# Gemini 3.1 Pro For Beginners - All New Features Explained (Gemini 3.1 Pro Tutorial)

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=1Sxxscn4Vfk
- **Дата:** 20.02.2026
- **Длительность:** 12:43
- **Просмотры:** 36,200

## Описание

🎓 Learn AI In 10 Minutes A Day - https://www.skool.com/theaigridacademy
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Wan to learn even more AI https://www.youtube.com/@TheAIGRIDAcademy

Links From Todays Video:
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

(For Business Enquiries)  contact@theaigrid.com

Music Used

LEMMiNO - Cipher
https://www.youtube.com/watch?v=b0q5PR1xpA0
CC BY-SA 4.0
LEMMiNO - Encounters
https://www.youtube.com/watch?v=xdwWCl_5x2s

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

## Содержание

### [0:00](https://www.youtube.com/watch?v=1Sxxscn4Vfk) Segment 1 (00:00 - 05:00)

So, Google Gemini 3. 1 Pro is here and of course, we have to talk about it. So, in today's video, I'll be showing you guys exactly how you can get the most out of this model after stress testing it for a few hours and how you can actually utilize this model with some of the best use cases that are unique to Gemini 3. 1 Pro. So, first of all, make sure that when you are on Google Gemini, you scroll down and you check the pro tab because there are three separate options that you might misclick. And oftent times, considering that the model picker is a little bit different to every other app, just make sure you select this so that you're on Gemini 3. 0 Pro. So, one of the key things that I've realized that most people just completely didn't realize was in Gemini 3. 1 Pro is of course the Aentic Vision. Aentic Vision is now enabled by default. And let me show you guys why this is so powerful and why this just proves that Gemini 3. 1 Pro is arguably one of the most powerful LLMs today when it comes to multimodal vision. Now, before I dive into this quick example, Google's Gemini Aentic Vision is essentially a new capability that turns image understanding from a one-shot glance into an active multi-step investigation. So when you put images in, it combines visual reasoning with code execution. So the model can crop, zoom, annotate, and analyze images step by step instead of just looking at what it sees once. So essentially, the model will have a think, act, observe loop, and the model plans how to inspect the image, executes Python code to analyze it, and then looks at the updated image before answering. Now, this matters a lot because prior models didn't do this and most people won't realize that this is now an embedded feature that you can actually activate with Google Gemini 3. 1 Pro. And I'm going to show you guys how you can do that in a moment. And remember guys, this is super useful in those use cases where you need to zoom into tiny text serial numbers and instead of guessing, this just overall reduces the hallucinations. So, let's take a look at this example with Gemini 3. 1 Pro Aentic Vision. So if you look at this image, it's quite evident that the two characters in said image are Peter Griffin and Brian Griffin. Okay? And this is one of those images that are pretty hard to see, but for humans, as long as you squint your eyes, it becomes a lot easier. But for AI LM, it is pretty difficult for them to see this the way their vision processes the model. Now, if you don't believe me, you can actually ask Chat GBT, what with this prompt, and actually found this example on Twitter. Which two characters can you make out there to be? Even though that there isn't actually anyone there, the arrangement of the objects actually make it seem like they are. And you can see that it doesn't get who the characters are. It says a hooded grim reaper and a Batman symbol, which is a complete hallucination. However, if you input the same image into Google Gemini and you add the agentic vision onto that in the AI studio, which I'll show you guys how to do in a second, it actually identifies Peter Griffin and Brian from Family Guy sitting next to each other. So, let me show you guys how you can activate this and why this is so powerful. So, now if you head on over to Google AI Studio and you try Gemini 3. 1. And I know you guys might be thinking, why are we not using this in the standard mode? It's just because you always want to make sure that this tool is always called. It's not always called by default. Google's Gemini models aren't exactly always the best. Sometimes they do get confused when it comes to tool calling. So, we're using this to ensure. So, now what we'll do is we'll go to tools and then we'll do code execution. And so now that is essentially adding on that image analysis capability. And remember this is state-of-the-art in terms of multimotal reasoning and this is literally number one across the board. So let me use a different example and show you guys what happens when we analyze those images. So this image has trumped basically every single AI LLM. If you ask LLMs what how many fingers are on that image, they always get it wrong. There were actually six fingers on that image. And you can see here that ChachiBT even after thinking for the highest time possible the extended reasoning setting it says that there are five fingers shown four straight fingers plus the thumb five digits total which is completely wrong it is six now considering Gemini 3. 1 agentic vision the standard model gets this correct if we actually go one step further once we actually paste this image in and I actually add the prompt how many fingers are on this image annotate it to help reasoning and then we click run right here you're going to see that this model does something completely different and so with aentic vision you can actually see here that it pulls up the image, annotates the image, and gives you a final answer that has reasoning. Now, of course, this doesn't mean that it's never going to hallucinate. But the point is that with Gemini 3. 1 Pro Preview, and if you add code execution on in the AI studio, you're basically going to get a 10 to 5 to maybe even 12% boost depending on which reasoning task you're doing. Remember, Gemini 3. 1 is state-of-the-art in terms of visual reasoning. Okay? There is no other AI model that is better than visual reasoning from this standpoint. And when you add on the agentic reasoning, it just takes it to a completely another level. So now something that Gemini 3. 1 Pro is actually pretty good at is of course coding and 3D visualizations. Now if you actually want to enable this, of course, remember make sure you're on pro and make sure that you click canvas. The reason you want to make sure you click canvas is because as you can see there

