what is genie 4, google's world model?
genie 4 is google deepmind's world model: it generates an interactive world from a text prompt, an image or both, with synchronized audio, and you can walk through it with the keyboard and change it with new prompts while it runs. google started rolling it out to eligible google ai ultra subscribers in project genie on november 18, 2026, and made it available to developers in the gemini api on november 19, 2026.
it is still a preview. google's model card says google is sharing genie 4 as a preview to learn alongside its community.
download the ReMynd desktop app to search text you saw in a focused window, by app and date.
what does genie 4 do?
genie 4 renders a world frame by frame as you move. google deepmind's genie 4 announcement, dated october 27, 2026, says you can use natural language to modify a simulation in real time, like changing the weather, moving the camera, or introducing a new person or object.
the model card lists what goes in and what comes out.
| field | what google's model card says |
|---|---|
| inputs | text strings, images, and real-time user actions such as keyboard and mouse navigation |
| outputs | real-time 720p video at up to 24 frames per second, with synchronized audio |
| built on | gemini omni flash, using gemini for reasoning and real-time steering |
| training data | video, audio, image, and text data |
google's project genie post says that for the first time, genie 4 pairs these visuals with real-time native audio, from ambient soundscapes to character dialogue.
who can use genie 4?
access depends on where you use it.
| where | who google says can use it |
|---|---|
| project genie | eligible google ai ultra subscribers; google ai pro members can join the waitlist |
| gemini api | developers with a paid tier api key, through the interactions api |
| google ai studio | developers, through its interactive canvas and starter apps |
the project genie site calls it an experimental research prototype rolling out to google ai ultra subscribers (18+). google's developer post says developers can start from a text prompt, a reference image, or both.
how much does genie 4 cost?
google's gemini api pricing page lists these paid tier rates per 1m tokens. the free tier is marked not available.
| item | paid tier price per 1m tokens |
|---|---|
| input (text or reference image) | $1.50 |
| output (text) | $9.00 |
| output (video, audio included) | $17.50 |
| context caching | not available |
google says billing is based on emitted video output tokens at 240 tokens per frame, which it puts at 5,760 tokens per second, or an effective rate of about $0.10 per second (about $6.05 per minute) at standard pricing. it adds that genie 4 does not support batch api, flex inference or priority inference, and that you should always call complete() to end a session and stop billing.
what are genie 4's limits?
google's genie 4 api guide lists these constraints.
- output is a fixed 16:9 frame; portrait video and configurable frame rates are not supported.
- visual stability may degrade past 60 to 90 seconds, with a hard cap of 120 seconds.
- keyboard moves take effect within about 0.2 to 0.5 seconds, while text prompts take about 1 to 2 seconds to appear.
- it does not accept live microphone or webcam input, and does not support function calling, google search grounding, structured outputs or session resumption.
- signs, gauges and clocks deform over time, and exact object counts drift across camera pans.
google deepmind's genie page words the length limit differently: it says the model currently supports around a minute of continuous interaction, rather than extended sessions. google's api guide also says genie 4 serves from data centers in the united states, western europe and japan, with round-trip latency under 150 ms expected in north america, western and central europe, and east asia.
how is genie 4 different from genie 3?
google keeps both in project genie. its project genie post says genie 4 delivers its highest-fidelity visuals, physics and native audio, while genie 3 stays in the model picker for longer runs or panorama-seeded spaces.
| genie 4 | genie 3 | |
|---|---|---|
| strength google names | highest-fidelity visuals, physics and native audio | longer runs and panorama-seeded spaces |
| session length google gives | see the limits above | up to 3 minutes per generation |
| seeding | a text prompt, an image or both | adds 360-degree panorama images |
the model card says genie 4 improved significantly compared to genie 3 on photorealism, fluid movement, and real-time steering, based on human and automated evaluations. google's developer post says every output includes synthid watermarking.
where does ReMynd fit?
ReMynd records your screen on a mac, runs ocr locally with apple vision, and keeps a searchable index on your computer. text comes from the focused window only, and video covers windows you have not excluded.
in our view, it helps when you tried a dozen world prompts in a browser and lost the one that worked: press control space, search in plain language, and jump to the exact moment, provided ReMynd was running then and the window was not excluded. on a mac it needs macos 14 or later on an apple silicon mac.
ReMynd also runs on windows 11 (x64 or arm64). for google's other model news, see what is gemini 4 neon.
limits of this guide
- it follows google's announcement posts, the genie 4 model card, the gemini api docs and pricing page, and the project genie site, as of november 2026. we did not test genie 4.
- the prices are gemini api list prices. google ai ultra and google ai pro plan prices were not checked.
- google's pages give two session lengths for genie 4, and this guide reports both as written.
download the ReMynd desktop app and search what you saw in a focused window, by app and date.