Every model on HeyOz makes a clip you watch. Genie 4 makes a world you move through. Describe a place, step inside it, change the weather while it runs — and it keeps going, with sound. We are getting it into HeyOz as soon as we can.
Scenes from Google DeepMind's Genie 4. Not clips rendered once and finished, but environments that keep running while you move the camera and change what is in them.
Google DeepMind calls it their most realistic and interactive world model yet. The difference from a video model is that it does not stop when the clip ends.
Describe landscapes and events, move the camera, add a character or change the weather — live, in natural language, while the simulation is still playing. Not a re-render. The world reacts and carries on.
Environments where things behave: water moves like water, trees sway, and characters act naturally rather than drifting through a scene. Google describes plausible cause-and-effect running through the world.
Sound is generated with the world, not added afterwards. Leaves crunch underfoot, crowds cheer, and characters can speak — with you, and with each other.
Genie 4 builds on the Gemini family. Gemini Omni anchors each simulation in real-world appearance and behaviour, while other Gemini models handle the reasoning that keeps a world coherent as it changes.
Genie 4 is a research model today, not an ad tool. Here is the part we are actually excited about.
Genie 4 is Google DeepMind's world model. Instead of returning a finished clip, it generates an interactive environment you can move through and change in real time using natural language, with audio generated alongside the picture.
Not yet. Google has it with trusted testers, with wider access planned for Google AI Ultra subscribers through Project Genie and for early-access partners of the Gemini API and Google AI Studio. We will add it to HeyOz as soon as access allows.
A video model gives you a clip: once it is generated, it is fixed. A world model keeps running. You can move the camera, add a character or change the weather while it plays, and the environment responds instead of starting over.
Yes. Google describes audio generated natively with the world rather than added afterwards, including ambient sound and characters who can speak with you and with each other.
Google lists sessions lasting only a few minutes, some visual drift over longer runs, difficulty with multiple agents at once, imperfect recreations of real places, and text that is only reliably legible when it was in the input. It is a research preview, not a production tool.
Yes. Google states that video and audio generated by Genie 4 carries SynthID watermarking, the same provenance marking used across its generative models.
For cinematic motion, Kling 4 runs thirty seconds with stereo audio and Runway Gen-4.5 leads on physics. For presenters, HeyGen puts a person on camera from one photo. All of them are in the studio right now.
Pick the right engine for the job — all in one place.
When Genie 4 opens up, it will land in a studio that already knows your brand.
HeyOz reads your brand and keeps every asset on-color, on-voice, on-message.
Seedance, Veo, Kling, GPT Image, and more — no juggling subscriptions.
Sized and formatted for Meta and TikTok, straight out of the box.
Not just a model — templates, avatars, and an agent that does it all.
Every other leading model is already in the studio. Start with those, and Genie 4 will land next to them.