The polished talking-head clip at the start of my latest video isn't me. I never opened the camera. I made it on my phone in about two minutes using Gemini Avatars. No studio, no green screen, just the Gemini app.
This post is the written companion to the full walkthrough. If you'd rather watch me set it up and run the live demo, the video is right above. If you'd rather read and follow along, everything's below.
What are Gemini Avatars?
Gemini Avatars are part of Gemini Omni, Google's new AI model announced at I/O 2026. The feature trains a digital version of you, your face and your voice, right inside the app. Setup takes roughly two to three minutes. After that, you type a prompt and get a short clip of yourself in almost any setting, saying almost anything.
The key difference from other AI video tools: you're not picking a stock character from a library. You are the character. The avatar is you.
What you need before you start
This feature has real requirements, and a few of them will rule people out. Check these first:
- A paid Google AI plan. AI Plus, Pro, or Ultra all work. The free tier does not include avatars right now.
- You must be at least 18.
- English only at the moment.
- Not available in the EU, Switzerland, or the UK due to regulations.
- A modern phone with a selfie camera. Any recent Android or iPhone works. I used a Samsung Galaxy S23.
How to set up your avatar (step by step)
There are two ways into the setup screen:
- The quick way: tap the + button and choose the Avatar option, then "create your avatar."
- The other way: tap your profile photo, go to Settings, scroll down, and tap Avatar.
From there, the flow is short:
- Agree to the terms. Two things worth reading: you can delete your avatar data at any time, and you are responsible for whatever you create with it.
- Record a short selfie video. Look directly into the camera lens, not at the screen. It asks you to read out a few random numbers, then turn your head left and right so it captures your face from different angles.
- Wait for processing. The clip uploads to Google's servers and your avatar is built in a couple of minutes.
One detail I like: the avatar is private to you. It's tied to your account, and only you can call it. It isn't a public avatar that anyone can grab.
How to generate a video with your avatar
Once it's ready, you call your avatar with @me or @yourusername. In a new chat, tap +, pick your avatar, and write a prompt describing the scene and what you want to say.
A few things I learned on the first try:
- The prompt matters, and not every prompt works. My first attempt, reading a news headline in a TV studio, came back with "cannot generate that video, try describing another idea."
- Reframing usually fixes it. My second prompt ("place me on a futuristic Tokyo street at night") generated fine.
- Generation takes 1–2 minutes. It runs on Google's video model (the Veo family), so give it a moment.
- You can regenerate the same prompt. Each run produces a fresh clip, so if you don't like the first result, just run it again.
The safety side that actually matters
This is the part I'd want anyone trying it to understand, because it's genuinely well thought out:
- It needs a live face, not a photo. The recording step means someone can't build an avatar of you from a photograph they found. That's a deliberate guard against deepfakes.
- Every clip carries a SynthID watermark. Google embeds an invisible SynthID marker into every output, and it survives cropping and compression. You can check a clip's origin in Chrome or through Google Search. In other words, Google can always tell it was AI-generated, and so can you.
- You stay in control of the data. Delete the avatar whenever you want and recreate it if you're not happy with the result.
What it's actually good for
Once your avatar exists, you can spin up intros, shorts, and social clips in bulk without ever setting up the camera again. For anyone making regular content, that's the real value. You can generate yourself on camera while you're nowhere near a camera.
Is it perfect? Not yet. The output still has rough edges. But this is version one, and for a v1 it's genuinely impressive. It's the kind of feature that gets noticeably better with each update, so it's worth learning now.
The bottom line
Gemini Avatars turn your phone into a personal video studio: record yourself once, then generate short clips of yourself on demand, each one watermarked so its AI origin stays traceable. If you're on a paid Google AI plan and outside the restricted regions, it's worth two minutes to set up and play with.
Watch the full hands-on demo above, and for more AI tool walkthroughs, subscribe to the Neel's World YouTube channel. If you make an avatar, tell me in the comments how you're using it.