Genie 4 is a world model from Google DeepMind. The research post went up on 27 October 2026. It describes a simulation you can steer while it plays, with sound generated at the same time as the picture. It is not a face-swap tool. HiVideoStudio does not host it, and this guide does not sell a render. The short page is Genie 4.
HiVideoStudio is not affiliated with Google or DeepMind.
What the preview includes
The DeepMind post, by Raia Hadsell, Diego Rivas, and Hang Qi, says Genie 4 generates music, ambient noise, and dialogue together with the world. A person in the preview uses ordinary keyboard controls to move and to aim the camera. A text prompt can change the weather, move the camera, or bring in a new person or object. The post says Gemini fits that prompt to the scene that is already running.
Gemini Omni is named as the piece that anchors how things look and behave. Gemini 3.8 Flash is named as an agent in the loop: it watches frames, breaks a goal into steps, and checks the steps. A second experiment in the same post pairs Gemini 4 Argon with Genie 4 on a robotics task. The picture they describe is a humanoid playing checkers, with the steps written in as the simulation runs. That is their research example. It is not a template here.
Access, as the post states it, started with trusted testers. A later step starts with Google AI Ultra subscribers in Project Genie, and with early-access partners of the Gemini API and Google AI Studio. TechCrunch's report the same day, by Rebecca Bellan, says that wider step was described as the coming weeks. This site has no login for any of those programs.
TechCrunch also reports that a scene can stay consistent for up to two minutes, and that this is twice the limit they give for Genie 3. The DeepMind post's limitation section does not state that number. It says detail and object permanence can drift as a run gets longer, after several edits, or while you move through the scene. It lists audio accuracy, legible text, spatial accuracy, and long-term consistency as open research. The two-minute line is TechCrunch's report of a briefing. It is not a measurement made for this guide.
The post says Genie 4 video and audio carry SynthID, DeepMind's watermark. Tester notes in the post include filmmakers trying camera and dialogue changes inside a live scene, and photojournalists trying a single archival photograph as the start of a simulation of that place. A photograph used that way is the first frame of an environment. It is not a face pasted onto another person's body. The post never describes a head swap or a character swap.
What it is not
Genie 4 is not Henge. Henge takes a guide video you already shot and replaces or retargets a person in it. Genie 4, in the research post, builds the world and then lets the tester steer it.
Genie 4 is not Echo-3. The Echo-3 posts checked for that page show a phone video of a real room becoming a 3D Gaussian splat, rendered in one demo by PlayCanvas. That is a reconstruction. Genie 4 is a simulation. The comparison guide is Echo-3 and Genie 4.
Genie 4 is not a HiVideoStudio effect. A sold clip here is a photo in a fixed template: Survival Horror AI at 50 credits, AI Zombie at 45 credits, Ozil Goal AI at 50 credits, Thriller Dance at 50 credits and silent. Each of those is about 15 seconds. None of them is a world you walk through, and none of them generates the kind of live prompt steering the Genie 4 post describes.
How to place Genie 4 next to a video tool
- If the clip is a world that changes because someone typed a new prompt or pressed a key, the October 2026 research name for that is Genie 4. You cannot run it on this site.
- If the clip is a real room you can orbit as a splat, read the Echo-3 page. That is a capture, not Genie 4.
- If the clip is one face dropped onto a performance you already filmed, read the character-swap guide. Genie 4's post is not that product.
- If you have one photo and you want a 15-second template, open the effect that matches the scene and read its credit price.
Quick answers
What shipped on 27 October 2026? A research preview of Genie 4 for trusted testers: photoreal steering, keyboard and prompt control, and native audio, as DeepMind's post describes it.
Is Genie 4 a face-swap tool? No. The post does not describe swapping a head or a character onto existing footage.
How long does a scene hold? TechCrunch reported up to two minutes. DeepMind's post says details can still drift on longer runs. This site did not time a session.
Can I buy a Genie 4 render here? No. The model page is an introduction. It is not for sale.
Sources
Google DeepMind, 27 October 2026. TechCrunch, Rebecca Bellan, 27 October 2026. Google's announcement, 27 October 2026. Last checked 2026-10-29.