Cut.ProCut.Pro

Genie 4 comes to Project Genie: playable AI worlds, now with sound

Google released Genie 4 to eligible Google AI Ultra subscribers inside a new Project Genie, with real-time generated audio, an agent that builds the world through conversation and a community gallery. Here is what you can do with it, what it costs in Brazil and what content it can produce.

Genie 4 comes to Project Genie: playable AI worlds, now with sound

Genie 4 comes to Project Genie: playable AI worlds, now with sound

On Wednesday (November 18), Google released Genie 4 to eligible Google AI Ultra subscribers, inside a new version of Project Genie. The model generates a world you move through with the keyboard, now with sound generated alongside it, and it accepts prompts in the middle of a session to change what is happening.

Instead of handing back a closed clip, it builds an environment that reacts to you, and that changes the kind of content it can produce. It is also expensive, and you still can't try it outside Google's top plan, so it is worth separating what is useful today from what is a showcase.

What is Genie 4?

It is a world model: instead of delivering a finished video a few seconds long, it generates the next frame in real time as you walk, turn the camera or interact. When DeepMind introduced Genie 4 on October 27, still as a preview for testers, the headline feature was native audio: music, ambient sound and character speech that adapt to what you do. According to DeepMind, it is the company's first real-time world model to generate native sound.

In the middle of a session you can type a new request: make it rain, put a person in the scene, change the direction of the action. Characters move, react to the environment and can be guided on the spot.

What came with the new Project Genie

Three pieces, according to today's announcement:

  • Project Genie agent. You describe the world in a conversation and it builds the scene, with a background soundtrack. Gemini can follow the screen, suggest next steps and trigger events inside the world.
  • Community gallery. Curated worlds and worlds from other users to explore, remix and publish your own. Google cites examples made with creators such as Will (BLVCKL!GHT) and Anand Gandhi.
  • Genie 3 stays in the model picker, with sessions of up to 3 minutes (they used to be 60 seconds), worlds chained into longer journeys, creation from a 360-degree panoramic photo and text-based changes to the environment while you explore.

Every generated video and audio track comes out with the invisible SynthID watermark.

Who can use it, and whether it is worth it in Brazil

For now, only people who pay for Google AI Ultra, and not every subscriber: Google says "eligible" ones. If you have Google AI Pro, you can join the waitlist at labs.google/projectgenie. Today's announcement says nothing about an API; according to the October post, access through the Gemini API and Google AI Studio is going to early access partners.

As for Brazil, here is the picture:

  • The Google One plans page, in Portuguese, already advertises Project Genie in Ultra and says it is available in more than 140 countries.
  • The Google One help center limits the feature to people 18 and older, but does not include a list of Project Genie countries.
  • The Genie 4 announcement does not list countries. There is still no Genie 4-specific confirmation for Brazil.

And the price matters. Google contradicts itself about which Ultra version includes the feature: a note on the plans page mentions only "Ultra 20x", and the FAQ on the same page says just "Ultra". In May, Tecnoblog found that, in Brazil, only the R$ 999.90 per month version gave access to Project Genie (the R$ 779.90 one did not). Check Google One before subscribing.

What you can do with it as a creator

The most direct use is the exploration video: record the session and post it. "I walked into a world the AI created on the spot" is a format that already works in gameplay and reaction content, and native sound takes away the silent-video look that Genie 3 worlds had. In the first version of Project Genie, from January, you could download the exploration video; the Genie 4 announcement does not detail exporting in the new experience, so check before planning a series around it.

Two other uses I see:

  1. Backdrop for clips and narration. If you tell stories, make true crime or explain places, you can generate the environment and move through it with the camera, instead of piecing together stock footage. The difference from a regular video generator is control: you decide the path in real time.
  2. A gallery world as a topic. Remixing a popular community world and showing your own version is the same trend mechanism as on Reels and TikTok, just inside Project Genie.

The catch is in Google's own list of limits: camera latency, complicated physical interaction, real-time audio accuracy and scenery that changes after a while. The October technical post adds illegible text and details that get lost after several edit requests. In practice, long sessions and objects that need to stay the same are exactly what still fails. Google calls both models "frontier research models", and it shows on screen.

One more caution: a realistic scene that never existed has to be labeled as AI-generated content on YouTube, and SynthID is there for anyone who wants to check. Don't post a Genie world exploration as if it were footage of a real place.

Is it worth subscribing to Ultra for Genie 4?

I wouldn't subscribe just for this. It is close to R$ 1,000 a month for a feature that Google itself calls research, with no confirmed country list and no open API. If you need generated scenes for short-form video, paying per generated second is much cheaper. In the Cut.Pro Generator you pick from several video models, pay in credits for what you generated and send the clip straight to the editor or to publishing. For an idea of per-second pricing on a Google model, see the test of Gemini Omni 1.2 Flash. Genie 4 isn't there yet, so I couldn't run a real test for this post.

The exception is anyone who already pays for Ultra for other reasons (like the 20 or 30 TB of storage). In that case, test it today and see whether your account is among the eligible ones.

The bigger move is what I called the workflow became the product: Google is stacking an agent, a gallery and remixing on top of the model, as it did with Flow and the artist tools. The winners are the people who can get the result onto social fast, not the ones who get access first.

What to do starting today

If you have Google AI Pro, join the waitlist, which is free. If you have Ultra, open Project Genie, record a short session and see whether it holds 30 seconds of attention without the scenery melting. If you have neither, wait for the API or for it to reach Pro: until then, generating scenes by the second is still the cheapest path, and content recorded by real people still carries more weight, as we showed in provenance as a signal.

Share

Keep reading

More insights and tutorials to help you grow as a content creator.