles your own words into the structure the model expects, and marks anything you have not described yet with a bracketed slot so you can see what is missing.
- · Environment — surfaces, style, landmarks, behaviour
- · Character — who you control, one action, its effect
- · Camera — one mode, no cuts
- · Audio — one ambience layer, plus what must not appear
Not affiliated with Google. Genie, Gemini and Project Genie are Google's products. We do not resell access.
The four blocks, and what goes in each
| Block | Answer this | Example |
|---|---|---|
| Environment | Surfaces, style, two or three landmarks, how the place behaves | "A rain-soaked night market, wet asphalt reflecting red signage, stalls under tarpaulins, steam rising from a griddle." |
| Character | Who you control, one action, its effect on the world | "A courier in a dark rain jacket walks forward, boots splashing through puddles that ripple outward." |
| Camera | One mode: static, slow pan, or follow | "Follow, at chest height, slow." |
| Audio and constraints | One ambience layer, one foreground sound, what must not appear | "Steady rain on canvas with a low market murmur. No music, no voice-over, no on-screen text." |
Long, layered sentences get re-interpreted while the world runs, which is why every block above is written as short statements rather than a paragraph of mood.
Worked example: one product photo, four blocks
Start with a plain studio shot — a coral running shoe on an off-white tabletop, soft side light, nothing else in frame. It is a good first frame and a bad world, because there is nothing to walk through. The builder does four things to it:
| Step | Weak version | What the builder returns |
|---|---|---|
| Environment | "A studio with a shoe." | "A concrete studio floor with a low off-white backdrop, soft daylight from the left, fine dust on the surface. The room is quiet: dust drifts in the light, nothing else moves." |
| Character | "The coral running shoe." | "A coral running shoe that hops forward on its heel, landing with a small puff of dust that settles after each step." |
| Camera | "Make it cinematic." | "Follow, at shoe height, slow." |
| Audio | "Add sound." | "Quiet room tone with a soft low thud on each landing. No music, no voice, no on-screen text." |
The change from left to right is not decorative. Each line on the right answers a question the model would otherwise answer for you — and every one of them is copy-pasteable into Project Genie today, where the same image can be uploaded as the starting world.
Check the first frame before you paste
A first frame decides layout, light and what has to stay put for the next two minutes. Google's guidance for uploading your own image is to use a high-quality image and keep the character centred. Three checks catch most problems:
- Centred with margin. A subject cropped at the edge has nowhere to move.
- One readable surface. A floor, table or path gives the model somewhere to simulate contact and shadow.
- Short edge at least 1024 px. Not an official Genie 4 requirement — Google has not published one — but small frames fall apart once the camera moves. Upscale before you paste.
When Genie 4 opens to you
The plan for this page, stated in advance so you can hold us to it:
- 1.Run in Genie 4 turns on once we have API access, using the same four blocks you built here.
- 2.The status table is replaced by a compatibility table (what the API accepts, what it returns, what it costs per session) with the source and the date.
- 3.Nothing is deleted. The prompt builder stays free and stays usable for Project Genie, Genie 3 and other world models.
If Google changes the access path first, the changelog records that instead. We publish the date either way.
What we will not do
- We will not label another model's output as Genie 4.
- We will not sell waitlist positions, invites or "Genie 4 access" keys.
- We will not publish a resolution, frame rate or price Google has not published.
- We will not clone a "Genie 4 online free" runner. Anyone offering one today is either reselling something they do not own or generating something that is not Genie 4.
FAQ
- Can I use Genie 4 online for free right now?
- No. Genie 4 is a research preview for trusted testers. Google has announced no free tier and no public endpoint, so any site claiming to run Genie 4 for free is not running Genie 4.
- Is there a Genie 4 app or download?
- No. Genie 4 runs inside Google products; there is no app, installer or local build to download.
- What can I do while I wait?
- Build two or three first frames and write the four blocks for each. Project Genie already generates worlds from text or image prompts, so the brief is testable today, and the same structure carries into Genie 4.
- Will this page run Genie 4 when it opens?
- That is the plan, but only with real API access. Until then the button stays disabled and the builder keeps working — we would rather ship a useful page with an honest gap than a fake demo.
- How is this different from Google's prompting guide?
- Google’s guide documents the environment/character structure (it is still titled for Genie 3). This page adds the access status you cannot get from a launch post, a first-frame check, and copy-ready blocks — plus a record of what changes when Genie 4 opens.
The structure this page builds on is explained with a worked example in the Genie 4 prompt guide; if your source photo needs more pixels first, upscale it.