GPT 6 Astra is the Best Model for Building Games (4 Real Examples) transcript

Hey everyone, GPT Astra just launched. And I could talk about how insane the benchmarks are or hype it up like all the other influencers, but instead, I want to share how it's brought back a part of my childhood. Because this is by far the best model I've ever used for building games. In this video, I'm going to show you how I created a Starfox style space shooter, a first-person shooter on a moving train, a real-time strategy game like Starcraft, and my favorite category, a roguelike deck builder featuring an AI founder trying to build a generational company.

I worked on all four games at the same time on Friday night using Astra on medium. I think you can get quite far even on the $20 ChatGPT Plus plan. You really don't need Astra higher or ultra. Medium can go a long way. And at the end of this video, I'll also talk about why the Astra launch got some backlash and why I think it might not have been entirely Open AI's fault. But first, let me show you how to build some games.

Okay, so here's the setup. To build great games, you need the ChatGPT desktop app. You need Astra selected here in the model selector, and you also need two additional tools. Blender is a free open-source tool to create 3D models and animations. And Godot is a free open-source game engine that turns those models into something that you can play. And you know the funny thing is, I actually have no idea how to use Blender or Godot.

I mean, just look at Blender's UI. It looks very complicated. And all I did was I just asked ChatGPT to install Blender and Godot for me so that AI can actually drive the apps itself, right? Simple as just saying let's install Godot MCP and Blender MCP. All right, so now let me show you the four games I built and how I built them, starting with the Starfox style space shooter. So, Starfox is a space shooter from the Super Nintendo days.

And I really loved playing it as a child. I want to start with this because I've been using it to test models ever since GPT 5.5. Here's GPT 5.5's version of Starfox. As you can see, it looks pretty great. It's got the polygon graphics. It's got enemies approaching me. And it looks pretty fun. Now, here is GPT 5.6's version. And I've also compared it with Fable 5. So, the graphics look a little bit better. Look a little bit more polished.

There's also reactions here from your wingmates and so on and so forth. And to be fair, in these early demos, I just used 3.js, which is a JavaScript library for 3D in the browser. You can get quite far with 3.js. But for Astra, I decided to use Blender and Godot, too. And that gives it a lot more tools to work with. I can make the graphics a lot better. Here is Astra's version. As you can see, it looks a lot more impressive right off the bat.

Let's go ahead and play it. So, I got the wingman. I got ChatGPT images to generate some wingman profiles for me. There's ship coming. The levels are more interesting. I think there's more ships here that I can attack. And if you keep going, there's also obstacles. There's different stages, right? There's a stage over here. And there's falling blocks that you have to avoid. And I also added a boss at the very end. So, let's go ahead and get rid of the ships.

And here it is. This is kind of like a starship destroyer kind of boss. And if you shoot at it, different parts of the ship will hopefully break apart. So, let's try this. All right. So, let's shoot at it. So, here you go. We broke the wings. And the other wings. And you keep shooting at it. And eventually, the boss is going to die. All right. So, that is basically the game that I built. And now, let me show you how I built this, right?

So, I asked ChatGPT and Astra to install Godot and Blender, and it installed. And then I pointed it to this blog post that an OpenAI employee wrote on how to build games with Astra. So, basically, Astra read the blog post, and the best practices are describe the player experience first, establish a look with concept images, build repeatable play tests early, and use Blender for distinctive assets. Okay. So, then I asked it, "What kind of visually impressive and super fun games should we make?

Give me your top five ideas." And it gave me a bunch of ideas. And then I decided to just tell it to build Star Fox. Okay. So, then it gave me some more ideas. Star Fox meets a collapsing mega structure, which is what we ended up building, the game that I just ended up showing you. So, then I just told it to go ahead and start building it, right? Then it built some sort of game that actually was not that fun. And I actually got ChatGPT in another thread to generate images for what the game could look like.

