Gemini avatars are heading to Google’s desktop app, and they are arriving alongside two other quiet construction projects: a hidden Customize tab and the first client-side signs of Gemini 4. On 24 August 2026, TestingCatalog reported that Google is preparing the Gemini desktop app for avatars with a dedicated Settings section, a concealed discovery screen for apps, skills and plugins, and new Gmail and Calendar response widgets — none of which Google has announced.

The desktop app in question is the native macOS client Google launched globally on 15 April 2026, a free download for macOS 15 and later with an Option+Space shortcut, screen sharing and local file access. There is still no full Google Gemini desktop client for Windows — Windows users get a separate search-style launcher instead — so everything spotted here concerns the Mac app. Every feature in the report remains hidden and in development; the value of the discovery is what it signals about where the platform is going.

This article separates what is verified about Gemini avatars on desktop from what is speculation, explains how the officially live avatar feature works today, walks through the Customize tab and widget changes, and weighs the evidence that Gemini 4 — a model Google has publicly confirmed is in training — is the release this client work is preparing for. For businesses planning around Google’s platform, the pattern matters more than any single feature.

Gemini Avatars Come to the Desktop App

gemini avatars desktop customize tab gemini 4 b wardrobe two doors

The headline change in the TestingCatalog report is that Gemini avatars are being wired into the desktop app itself, rather than remaining a web and mobile capability that the Mac client merely displays.

A dedicated Settings section

According to the report, avatars on desktop get “a dedicated Settings section” where users can create and manage them directly from the app. That is a meaningful step up from the current arrangement, where creating Gemini avatars starts on a phone. Managing Gemini avatars natively would put the desktop app at feature parity with the web on one of the platform’s most personal features — and it suggests Google sees Gemini avatars as core to the client, not an accessory.

An Avatar option in the attachment menu

The second change is a new “Avatar” option in the attachment menu, tied to the previously spotted Likeness feature — a 3D portrait of yourself for use in image and video generation. In practice, that would let a Mac user drop their own likeness into a prompt the same way they attach a file today. TestingCatalog is explicit about the status: “For now, these desktop changes remain in development.” Nothing here is live, and Google has confirmed none of it.

Everything spotted in the desktop app, in one table

The report and TestingCatalog’s earlier desktop coverage from 1 August 2026 add up to a clear picture of a client being built out fast. Here is the full set of in-development desktop findings.

Feature spottedWhat it doesStatusFirst reported
Avatars Settings sectionCreate and manage Gemini avatars inside the desktop appIn development, hidden24 Aug 2026
Avatar attachment optionAttach your Likeness 3D portrait to promptsIn development, hidden24 Aug 2026
Customize tabDiscovery screen for apps, skills and pluginsHidden on desktop and web24 Aug 2026
Gmail and Calendar widgetsStructured response cards for mail and schedule queriesIn development24 Aug 2026
Image and video generation tabsDedicated creation surfaces in the Mac clientIn development1 Aug 2026
Custom MCP server panelConnect external tools to the desktop clientIn development; broad support still missing1 Aug 2026

We covered that earlier wave of Mac-client work — the media generation tabs and MCP panel — in our article on the Gemini desktop app’s MCP support and media generation tabs, and the new findings extend exactly the same trajectory.

How Gemini Avatars Work Today

gemini avatars desktop customize tab gemini 4 c control board switches

Desktop support may be new, but Gemini avatars themselves are already an official, documented feature of Gemini on the web and phones — which is precisely why the desktop work is credible.

Creating an avatar means a selfie video and your voice

Google’s help documentation describes the flow plainly: “You can record your face and voice to create your avatar to use in Gemini Apps.” Creation starts on a computer by scanning a QR code with your phone, then recording a short selfie video and voice sample. The result is a persistent likeness tied to your personal Google Account, usable in videos generated with Gemini Omni and images generated with Nano Banana — the same image model at the centre of our report on Gemini watermark removal.

Where Gemini avatars are allowed — and where they are not

The restrictions are as notable as the capability. Gemini avatars require users to be 18 or over, work only in English for now, and are not available in the EEA, Switzerland or the United Kingdom. UK businesses therefore cannot use the feature at all today, a point we return to below.

RequirementOfficial rule today
AccountPersonal Google Account only
Age18 or over
LanguageEnglish only
Regions excludedEEA, Switzerland, United Kingdom
Creation devicePhone camera via QR code, managed from a computer

From Characters to a persistent identity

The feature has a history. TestingCatalog first spotted 3D avatar work in March 2026 under the Likeness banner, and reported in May 2026 that avatars — “previously branded as Characters” — would act as a persistent identity tied to the account, launching alongside the Gemini Omni agent. The desktop Settings section is the latest chapter of a programme Google has been building all year.