### [5:00](https://www.youtube.com/watch?v=1Sxxscn4Vfk&t=300s) Segment 2 (05:00 - 10:00)

are seven different tools that Gemini will decide to fetch given your prompt and it can often make mistakes. So if you're coding something make sure you've got the canvas ticked on and then you can start to go. So canvas can be pretty useful to see different things whether it be a visualization, a 3D object, some kind of education. There are tons of different use cases which I'll get into in a moment. But all you want to basically do is prompt Gemini and basically ask it to visualize something using the correct libraries. So I've said, "Show me what a gunfiring animation is like. Make it beautiful, make it a cross-section. " And for example, it created this object. So this is pretty cool because if I, you know, click discharge, you can see there it discharges the round and you can see that it hits the firing pin, the thing explodes, and it comes out of the chamber. Now, I do know that this is of course not the arguably the best kind of animation, but I did ask for it to show me how it would look with a cross-section. That's why this is visible. And of course, I'm pretty sure I can make this better from a few rounds of prompting, including adding lighting, making sure that this is visually more appealing. The only thing I would try to say here is that the reason that I'm saying that you can do this is because in the next section of the video, I'm going to show you guys multiple examples where individuals are using Gemini 3. 1 Pro and the canvas feature to code things that enable a deeper level of education. Often times when we are learning things with LMS, it can be a bit difficult because there's no visual aspect to it. That's why when you're trying to visualize things with this, it is pretty useful to be able to say make it in 3D. Now, of course, like I said, you could always make this better with your prompting. Maybe you could say add some annotations on screen so I know exactly which parts are which parts. But overall, I think something like this is pretty cool for a rough draft just for understanding how different things work. So, now let's take it one step further again and look at some of the other use cases. And some of these are done by the Gemini team. And for some of these use cases, I've actually replicated them. And I'm going to be sharing the code with you all. The link will be in the description. What you're looking at is a tiny little app built in Gemini running on the web browser. And the goal of the app is to eventually produce a believable looking city. Every step along the way, I'm describing to Gemini my goal. Create me many programs to solve individual problems. Then I'm also asking Gemini to assemble back into the larger picture. The very first step is to generate a terrain. Human civilization rests on areas of where resources are. We need to figure out where people are going to live. Where is it that water is? the lowest flattest area is? Because we have these hubs, we can simulate how people travel. And every time they walk, we can create a road path. So the last step is to generate a satellite image. So this is like this drum roll time, right? This is like the most satisfying part of this whole project. I think it followed our road map super duper well, right? And what you're going to see is this the city but from a different perspective. And that's wild to me because we went from math to a generated image and now we're sort of exploring this fictional world just Gemini imagining how the city might look. With Gemini 3. 1, the model seems to be caring a lot more about the user intent. It's a lot better at improving things that you are intending to change. Now, I did come across this really specific use case of where you're essentially using a 3D model with Hyper 3D and then essentially you're basically taking that model with Gemini 3. 0 Pro and then you're essentially just altering the model. Basically 3D parameter fine-tuning. You can see right here that you're essentially adjusting the parameters and then once you do that for whatever reason you may want to do it, you're refining it. You're once again importing it and then you're able to edit it once again. So, this is of course one of those really niche things, but I did see it and come across it on Twitter. So, if you are into the 3D space, of course, there are different 3D tools that are useful, but this might be useful for you. Now, here is another use case that was really cool. An interactive boy simulation. I've actually got the code for this. But first, I'm going to let Google walk you through this because, you know, hearing him talk about this is really cool. After seeing a video of a Starling mmoration, I was inspired to ask Gemini 3. 1 Pro to code a simulation of one for me. After seeing a video of a stalling mmoration, I was inspired to ask Gemini 3. The model created this cloud of voids that interact in similar patterns to a real flock of birds. I then asked the model to make the simulation interactive so I could move the birds with my hands. I also asked the model to create music that would change according to the movement of the mmoration. All of the features in the simulation can be controlled, including the behavior of the birds and even the environment. I was stunned at how beautiful the final result was. And so you can see with a small amount of talking to Gemini and a small amount of, you know, editing the code. You can see I managed to generate my own version here. That's still pretty cool. Maybe it's flying a little bit too fast. You can see this does actually follow my