And this is the image of what the game could look like. All right. So, you can see this is a collapsing space structure. There's obstacles along the way. And I asked it to add teammates and allies and make the graphics look better. Then it kind of went off and did a lot more work. It generated a representative example of what the what the game should look like here. And it also generated a profile images for my wingman.

And then I just started giving Astra me more feedback, right? So, the game looks better, but it's still boring. Add power-ups, make the enemies harder, the space station make it look like it's falling apart. And then the boss fight should be more interesting as well. And another thing that I did is like I just asked it, "Hey, how would you continue to improve this?" And it gave me a bunch of ideas here. And then I asked it to make the boss more intimidating and implement all of its ideas.

One thing that Astra is really good at is it can use browser use and use computer use to play test its own game and just make sure that it's fun. Uh so there you go. So, after basically 3 minutes of conversation, we ended up with the game that I just showed you. Now, let me show you the next game, which is a first-person shooter on a moving train. So, this game is called Dustline. And let's go ahead and play it. All right.

So, you see here that I'm on the cart of a moving train. There's enemies shooting at me. There's gun. I can reload. And also, I can aim the gun. Like this. And looks like there's a flamethrower. Let's go and pick up the flamethrower. I can shoot flames at the enemies. Also, rocket launcher. Let's shoot the rockets at these guys. All right. It makes bubble explosions. Let me switch back to the gun. That gun's the best, I think.

Let's get rid of these guys. And always with these games or maybe tech demos I like to build, I always like to add a boss battle at the end. So, let's get rid of these guys first. And let's see what the boss looks like at the very end of this game. All right. Here we go. Some of the enemies Okay, some of these enemies fly. They have jetpacks. They just shoot at me. All right. Hopefully, I don't die before I get to the boss.

And you kind of see the boss already, right? It's basically like a huge ship. All right. And I think for this boss, we probably should get some rocket launchers. Flamethrower is probably not great. Let's try using a rocket launcher to kill it. And let's see what happens after we kill it. All right. The boss is not really shooting at me. Maybe I need to make it harder. Looks like I ran out of rocket ammo. There we go.

And let's go here, take control of the train, and that's the game. Okay? I mean, look. It's really more of a tech demo than a real game. But once again, let me show you how I built this. So, I followed the same process, right? The first process is, "Hey, let's build a first-person shooter with Godot and Blender. Give me your top five ideas." And it gave me these five ideas. And honestly, I thought about maybe doing some sort of Left 4 Dead zombie shooter.

Kind of explore that. And then one of the ideas was the last train home idea. And that kind of inspired me to maybe make this a moving train shooter instead. So then I got ChatGPT generate an image of what it's going to look like. Now, the image looks a lot more impressive than the game that we built, even though the game that we built is, in my opinion, pretty good. So maybe the next version of OpenAI's model can actually build this in one shot.

But it's always good to provide a reference image so that AI knows what you're going for, right? And then here's the idea, Titanfall meets moving train. It loved the idea. And I asked it for a three-milestone plan. So you can try to get it to build in one shot, but I think it's good to kind of split it into like a couple of milestones so they can test it along the way. And it built something fairly basic, okay? It didn't quite look like the way I wanted it to.

So I provided a reference image here again. And I told it to make the graphics look better, make the gun look more detailed, and instead of robots as the enemies, make it actual humans. And add proper death animations and stuff. Okay? Then it kind of went off and built it again, and I provided more feedback, right? And this time I think I actually had to leave. I had to go out with my family. So I started providing feedback using my voice and ChatGPT on the mobile phone.

So this is like super stream of conscious feedback. Have more different types of enemies. The AI is pretty dumb right now. This is not like super quality feedback, right? And I also told it to make the rocket launcher better and add more bullets to it. And so on and so forth. And that's it. That ended up with the game that I just demoed to you guys, right? Okay, so the reason I want to go through these conversations is I really want to emphasize that anyone can do this, as long as you have ChatGPT, Astra, and you have Blender and Godot MCP installed, which are free and open source.

