
Gemini 2.5 just released and it's a stellar coding model! 🌟 In this video, I demonstrate its capabilities by building a simple 3D game using the Cursor platform. Watch as I effortlessly create a technical blog with a neo-brutalist theme, generate markdown blog posts, and develop fully functional Tetris and Asteroids games. 🕹️ The video showcases the model's impressive instruction-following and code-generation abilities. Tune in to see how far we can go with a full-stack application created entirely with natural language prompts! 🚀 For those asking about my voice dictation tool - it's Wispr Flow↓ https://ref.wisprflow.ai/ddai 00:00 Introduction to Gemini 2.5 00:23 Demo: Building a Simple 3D Game 01:15 Creating a Full Stack Application 01:54 Adding Blog Posts and Sections 03:36 Developing Game Pages 08:15 Implementing a Chatbot 10:55 Building an AI Image Generator 12:51 Conclusion and Final Thoughts
--- type: transcript date: 2025-03-28 youtube_id: d7H3HzWgsAs --- # Transcript: Gemini 2.5 Pro is Now Available in Cursor and It's Amazing... Gemini 2.5 just came out and it is an incredible model at coding in this video I'm going to show you what it looks like with en cursor which it was just released in and just a quick demo that I wanted to show you I was playing around with this model earlier this week after it came out and one thing that I found incredibly impressive with the model is its ability to follow instructions but also generate coherent code if we look at the example on the screen here now mind you this is sped up about four times what it's doing is writing a simple but very involved 3js game as we can see here there's a car there's a city I'm asking it a number of different questions related to the game to generate buildings and Roads and emulate other cars and with every single request that I put through to the model it was able to generate a very long but comprehensive and most importantly working example of what I had asked for this is just a very simple game but I built this in just about four prompts not even a syntax error what I'm going to do within here is I'm going to leverage a little bit of the agent mode with incursor but I'm also going to be leveraging some of the command K and just highlight some sections when need be and we'll just see how far we can get with a simple full stack application the first thing that I'm going to do is I'm just going to bun create next app within the root of our directory and what we're going to do within this is I'm going to start with our homepage and I'm going to say I want to build out a technical blog called developers digest and I want the overall theme to be Neo brutalist let's make sure we have a broken out header as well as footer with a number of placeholder links in both I'm going to go ahead and send that through and once we have that back I'll go ahead and take a look at the results all right here's the theme of our website and if you're not familiar with the Neo brutalist theme I think it was popularized originally by gumroad so they have all these fun sort of colors throughout within it so now what I'm going to instruct within this is I want a technical blog post section I want three three different example blog posts all within the markdown format let's have them about air related topics and I want a specific folder for all of those blog posts now the one thing you'll likely run into with Gemini 2.5 Pro is the rate limits they are working on trying to extend the rate limits with en cursor wind surf as well as their API right now but that's something that very well could take some time now if I go within the Articles tab we see these three different articles we have the rise of AI modern web development understanding large language models so on and so forth if I go within this we have our blog post here and the nice thing with markdown blog post is the way that they're structured is they're going to have a similar format to this where you'll have a title date summary sometimes you'll have some other metadata within this and that is the portion that is going to be parsed that is put Within These various areas next let's add in three other sections as well as some placeholder Pages I'm going to say I want to add in three cards below below that section that we had just created on the homepage and I want each of those cards to still be in the Neo brutalist theme but I want them to be different colors I want the first one to be games I want the second one to be chapot and then I want the third one to be AI image generator and with all of these I want them each to have placeholder pages with a really nice hero area that will have the title of each page and I'll go and I'll send that in now we have our three sections we have play game and we have this new games page we have our chatbot page and then finally we have our AI image generation what I'm going to do is I'm going to work through these different pages and we're going to see how far we can get the first page that I'm going to go with in here is I'm going to say I want tiles for three different games I want Tetris I want asteroids and then finally I want a really creative 3js game with a plane a Runway as well as some Scenic mountains now within here we have some cretive images Tetris asteroids as well as Mountain flave so now what I'm going to say is based on all of the games within the game style on the games page I want to build out all of those games in full on their own respective pages in a Neo brutalist theme it will be interesting to see if it will be able to create these games in one shot as well as how well it will actually fit the overall aesthetic of our site in this case it ask for some clarification and I'm just going to specify for it to proceed now in terms terms of the games there are Pages for each of the games but they're still just placeholders what I'm going to say here is let's actually make all of the games actually work set up all the game logic in an objectoriented manner we'll go ahead and we'll send that through and then we'll Circle back and see what it generates here it asks if we wanted to start with Tetris first I'm going to go ahead and say Yes it just went and wrote out a ton of code here we have about 500 lines of code or so I'm going to go ahead and save that out and now if I go over to our Tetris game we now have what looks to be a Neo brutalist Tetris game I have the yellow shapes here I have the green shapes here and it looks like it works I can even rotate these shapes like you would expect I I can have them go down faster if I hold down the button I can restart the game with the Press of R and then similar thing is I can do a hard drop to get them to the bottom of the screen it worked with Tetris without a flaw now let's instruct it to do the same thing on asteroids I'm going to say that was perfect let's continue onto asteroids and in a similar fashion here we have another impressive generation and here we go I can go and I can shoot all of these different asteroids on the screen here I can also go and I can Traverse we have game over I can click R and we have all of the instructions here again very impressive also very fun to be able to just create these old school Atari style games on the fly as well as it fitting the overall aesthetic of our site finally and possibly most impressively we're going to generate the Mountain Flight game now I want to generate the 3js flying game I want to make sure we have obstacles there's also Collision detection and there is also the ability to shoot from the plane as