Google opens its Genie 4 world model to developers in the Gemini API
Developers can now generate playable 720p worlds with sound from a prompt or image, at about 10 cents per second of output, with sessions capped at two minutes.
Key takeaways
- On November 19, Google opened Genie 4 to developers in Google AI Studio and the Gemini API.
- It turns a text prompt, an image or both into 720p, 24fps interactive worlds with native audio.
- Google's documentation puts pricing at roughly $0.10 per second of streamed output, with each session limited to 120 seconds.
- Prompts depicting real people, public figures or brand IP may be refused, PPC Land reports.
- Google pitches it for prototyping, spatial design and training robots and AI agents.
Google has opened Genie 4, its model that builds interactive worlds in real time, to outside developers, the company said on November 19. Developers can reach it through the Gemini API, using the Interactions API, and in Google AI Studio. The release followed a consumer rollout the day before, when Google began offering Genie 4 to eligible Google AI Ultra subscribers in Project Genie, its world builder, PPC Land and TestingCatalog report.
Developers give Genie 4 a text prompt, a reference image or a mix of the two, and it produces a continuous 720p world at 24fps with native audio, ranging from background sound to character dialogue, according to Google. People move around with keyboard controls and can type new instructions mid-session to alter things such as weather, lighting or objects, and Google says Gemini reasoning keeps the scene coherent. Each output carries an invisible SynthID watermark, the company says.
How much does it cost and what are the limits?
Using Genie 4 requires a paid-tier API key, and Google's documentation puts the price at roughly $0.10 per second of streamed output with stereo audio included, PPC Land and TestingCatalog report. TestingCatalog lists per-token rates of $1.50 per million input tokens and $17.50 per million video output tokens, while PPC Land reported that Google's pricing page did not yet carry a Genie 4 entry when it checked. A single session is limited to 120 seconds. According to PPC Land, the documentation also cautions that image stability can slip after 60 to 90 seconds, and the first frame takes roughly 5 to 9 seconds to appear.
Google serves the model from facilities in Japan, Western Europe and the United States, TestingCatalog reports, and apps connect over WebRTC in a browser or WebSockets from server code. Prompts showing public figures, real people or copyrighted brand IP may be refused or cut off partway, PPC Land reports, and output comes only in landscape 720p.
What is it for?
Google names three uses: prototyping interactive experiences, visualizing spatial designs, and training robots and AI agents. TestingCatalog says Google presents Genie 4 as the first of its world models that developers can build with.
Google DeepMind first showed Genie 4 on October 27 as a research preview limited to trusted testers, without giving a price, resolution or session length, according to PPC Land. TestingCatalog reports that Genie 4 draws on the Gemini family, with Gemini models doing the reasoning that keeps simulated worlds consistent.
PPC Land notes that 720p at 24fps is what Google gave for Genie 3 in August 2025, and that Google describes Genie 4's gains in realism, physics, steering and sound rather than resolution.
Questions readers are asking
What is Genie 4?
A Google DeepMind world model that creates explorable environments with sound in real time, which users steer with keyboard controls and typed instructions.
How much does Genie 4 cost for developers?
About $0.10 per second of streamed output with a paid-tier Gemini API key, according to Google's documentation as reported by PPC Land and TestingCatalog.
How long can a Genie 4 session last?
Up to 120 seconds. PPC Land reports the documentation says image stability can slip after 60 to 90 seconds.
Sources
- #genie-4
- #world-models
- #gemini-api
- #google-deepmind
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
AI2 min
OpenAI details its defenses against adversarial distillation
OpenAI says new safeguards cut attacks that pull hidden reasoning from GPT-6.2 Astra to under 1% success in its own tests, and that it acts against thousands of accounts a week.
AI2 min
Thinking Machines releases Inkling 2 Mini, a looped open-weights model
The 26.4B-parameter model runs its middle layers three times per token and needs about 20GB of memory, letting it run on laptops, the company says.
AI2 min
AI-made proof of Generalized Riemann Hypothesis compiles, awaits review
Anthropic's Levent Alpöge says its models produced the proof. Independent checkers say the code compiles, but Scott Aaronson expects understanding it to take years.
AI2 min
Mistral opens Asia-Pacific headquarters in Singapore, plans 200 staff
The French AI company will build custom models for public services, robotics and manufacturing, betting on regional demand for open-weight models it pitches as 'sovereign AI.'