The three-step process is just brainstorm some ideas, get ChatGPT images to to some concept art, and then ask AI to build and give it feedback along the way. And this whole process took me maybe like 30 minutes to hour to do. While at the same time building three other games. All right, for the next game I want to show you is basically related to my favorite game of all time, which is StarCraft. You know, I played StarCraft non-stop in high school.

I skipped classes to play it. And funny story, my defining moment in StarCraft was playing with my Korean friend three versus one stomping a computer, and then I dropped a nuclear bomb on his base and kind of ruined his perfect record. That That's my favorite story. So, there's rumors that Blizzard is working on a new StarCraft game coming out this fall. So, naturally I had to see if Astra could help me build a StarCraft level.

All right, so let me show you the game that it built. So, the game is called Ashfall, and let's join the fight here. And you can see here that this is a real-time strategy game. I got marines that I can control, and I got siege tanks. And the siege tank can actually siege if you ever played StarCraft before, and we can build marines, medics, and tanks here. And let's also build some turrets while we're at it. So, it's basically like a base defense game.

There's a zerg coming at me, and I got to capture the relays and defend myself against the zerg. So, next wave is coming soon. We got some medics that can heal my marines. And you can see the controls work pretty well, right? And I can also build some turrets, and here's another tank. Let's put it into siege mode. And finally, I think there's other options here like orbital, which I think it's like a orbital strike. Okay, that's pretty cool.

And I can also add stem to my marines. I can press E to put them into overdrive. Let's wait for the next wave and see what happens. All right, here we go. I press E to put into overdrive, and then they'll kill the enemies faster. Oh, and here's the ultralisks that came out. So, I probably should go back here. Okay, it looks like I'm actually losing right now. There's >> [laughter] >> too many zerg. I think my base is falling apart a little bit.

Maybe I can use the orbital strike again. There we go. So, I survived this level, but I probably won't survive the next level. All right, so that's the game. I I want to see if Astra can kind of build a real-time strategy game, and this is again just a cool tech demo, but I think it's pretty good, right? So, now let me show you how I built this. So, in this example, I knew that I wanted to build StarCraft. So, I asked it to create an image of a real-time strategy game like StarCraft, except with modern graphics, and this is what it made.

And once again, the concept art looks a lot better than actual game that we built, but I think it's useful to ground the AI on what we're looking for. Okay, so after it made the concept art, I basically just asked it to make a real-time strategy level with Blender and Godot. Player control the human Terran faction, and the enemy is the Zerg, and has to be super fun, you know. Once again, I feel like as the models get better and better, I become more and more lazy with my prompts.

So, it went off and started building the stuff, and then I said, "Hey, make the graphics better, and why don't we do a level where it's about Terran defending the base against Zerg." And then I mentioned that the siege tank obviously has to be able to enter siege mode. This is like StarCraft 101. Okay, then it made the game, and making a fun game isn't just about the graphics, obviously. It's about balance, it's about core loops, it's about progression systems.

So, I told it, "You made the game too easy. Add some more additional characters like medic, and then add three different types of Zerg enemies, and make the enemies more aggressive." Okay, and then I asked it for better lighting effects. So, quick tip, I'm not a game designer, but I found that if you want to make your 3D games look good, just asking for better lighting effects makes a huge difference. If you scroll down here, and then I play tested the game myself, and I told it that wave five onwards is way too hard.

Make it easier, and so on and so forth. And that's it. So, I maybe took like a dozen or so prompts to do this, and it's obviously not StarCraft, but it's a pretty a tech demo for what a real-time strategy game made with AI could look like. All right. So, the last game I want to show you guys is in my opinion the closest to an actual game. It is a rogue-like deck builder called No Moat. And basically, it's about the AI founder who's trying to build a generational company and trying to overcome obstacles along the way.

So, let's go ahead and try playing it now. So, we got three choices. We got a dropout, the professor, and the ex-Fan employee. And why don't we pick the dropout? And the first thing you see is a map of the levels that you have to beat. And [snorts] there's three different stages. And let's go ahead and get a companion first. So, we can get a designer, a principal engineer, or a research scientist. I'm going to have to go with a designer to start.

