Gemini 3.1 Pro is a surprisingly large upgrade from Gemini 3. The decimal point hides what feels more like a generational leap designed to tackle the messy, non-linear logic of the real world. In practice, Gemini 3.1 Pro offers fewer generic answers and more specific help. But to really showcase what it can do takes some creativity. To demonstrate, I came up with some prompts designed to push different aspects of the model, each leaning into a specific capability.
(Image credit: Shutterstock)
Gemini 3.1 includes a specialized reasoning layer that allows it to simulate adversarial scenarios. Instead of producing the first sensible answer, it slows down and stress tests ideas from multiple angles. By asking it to find the holes in a plan, you are effectively hiring a world-class strategist to sabotage your own ideas so you can build them back stronger. I decided to weaponize that to make a foolproof date night. I told Gemini:
“I’m planning a surprise date night involving a late romantic dinner and rooftop stargazing. Find three ways this plan could realistically go wrong and then revise the itinerary to include Plan B backups for each.”
You may like
The response noted that a late dinner before stargazing might create a massive “low-energy valley,” potentially ruining the romantic atmosphere. It suggested that a sudden cloud cover is not the only weather risk, but that dew point spikes can make outdoor furniture damp and uncomfortable for evening wear. The itinerary came back rewritten as a layered plan with decision points.
The same adversarial approach can be applied to business proposals, travel itineraries, or even big career moves where hidden weak points tend to surface at the worst possible moment. Letting the model simulate failure scenarios in advance turns vague anxiety into concrete contingency planning.
2. Cinematic geography
(Image credit: HBO Max/Studio Ghibli)
Gemini 3.1 can do some amazing video analytics, picking out patterns and details of people and locations, and connecting them to the real world. I wanted to see how well it could extract the aesthetic DNA of a video and pull out the intangible qualities that give it a certain feeling. I tested that with the prompt:
“Analyze this video of Studio Ghibli landscapes. Find locations in New York State that match the vibe of the scenes.”
Gemini broke down the visual motifs in the compilation and mapped those traits onto real places. It translated animation into geography.
The AI suggested places like the Ashokan Reservoir in the Catskills for its broad reflective water and mountain backdrop, and the village of Cold Spring for its river views and tree-lined streets. Each location came with a short paragraph explaining how the quality of light and terrain echoed specific animated scenes. This kind of aesthetic translation can help with everything from scouting filming locations to designing a themed party that captures a specific cinematic mood.
3. Lego intervention
(Image credit: Lego)
Gemini 3.1 Pro can analyze the space in images just as well as it can analyze the spirit of a video. Its spatial reasoning extends to seeing how things fit together, even if you can’t. I asked it to:
You may like
“Use this photo of a half-built Lego car and a pile of leftover pieces and show me exactly where these remaining technical pins should go to stabilize the chassis and allow the wheels to turn smoothly.”
The model identified the front axle as slightly misaligned relative to the rear assembly and noted that two blue technical pins in the leftover pile appeared to match open connector holes near the mid-frame. It then walked through a fix.
The attention to balance was notable, and the wheels spun without scraping after I followed the instructions. The spatial reasoning on display here can extend to far more than toys. Anything from home repairs, furniture assembly, or diagnosing why a shelf keeps leaning could be assisted by the AI. When a model can reason about forces, alignment, and structure from a single image, it becomes a practical troubleshooting partner rather than a passive observer
4. Instant animation
Native SVG rendering allows Gemini 3.1 tp serve as both front-end developer and graphic designer. Instead of just giving you code to copy and paste, it can now generate and preview interactive animations directly in the interface. For a test, I came up with a fun little gimmick, asking the AI to:
“Write the code for an interactive animated SVG for a kid named Orion. Create a night sky where the stars twinkle and the Orion constellation is highlighted, and the name is spelled out. When I click his name, have a rocket ship blast across the screen from bottom to top.”
The response included a complete SVG block with embedded CSS animations and a small JavaScript function to handle the click event. You can see a video of how it worked above.
It’s simple but polished. Gemini 3.1 Pro handled the animation timing cleanly and structured the code, so it was easy to tweak colors and speeds. It felt like a personalized digital card rather than a code dump.
Native code rendering opens the door to creating interactive invitations, personalized learning tools, or lightweight web experiments without hiring a developer. Instead of static designs, you can generate living graphics that respond to clicks and feel tailored to a specific person or moment.
5. Literary ambition
(Image credit: Tosnail)
Gemini 3.1’s Deep Research mode is built for projects that live in the space between hobbyist enthusiasm and professional standards. In this case, I wanted a real guide toward making my own book covers. I told Gemini:
“I want to create high-end, custom hardcovers for my existing paperback collection without removing the original covers or using glue on the books themselves. Run a Deep Research session to find the best ‘removable’ case-binding method and a step-by-step guide for calculating the ‘spine gap’ so the new cover stays snug when the book is opened and closed.”
The AI didn’t just skim a few high-ranking tutorials. It cross-referenced niche craft forums, supply store documentation, and long-form guides, then synthesized them into something usable. The most impressive section was its explanation of how to calculate the right sizes to maintain a snug fit without pinching. It was easy to follow despite the technical details. The guide respected the integrity of the original books while elevating them into something that looks and feels like a bespoke special edition. You can check it out here if you’re so inclined.
Deep Research like this can be applied to any specialized hobby where advice is scattered across forums, manuals, and niche communities. It helps synthesize professional standards and DIY ingenuity into a single, coherent starting point that lowers the barrier to trying something ambitious.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
The best business laptops for all budgets
Our top picks, based on real-world testing and comparisons