### [10:00](https://www.youtube.com/watch?v=1Sxxscn4Vfk&t=600s) Segment 3 (10:00 - 12:00)

mouse, which I find to be really, really cool. Of course, like I said, I'm going to leave the code in my community. If you just head on over to the prompt section and you scroll down to the Gemini 3. 1 interactive board simulation, all of the code is here. So, if you want to run anything from this video, all of the code is here because of course I can't put it in the description. And so, yeah, this simulation is pretty cool if you ask me. It just simulates the birds. And of course, since I don't have the hand tracking, it just simulates, you know, and follows your mouse movements, which is pretty interesting if you ask me. Of course, you've got the environment as well. You can change the sun height, all of those things. You've got numerous effects such as the bird size as well. I mean, it is really, really cool. I think interesting. You've got mouse attract and repeal. So, I think this is something that, you know, just goes to show in terms of visualizing different things, this may be helpful or useful. Remember guys, Gemini 3. 1 Pro is of course good at multimodal. So, if you're ever doing this, don't forget to input a picture. Of course, don't forget to describe it exactly how you want it because this model is basically really good across the board. And once again, if you do want to code with Gemini, of course, once again, if you have, you know, really in-depth props with whatever it is that you do want to use, activate the canvas feature, then of course, enter your prompt. And sometimes you might need one or two extra pumps just to get it and refine it. And you can see here they literally had an ISS orbital tracker. And this was from the Gemini official demo. So you can see here that this Gemini 3. 1 demo definitely showcases exactly the different ways that you can actually use this model in a vast variety of circumstances. Now I'm going to show you guys another use case that I won't personally be using, but I see people using this all the time. So I wanted to just make a quick video on this. Now, of course, there are SVGs, which some people do use for different animations and all these kind of different things. Me personally, I don't really use them. I mean, I think they're pretty cool. But I will say this though is that often on the first time, the SVG animation may not be correct. So, you can either decide to reprompt it again and tell it to fix whatever issue isn't working because often times it doesn't oneshot something. So, if that does occur and it just doesn't work, what you can do is you can hop on over to Google AI Studio because what I've tended to realize is that it does reason for longer over there. I'm not sure what the reasoning is, but maybe they just do extended thinking on the AI Studio platform because they want to give more compute to developers. I'm not really sure what the reason is. So, if you're having an issue with testing Gemini 3. 1 preview, you can actually go on over to the AI studio and you'll see that it does tend to reason for longer. This one ran for 700 seconds before it actually finished. I'm not sure if this is better or worse, but you guys can be the judge of that. In my honest opinion, I think the first edition is usually a bit better, and then after that, you can make adjustments. So, if you did enjoy this video, don't forget to leave a like, comment, and of course, if you want any resources from this video, I'll leave a link to them down below.

---
*Источник: https://ekstraktznaniy.ru/video/11975*