Gemini Avatars and the Hidden Customize Tab

gemini avatars desktop customize tab gemini 4 d rocket standing

The second discovery in the report is a Customize area that already exists, invisibly, on the web — and is now being replicated on desktop.

A discovery screen for apps, skills and plugins

TestingCatalog describes the desktop app gaining “the previously spotted ‘Customize’ area”, which “opens a discovery screen for apps, skills, and plugins.” It remains hidden on both surfaces. The framing is significant: a browsable catalogue of extensions is the kind of surface a platform builds when it expects third parties to ship things into it.

Skills management started on the web

This is not the first sighting of the underlying system. In July 2026 we reported that the Gemini web version was gaining a Skills management section, and the Customize tab looks like the discovery layer over that same machinery. What is still absent, per the report, is broad connector and MCP support in the desktop client — the plumbing that would let those plugins reach external systems.

Why a Customize tab matters more than it sounds

A customization surface changes who Gemini is for. Out of the box, an assistant is a generalist; with skills, plugins and Gemini avatars layered on top, each installation becomes specific to its user. That is the same direction every major assistant vendor is moving, and it rewards organisations that already know which workflows they want automated.

Gmail and Calendar Widgets Point to a Bigger Client

gemini avatars desktop customize tab gemini 4 e door in frame

The third finding is easy to underrate: the desktop app “is already being adapted for new types of responses, including Gmail and Calendar widgets.”

Structured responses instead of prose

A widget is a structured answer — a mail card or schedule block rather than a paragraph of text. TestingCatalog reads the work as Google “preparing the client to handle upcoming Gemini 4 capabilities”, which is an interpretation rather than a fact, but a reasonable one: richer model output needs richer rendering, and you build the rendering first.

The desktop app is closing gaps fast

Put the widget work next to the August findings — generation tabs, camera attachment, MCP panel — and the Mac client has gone from a thin wrapper to a serious surface in under five months. Google said at launch that Mac users prompt roughly twice as much as users on other platforms, which explains the investment.

Signs of Gemini 4 Inside and Outside Google

gemini avatars desktop customize tab gemini 4 f blank flag on pole

The report’s most-quoted line is about the next model. Here the evidence splits cleanly into what Google has confirmed and what remains rumour — and the confirmed part is substantial.

What Google has said on the record

Two official statements anchor the story. In July 2026, alongside the Gemini 3.6 Flash launch, Google DeepMind said: “We have already started our most ambitious pre-training run yet, for Gemini 4, and can’t wait to share more.” Days later, on Alphabet’s Q2 2026 earnings call, Sundar Pichai repeated it: “We are now training Gemini 4, and we’re being very ambitious with it,” adding that the next generation of frontier systems will need “much larger base models.” Gemini 4 is not a leak; it is a stated programme.

What remains speculation

The December arrival window is an inference from Google’s roughly six-month training cycles, not a date anyone at Google has given. The claim that the summer cycle’s Gemini 3.5 Pro was cancelled — clearing the runway for Gemini 4 — is TestingCatalog’s reporting only; what 9to5Google has established is that 3.5 Pro missed its June 2026 launch window. And the viral claim of “120+ new mentions of Gemini 4, up from 0” in internal references is an unattributed social media post quoted in the report, not measured data. None of these should be treated as fact.

Why the client work is itself a signal

The strongest evidence is circumstantial but concrete: Google is rebuilding the front end. Avatars in Settings, a Customize tab, structured widgets — these are the surfaces a materially more capable model would need. When the client sprints ahead of the model, a model release is usually close behind.

The Gemini 3 Family the Desktop App Runs Today

To judge how big a jump Gemini 4 would be, it helps to see the cadence that got Google here — and the scale of the app these releases feed.

ModelReleasedNotable detail
Gemini 3 Pro + Deep Think18 Nov 20251M-token context; flagship tier
Gemini 3 Flash17 Dec 2025Fast tier for the Gemini app
Gemini 3.5 Flash19 May 2026App rollout via release notes
Gemini 3.6 Flash + 3.5 Flash-Lite21 Jul 2026$1.50/$7.50 and $0.30/$2.50 per 1M tokens
Gemini 3.7 Flash13 Aug 2026Intro pricing $0.75/$3.75 per 1M tokens
Gemini 4In training, no dateConfirmed by DeepMind and Pichai, Jul 2026

Benchmarks the flagship already posts

Gemini 3 Pro’s launch numbers show how high the bar for Gemini 4 sits. The scores below are Google’s published figures from the Gemini 3 announcement.