well I want the sky to be blue the ground to be green and they're also to be a Runway since the first two games did do really well and obviously it can generate several hundred lines of code coherently without any issue what I wanted to do is just add some even further detail to our 3D game here we see within here the agent determined that it is going to need to install 3js as well as the types for 3js and we're going to see it just work through the structure of the page similar to how we did within asteroids as well as Tetris now we see this generation here now what's pretty amazing with 3js is what it's able to generate in just a number of lines of codes within here we have under 300 lines of code but you would be surprised if you haven't used 3js before what you can generate with that amount of code here is our game there definitely is an issue it does look like it's rendering below potentially the grass type of area but on refresh if I just refresh repeatedly it does look like we do have some objects within the regular area there it looks like the plane isn't actually really shaped as a plane first of all and more importantly it's actually rendering below the world I want to be able to make sure that it's above where the runway is and I want the perspective to be from behind the airplane but I also want the ability to control the viewport with my mouse this one definitely does look like it is struggling a fair bit if I did reprompt it I wouldn't be surprised if I do have a decent generation I have tried something like this before where it does give a relatively good generation but in this case it doesn't seem to be but with just a few shots in this example it doesn't seem to be doing it quite yet next what I'm going to do is I'm going to try and set up a simple chatbot within this style of UI now what I'm going to say is now on the chatbot page I want to Leverage The V4 version of the open AI SDK and I want to integrate GPD 40 as as well as gp40 mini within a dropdown now remember I want the chatbot to be in the Neo brutalist theme and let's make it colorful as well as all the messages that are sent back and forth have them be fun colors that are changing with each message I'll send that in now what this task will involve is we'll have to set up an endpoint I'm just going to plug in an environment variable for our opening ey API key what I'll do within here is I'll just put in a and then I'll grab my API Keys all right here is our chat bot we have our dropdown if I say hello world I'll send in a message we see that thinking there we have this is a simulated response from gp40 mini I've already added in my environment variable for the open AI API key let's wire that up to have the messages stream back as soon as they come in if I try this out a hello world all right so what's interesting with this is what it actually decided to use if I go over to the chatbot page is even though I specifically asked for the V4 version of the open AI SDK what it decided to use was the AI SDK from verel now that might be because we're within a nextjs project and it has a bias towards but it is interesting that even with it taking its own Direction it wasn't quite able to update this one of the things that I've noticed especially for newer libraries is sometimes the training data can be weighted towards different versions and sometimes it can get a little bit mixed up there could be a deprecated method for instance in the latest version when it goes to try and install that package and is trying to reference something in the training data that might not be the way that it's intended to be set up right now that's just one Theory overall it's almost there we could probably get there with just one prompt if I take this and I could say on the front end it's rendering the following and I'll paste all of that in now just to try this one more time I'll say hello world and there we finally go now we have a working chatbot I can say write me a short story on llms I'll send that in and there we go it's generating all for us we have these nice fun colors it's streaming out we have all of these nice fun colors and it's relatively fast as well gbd4 mini it's obviously a smaller less powerful model but what's neat with this is I've built this entire application with just natural language now just to move on to the final step here but on the image generation page I want to build out an image generator on the left hand side I want a pan to have a user put in the description of what they like and on the right hand side I want to show the image and finally I want to store the previous generations below and I just want those to be referenced from the URL that we get back from open AI I specifically just want to use Dolly 3 in this example another question that I do get a fair bit on my videos is what program do I use use to transcribe my voice and I use whisper flow so I'll put a link to that within the description of the video if you're interested within one shot we have this generation here within here we have this new API m point for image and within this request it looks like we're making that request to Dolly 3 we're using the openingi SDK in this case and it does look like it potentially is exactly what we want let's just go and save this out and now here we go we have our AI image generation model I'll click generate we'll wait for it to generate we also have this nice loading state where it's fading in and out within the area where it's going to generate an image and now this is actually a good error because it's just indicating for us to set within our nextjs configuration here to include this I'm going to say include the following domain and I'll just paste in the error that we have there so within nextjs you just have to make sure you add in the images as well as the domains that coming from now I'll try a cute cat wearing sunglasses in a Neo brutalist style I'll go ahead and I'll send that in and then there's our generation history now I'll say a cute dog wearing sunglasses in a Neo brutalist style and hopefully we'll have those Generations just below on the page here and there we go so we have our dog we have our cat overall very impressive I'd probably say this is one of the best if not the best coding model that I've used for these day-to-day type of web development Tas we were able to generate two successful 2D games we were able to get started at least with a third 3D game and then finally we were also able to set up the basis of a simple blog and we have this full stack site now otherwise that's pretty much it for this video hopefully you found this video useful maybe you learned something potentially on how you can prompt these different models now obviously I'm coming at this from the perspective of coding for many years so being able to direct llms in a particular way I am biased to having a little bit of experience in building applications myself but hopefully for people that are maybe a little bit less experience you could take something from some of the prompts potentially that I use overall just see what Gemini 2.5 Pro can do as well as what you can do within cursor that's pretty much it for this video if you found this video useful please comment share and subscribe otherwise until the next one
Weekly deep dives on AI agents, coding tools, and building with LLMs - delivered to your inbox.
Free forever. No spam.
Subscribe FreeNew tutorials, open-source projects, and deep dives on coding agents - delivered weekly.
Technical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.