New Absolute King of 3D & Blender - GPT-6 Astra transcript

GPT -6 Astra is insane. I'm seriously, it's absolutely mind blowing. We was just recently discussing Fable 5, Metis 5, and now GPT -6 surpasses it significantly. We have crazy good computer use and something no one was expecting. OpenAI shows trailer when chat GPT works in Blender and it's insanely good. I did some tests guys and I guarantee it will blow you away. Less talking, let me show you. And I want to start with something that I was actually working on and it was character like Animal.

And I was actually building my own game and I needed it to rig it. And I was thinking, OK, maybe that's a good chance to test it out. And what happened next, I was not prepared for. But for a bit comparison, let me first show how Fable 5 .1 did that. I'm going to give all the APIs and stuff. I can explain setup just in a bit. And the task is simple. Generate a character, rig it in animate. Animation like simple walking cycle, maybe this animation.

Here's the Fable 5 .1 result. And it was actually not bad. You can see here's how it looks and we can actually see the animation. First of all, I was surprised that weights are not really broken. But you can see like a little bit here, there like feels a little bit rigid, but not bad, not bad. And here's how GPT -6 did it. Dang, look, I mean, boom. It also added like this animation. Well, this animation is not that cool, but walking animation, I was like, dang, what?

And there is no problem with weights. I don't even know how it actually managed to do that. And what was surprising me, I was not really asking to add rig and it actually created control rig. I mean, simple. And it actually added like inverse kinematics here. Dang, like I was no way, no way. I mean, it's so much that I will be able to test. And I mean, I was actually working with the animation and stuff. So just look at that.

I mean, that's a simple control rig and I was not even asking about it. And later, GPT also did some renders. I mean, it's crazy. But what you're going to see next, guys, I was not prepared. But first, let me quickly explain my setup, how we're going to do the test. So I have Codex CLI. That's much more convenient. And we're going to do test with quite cool Blender plugin by Hicksfield. This way, we're going to get really, really far compared to computer use.

I mean, you probably saw the trailer from OpenAI where they show like, you know, like, do this, do that. And it indeed can do that. But but hear me out. I set the plugin. I gave it a test. And guys, it was saving a file for 10 minutes. You can see the time lapse on your screen, how it does it like it takes eternity. So the task was like to create a samurai from the image. And then like, I mean, that was taking like so long in one hour.

It was able to do this like this because it was clicking and stuff. So computer use is cool, but it's so, so slow. And I mean, if you provide it with MCPs that expose all the engine functionality, it will do way more than that, which I'm going to show you. And here's, by the way, how GPT did the samurai with MCP. Well, building characters from primitives is not really working well in general. Right. So we're going to wait until GPT learn how to work with polygons and stuff.

But for now, we have amazing 3D AI that can generate really, really good polygonal mesh. And we can actually provide all access to GPT. This new Hicksville plugin is actually free and it provides a lot of tools for the agent. You can plug it like with anything, like with Claude, with GPT. It's very easy to install. I'm going to drop the link and I combine it with Hicksville CLI as well, because now my codex has access to 3D AI, image generation AI, and plenty of other options that it can trigger from Hicksville.

And if you want to use 3D, 2D video generation, you're going to need to log into your account or you can use it for free as an MCP. Now, next test and here we're going to see a huge difference between Fable 5 .1 and GPT 6. Idea is very simple. I have my creature. How about we build some location? Let's start with simple location. And I'm going to be generous and I'm going to allow it to use any other APIs to test Fable 5 .1.

Now, I was actually using the Hicksville plugin. You can access it right here and GPT 6 will be also available here in plugin directly, which is cool, but of course it's optional. OK, work is done and here is the result. So Fable 5 actually generated few objects and it actually reminds the scene. It's not like far in shape, but it looks like very rigid, right? And it's still impressive, but not as impressive as GPT 6.

So let me finally show it to you. And just so you believe that was indeed one prompt, I have complete time lapse. So here's how GPT does it. And I actually really like this approach. It does some blocking and start gradually generate some objects somewhere. It did it procedurally somewhere. It generate objects actually for this whole scene. It generated just few objects. It generated main islands and this shrine itself and position.

All the backgrounds was procedurally generated. There are flames. They are animated. Also procedurally done and this amazing video. I mean, I was actually blown away when I saw that this look really, really good. And, you know, I'm the guy who actually see this firsthand. And this jump in quality is insane. Now, let me show you one more test. Here's the reference image. And the task is to create all of this procedurally.

So in this case, we're not allowing it to use any 3D AI or generating images. Just do everything procedurally. So we're going to see how it does. Also, again, I recorded complete time lapse and you can see the approach, right? It starts with primitives, then it adds more and more and more details. And you can see like how it already looks so cool. Right. And it also did such a great job setting up the light in the scene.

I mean, the render is set quite professionally and also look how it organized the project, like and prepared, like everything, like put the reference that we use, it generated the videos. It all took like one hour. You can see all the renders that it created even on multiple stages. So you can track the progress only by here. And here's the final result. I mean, that is insane. As a 3D artist, I'm shocked and scared.

Not going to lie because only three months ago, I was making video testing Claude Opus 4 .7, 4 .8 and GPT -5 in Blender. And they were struggling to place objects with right rotation, you know, not to say about everything else. So like just placing ready objects around was a huge problem for them, both for Claude and GPT. And now it not only places them right, but also creates them from primitives like crazy. Some of you guys always ask about the mesh and this is like most reasonable use of polygons I ever seen in my life.