Gemini 3 Pro published benchmark scores (Nov 2025)
GPQA Diamond 91.9%
SWE-bench Verified 76.2%
SimpleQA Verified 72.1%
ARC-AGI-2, Deep Think 45.1%
Humanity’s Last Exam 37.5%

For a deeper look at the current flagship tier, see our guide to Gemini 3.1 Pro.

An audience of a billion

The platform these releases serve crossed 1 billion monthly active users on 11 August 2026, making Gemini Google’s fastest-growing product ever and its 14th product to pass that mark. Growth has been steep: Google’s stated figures put the app at 400 million monthly users in May 2025 and 950 million in July 2026 before the billion. The same announcement noted that 63% of users talk to Gemini by voice and that more than 150 million images are generated daily.

Gemini app monthly active users (Google-stated figures)
May 2025 400M
Jul 2026 950M
Aug 2026 1,000M

What Gemini Avatars and Gemini 4 Mean for Business

A leak about hidden desktop features may seem far from a business decision. It is closer than it looks, for three reasons.

The personalisation wave is coming to work

Gemini avatars, skills and a Customize tab all push in one direction: assistants shaped to the individual. The machine learning behind a personal likeness — a model fine-tuned on your face and voice as training data — raises immediate questions for brand control, consent and impersonation risk. Firms that decide their policy on staff likenesses before the feature reaches them will be glad they did; the neural network does not wait for your governance to catch up.

UK availability is a planning fact, not a footnote

Because Gemini avatars are currently unavailable in the UK and EEA, British teams get something rare: lead time. The sensible use of it is unglamorous — reviewing where a desktop AI client fits your estate, which connectors you would allow, and what an avatar policy should say. Our AI strategy service exists for exactly this kind of before-the-wave planning, and our AI models and tools hub tracks the model releases as they land.

Watch the client, not just the model

The practical lesson of this report is a monitoring habit: platform intentions show up in clients months before announcements. The desktop app gained media tabs, an MCP panel, widget scaffolding and now Gemini avatars groundwork in one summer. Whether Gemini 4 lands in December or later, the shape of what Google is building is already visible — and it is a personalised, extensible, desktop-native assistant.

Getting Ready for Gemini Avatars: A Practical Checklist

You cannot switch Gemini avatars on in the UK today, and the desktop features are unreleased everywhere. What you can do is make the decisions that will be waited on later. Three areas cover most of the ground.

Decide your likeness policy first

A likeness policy answers the questions Gemini avatars will force: may staff create an avatar from a work account, may a personal avatar appear in work output, and who signs off marketing material featuring a synthetic likeness of a real employee? Write the answers down now. The organisations that struggled with generative image tools were rarely short of technology — they were short of a one-page policy when the first awkward example landed on a manager’s desk.

Map where the desktop client would sit

The Gemini desktop app can see screens, read local files and — once the MCP panel ships — reach external tools. That is an endpoint-security question as much as a productivity one. List which teams would benefit, which data those Macs can reach, and whether your device management would treat the client as sanctioned software. Doing this before general availability turns a scramble into a rollout.

Assign someone to watch the release cadence

Google shipped five model releases in nine months and rebuilt its desktop client in one summer. Somebody in your organisation should own the job of tracking what actually ships — Gemini avatars reaching desktop, the Customize tab going live, any Gemini 4 announcement — and translating it into a monthly one-paragraph update. The habit costs an hour a month and removes the surprise from every launch this article anticipates.

Gemini Avatars FAQ

Are Gemini avatars available on the desktop app now?

No. The desktop Settings section and Avatar attachment option are hidden, in-development features spotted by TestingCatalog on 24 August 2026. Google has not announced them, and there is no release date.

What are Gemini avatars, exactly?

An official Gemini feature that turns a short selfie video and voice recording into a persistent 3D likeness tied to your personal Google Account, usable in generated images and videos. On current documentation it requires users to be 18+, supports English only, and is not offered in the EEA, Switzerland or the UK.

Is Gemini 4 confirmed?

Training is confirmed; nothing else is. Google DeepMind and Sundar Pichai both said in July 2026 that Gemini 4 is in training. The December launch window is analyst and press inference from Google’s release cadence, and the reported cancellation of Gemini 3.5 Pro is unconfirmed.

Does the Gemini desktop app work on Windows?

No. The native Gemini desktop app is macOS-only (macOS 15+, launched 15 April 2026). Windows has a separate Google search launcher app, which is not the full Gemini client.

What is the Customize tab?

A hidden discovery screen — spotted on the web and now in the desktop app — for browsing apps, skills and plugins that extend Gemini. It is not live on either surface yet.

References