And here is our first enemy fight, right? So, basically, it's a rogue-like deck builder. I got a bunch of decks here, and I can go ahead and fight the enemy. Let's go ahead and just use Ship It to get rid of some these rapper startups, and let's get our designer on the board. All right. Let me rate limit this scope creeper so that it doesn't actually attack. And Ship It, get rid of this thing. And yeah, there you go.

So, there's a new player coming, and I can also press this to refresh my cards. So, let me just go ahead and play it for a little bit longer. Let's add a GPU rack here to block the enemy's attacks. And Ship It again. And there you go. Here's the mini boss, the Acquire, that wants to acquire our startup before we even got started. So, let's go ahead and try to beat it. Let's move our players around. And let's hit this thing, and let's do hype tweet to increase our attack.

All right. And since it's got three health left, let's just kill it directly with Ship It. There you go. So, we won the game, a victory victory. And now we have a choice between adding more companions or getting more cards. So, let's say adding more companions. And this time we want to add a research scientist. We're going to have a research scientist. All right. And here's a bug investigation. And again, we can kill the bugs, bring our pieces onto the board, and see what other enemies are coming out.

So, we got a memory leak. Let's use Hypetweet to kill it. And yeah, so that's basically the game. There's different levels, and every level has a little mini boss boss fight. So, now there's like a may major bug that we have to beat. And unlike the previous tech demos that I showed, I just spent a lot more time into this game to make it good. Because as you know with roguelike deck builders, they got to look great, but they also have to be somewhat balanced, right?

And by the way, for this game, I'm going to include a link for you to play it directly in the YouTube description and the first comment. I'm actually pretty proud of it, and please give me feedback, and I'm going to improve it over time. All right, so for this game, I actually started in Claude using Fable, but that's only because I didn't have access to Astra for the first day. So, you can pretty much do this using Astra and ChatGPT.

But basically, I followed the same process. First thing is I have to build a game like Slay the Spire and Wildfrost. It needs to be super fun. Ideally, there's some memes about the AI race in there, too. And give me your five best ideas. And the very first idea that they came up with is No Moat. You run a AI lab climbing a ramp from seed round to AGI. Units are researchers and GPU clusters, and there's different types of cards.

So, then it recommended a bunch of other ideas, and I told it that I want to do No Moat. So, then I asked it like, is this very complicated to build? And it said it's pretty easy, and here's a rough plan. And this time I actually told it to write an HTML spec for the game. And let me kind of show you what the spec looks like. Okay, so here's the spec that AI generated. I have a spec scale where I split the spec into product design and tech stack.

Right, so product here is the solution, here's the core loop, and here's the requirements, and design. This is what it looks like. So, the thing about Claude, the big kind of drawback, is that Claude doesn't have an image model. So, in terms of making the designs, it has to generate stuff probably using HTML and using emojis. So, the first version of the game actually just looks like this with emojis and different cards, just to kind of get the gameplay and get a feel of the game.

Right? So, here's the design, here's the emojis, and so on. And then I kind of realized that I really wanted better images than just emojis. So, after playing the first version, I decided to move over to ChatGPT and Astra became available then, too. So, let's go ahead and move over to ChatGPT and let me show you the thread. In ChatGPT, Claude had already built the initial version of the game with emojis. So, I told it to first export the code base and build a playable version of the game.

And I said, "Hey, should we use Blender MC for this? I have no idea what I'm doing, so I need your help." And it actually didn't recommend using Blender. It it recommended using images to just generate art. And scrolling down here, and I say, "Hey, why don't we figure out one convincing fight first? Let's make the graphics great. And I really like the diorama look that you just saw in the demo that I just played." So, I kind of gave it instructions to do that.