Look, everything is built from primitives as much as it can be. Like the whole scene is just 140 triangles, which is extremely low. It's like 70K faces. And look, every text here makes sense. No gibberish. But yeah, I mean, like it's insane. But let's take it even further. On X, I saw crazy examples, how someone do complex objects from blueprints. And I thought we also going to need to test it. So here's the car that I going to ask again, built from primitives as close as it can be.

And by the way, guys, if you still not follow me on X, this is best time to change it. You're going to be first to know any updates related to 3D AI because putting videos on YouTube takes much more time. Here I also recorded the process, not from very beginning. So it already created these tires. And again, like those are primitives. It was looking so good. So the workflow of GPT is the same. It starts with blocking and then upgrade and upgrade.

And I also asked to create this assembly animation, like from pieces around. So it puts all things together. And guys, you can see the results. It's absolutely mind blowing. And again, it's really sets the render really nice. And this assembly animation is something insane. I mean, of course, when you start looking into details, it's not exactly same card. There are a lot of discrepancy that you can find. I was not doing any behind the scene tweaking on top.

It's all was created in one prompt. I actually feel that GPT 6 works longer than Fable. But the interesting thing that I made so many tests already today and it is way less expensive that Fable. Like Fable 5 .1, you can do one task and all your limits disappear on Max Plan. Here I did more than 10 tests and I spent only 7 % of my weekly limits. I also use Max Plan on GPT because I feel like I'm going to switch to GPT for a while.

And I remember I saw this crazy showcases on X and I was thinking, maybe that's like few prompts, maybe something prepared. No, that's indeed how it does. And this particular case was created with Hicksfield MCP because it exposes a lot of 3D AI APIs. Like there's plenty of things like it generates a lot and it uses in a smart way, which is cool. Like Hicksfield MCP has pretty much all the APIs related to 3D, to animation and to image like this all image models in one place.

So just so you know how it's done, it's definitely not procedural. It generates some objects with 3D AI and place it in scene nicely. And speaking of that, next test will be related a little bit to 3D blocking and making video of it. And it actually takes it to another level, guys. So I have this idea like there will be dynamic race because I saw lots of these examples and like it's too dynamic. Like usually 3D AI is not capable to do that, because always there is someone behind the scene actually adjusting all these things.

So I actually use this plugin with GPT and man, it actually created the complete track with a lot of objects. I mean, as a play blast. So the play blast is used later for the video to generate. And you can see it on the screen. And it also set crazy dynamic camera animation. In this case, it was more than one prompt just to make sure actually getting there. But 95 % of that was made by AI. And here what you see is a play blast.

And the first thing you can notice is like how dynamic this scene is. Yeah, I mean, it's very simple primitives and stuff, but that's the whole point. Now, here's what is going to happen. It would take this play blast and put in C Dense 2 .5 and Hicksville and just look at this result. It's absolutely crazy. I mean, AI filmmaking is real and you can actually direct it. I mean, you can control it like, look, this was done in like few hours.

I mean, directing the scene, play blast, the environment and then rendering at once with C Dense 2 .5. So what I was talking about, not only 3D artists are doomed, but filmmakers as well. But honestly, AI video is on the stage already for a while, and it's quite, but I mean, there's not much things that are really enjoyable, right? Or image generation is absolutely on the highest level right now, but still there's few good ideas.

So true artists will remain anyway. At least I believe in that. If you guys disagree, let's discuss that in the comments. Here's one more example setting the crazy camera animation. And the idea was like really funny. Like we have a bee and you know, like in some movie openings, there is something happening like small insects going to like two holes, etc. And I just want to test how good it will handle like this motions, like follow up motions, the dynamics, etc.

And yeah, that was actually really cool. And again, combining that with C Dense 2 .5, like just insane. And the amount of cases where you can use that is absolutely insane. I'm going to be making another video testing GPT -6 in game engines, such as Unity Unreal Engine. So subscribe to not miss that one. But I also tested how good it will handle the logics, etc. and ask it, for example, to create me an add -on for Blender that will create LC Soins.

I mean, you all know this guy and I want to create LED Soins configurator right in Blender and just look what it created for me. I was like expecting maybe like this much parameters, right? But look, it created me a whole bunch of stuff like for, I don't know, I can do really anything. So here's how it works. Like there are three presets right now. Like you can say, set this preset. OK, like this preset. There's also this preset and this preset.

There are plenty of settings that you can tweak. You can even upload your own logo. Oh, this one, I don't think work. OK, let's do this. You can preview it as LED. And I mean, like I didn't even check all the parameters. I mean, that looks so impressive. This one actually looked cool. And I also spot in this preset, it was actually calculating connectors. Look how it connects through the glass. I mean, that's just insane.

I mean, wow. Just think about what crazy things you can do with it. While I was working on Blender test, my friend actually took it a little bit further and created like 3G application to create sprites for games. It took like one hour and just look how crazy it can be like this. It's absolutely next level of understanding 3D. I mean, this case, again, shows really well that it understands how to set the logic around it to get some reasonable results.

So here the idea is that you load a 3D object and it creates you a sequence of sprites. Just like that. Usually you actually have to go and pay for a subscription on some website to get access to this feature. And now whenever you need it, you just wipe code it like in one hour and tune it as you need or how about creating web application that will show us prediction on climate change around the earth. And all this data GPT found from some open sources, like from open APIs, etc.

And just look at this. I mean, I don't know. I've been showing like tasks like this with Fable 5 also. And I don't know. It doesn't feel like a small jump, like small improvement. It's like significant jump. Anyway, application in games. I'm doing crazy tests right now. Can I show them to you very, very soon? Don't forget to follow an X to see them first and subscribe to see them on YouTube very soon. That was Steph.

Like and subscribe and see you the next one.