Google upgrades Gemini with custom Gems and Imagen 3

If you've been using OpenAI's Custom GPTs, Google's pace might have felt a bit sluggish. First teased back at I/O in May, Google is finally turning its custom AI concept into a real feature.

Google began rolling out "Gems" to Gemini Advanced, Business, and Enterprise subscribers, alongside a system-wide upgrade to its new Imagen 3 image generation engine.

Build your own Gemini dream team

Put simply, Gems are custom personas tailored for specific tasks. Setting one up requires zero coding: give it a name, type in a set of instructions defining how it should behave, and start chatting.

Once created, a Gem remembers those guardrails across conversations. That means you no longer have to paste the same tedious prompt context every time you want to review code or outline a blog post.

To jumpstart adoption, Google bundled five premade Gems: a Learning Coach for breaking down complex subjects, a Brainstormer for generating creative ideas, a Career Guide for resume and skill planning, a Writing Editor for tightening prose, and a Coding Partner for debugging and building projects.

Gems are rolling out globally across web and mobile in over 150 countries, covering most major languages, strictly for paying subscribers.

Imagen 3 brings sharper visuals and cautious person generation

The update also brings Imagen 3 into the core Gemini app, replacing older visual systems for both free and paid users over the coming days.

Google claims Imagen 3 sets a new benchmark for detail and prompt accuracy. It handles varied visual styles—from photorealistic close-ups with glistening water droplets to rich oil paintings and claymation textures—much more reliably. All outputs continue to carry Google's invisible SynthID watermark.

The elephant in the room, of course, is human generation. After earlier controversies forced Google to pull the feature offline, Imagen 3 is cautiously reintroducing people pictures, starting as an English-only early access preview for Advanced and Enterprise accounts.

The safety guardrails this time are strict: Gemini will not generate photorealistic, identifiable real people, depictions of minors, or excessively violent or sexual scenes.

Why it matters

Gems doesn't reinvent the wheel, but it patches a glaring hole in Gemini's daily utility. For paying power users, eliminating prompt repetition in structured workflows like coding or editing makes the $20 monthly subscription far easier to justify. If you already use Gemini regularly, turning your repetitive chores into standalone Gems is worth doing right away.

Competitively, this keeps Google in lockstep with OpenAI. While OpenAI has tried—with mixed success—to turn GPTs into an open app store, Google is treating Gems purely as a workflow utility, emphasizing corporate readiness, language breadth, and Workspace integration.

The cautious return of human image generation shows Google is still walking on eggshells. We think keeping tight restrictions on identifiable people is a necessary compromise to avoid another public relations disaster, even if aggressive filtering occasionally frustrates users pushing for sheer creative freedom.