And so on and so forth, and here we go. And you can see here that I started generating the artwork. So, this is like, again, one of the biggest benefits of ChatGPT and the whole OpenAI ecosystem, the ChatGPT image model, it's just great for generating game art. So, it generated both the bad guys and my hero characters. And it also generated the game stage, some of the bugs, and I items, and so on. Okay? And then I really wanted each character to appear animated.

So, we kind of went off on a side track. We tried to use Blender to generate the characters, but the 3D characters didn't look anywhere near as detailed as this 2D art. So, I decided to stay on the 2D side. All right. Then, I asked it to generate more heroes. So, you have some characters here and then you have probably the most stereotypical image of a PM and a designer I've ever seen. So, this is the PM. It's got the new balance shoes.

It's doing sticky notes. And the designer looks something like this. Right? Uh so, yeah. So, AI can be pretty stereotypical. Then, I started playing the game and just give me some more feedback. Just like asking it to check the balance, look at other proven games in this genre, and see if there's any best practices from them. And also generating more characters, more enemies, asking it to make each enemy unique, and so on and so forth.

And obviously, I did not look at the code a single time in this whole process. So, I asked it to generate a vibe coder and a principal engineer. And going down here. And I told it to add the X fan employee as one of the hero characters. All right. And then here's the illustration for the X fan employee. Looks pretty burned out. And finally, I told it to ship this game in ChatGPT sites. So, unlike the 3D games, which are in Godot, like this is a 2D game, so it's pretty easy to ship to ChatGPT sites.

And then I said you can play it here. So, real quick with ChatGPT sites. Once you generate the web page, by default, it's only available to people you share with. So, you got to hit share here and change the permissions to public so that anyone on the web can play it. Then, I asked it to fix some more stuff and so on and so forth and then do a polish and balance pass and make it ship ready. And there you go. Oh, another thing that it did was a lot of the enemies were rival founders, but I think AI founders generally try to help each other.

So, I wanted to make the enemies more realistic. So, now the enemies became stuff like Skull Creeper, Procurement Manager, Reply Guy, so on and so forth. And of course, I also got AI to generate images for each enemy type. I think this is the predatory VC. It's pretty funny. Looks like the Joker. Yeah. Oh, the last thing I told it to do is I wanted to make it playable on both web and mobile. So, I tested it on mobile and it looks pretty bad, as you can see here.

So, I asked it to make it look better on mobile and just make a bunch more improvements along the way. And there you go. That is the game. So, just like the other examples, it's just like a lot of back and forth, giving it a feedback, and making it better over time. For this one, it took me maybe like 2 hours of back and forth to make a actually playable game instead of a tech demo like this. And like I said, I love this genre of roguelike card builders.

So, I'm going to keep improving this game and make it better even just for myself. All right. So, to wrap up, it's funny because just a few months ago, I was using things like plan mode and elaborate prompts to try to get AI to build games and not really succeeding. But as you can see here, most of these games I built from just exploring an idea with AI, generating a reference image, and giving feedback to Astra along the way.

And half of the prompts that I showed you were just lazy voice dictation while I was out with my family. The thing is, these models have become so good that you can just talk to it like you talk to a capable game designer or developer, and they'll just get it done. All right. So, now I want to talk about Astra's launch. So, the launch actually got a little bit of a backlash because what happened was OpenAI made a blog post and did press releases and did influencers saying, "Hey, I've had access to this model for a week and here's how cool it is." And the problem was no normal person could get it.

And in OpenAI's defense, this may not have been entirely their fault. As you know, there are probably third parties involved in deciding when a state of the model can get released. But, personally as a creator, I think hyping this stuff up is completely unnecessary. I think launches should happen only when the model is actually available to everybody. The reality is that there's so many models coming out almost every week now that not every model is going to change your life.

So, I don't want to overhype this stuff. Instead, what I want to do is just give you my honest opinion and share what I build with these models so that you can try building them, too. So, now that we all have access to Astra, I encourage you to build the games from your childhood and also your imagination. Build them with your kids, too. It's an amazing family activity. And as always, thanks for watching and please like and subscribe if you enjoyed this video.