My Full AI Workspace - A Year in the Making, Now Yours for Free transcript

This week we crossed the line no one was expecting. Tripo released a smart UV finally solving UV unwrapping and it looks crazy. MeshE announces version 7 .1 introducing 4K mesh resolution with up to insane 80 million polygons before simplification. And GPT -6 Astra still surprises with how much it can do in 3D and animations. All of that and more today with Philip. Next big news is that MeshE 7 .1 was just released. And the great thing is not only it released overall improvements but they also announced as the first 3D AIR generator like 4K detail.

And this is really cool to see because you can see here that the typical render resolution with all those 3D AIR generators was so far like having a 2K render resolution over here which has a good detail richness already but 4K brings it to a new level of course. And this is coming with MeshE 7 .1. You can see for example here a detailed side by side comparison where you see how much detail Ultra 4K is actually bringing compared to the default one which is here probably 1K and you can see it here side by side.

This is of course especially useful for 3D printing but it can be also great for creating displacements or a high Skyped geometry that you can then take further into in your 3D software. So really exciting to see that changes. You can also read about it here in The blog what they exactly worked on and how much detail richness you actually get. And 4K is already on a really high standard and it's of course great because with more detail and preciseness you then of course obviously also have more high quality meshes that you can use.

You can also see here that MeshE scores in compared to other 3D generators here the highest detail richness when it comes to that. So they really put focus on the detail itself that you can generate now and MeshE is also claiming here on the blog that they get more detail without more noise which was also an issue that when you upscale the resolution that you then also receive more noise with you know your generated meshes.

But here they improved that quite well and you will get higher details here without introducing new noise which is of course great. I also gave this a quick test myself you know on the website since MeshE 7 .1 is now available for everyone to use. So I selected here the MeshE 7 .1 flagship resolution with Ultra 4K and it costed me like 35 credits for this generation itself. And you can see that the detail is really stunning here and impressive I would say.

When it comes to that and you will see overall like in the scales of the dragon but also from you know what we get here on the ground we get quite a good. Detail richness here overall so yeah really cool to see that though. Probably the coolest thing that you already can touch is Tripo Smart UV, and along with that they also updated Tripo P2, which is now public instead of beta, and there are cool stuff like four generation at once for same price that they added.

But let's talk about the UV. This is first AI UV unwrap that really works. I tested it on multiple cases, and I'm still shocked. First of all, again, like it's extremely fast. I would say it's like quite cheap because like for 20 credits you get like up to four variations, and you can basically choose which did the better job. And like almost all the cases that I tested, it was like I mean 99 % there. It works for organic, it works for hard surfaces.

I mean for hard surfaces it works so well that I don't know like I was like just left speechless. I already did complete breakdown on my channel, and it is really worth trying. Closest alternative is Hunyuan 3D Studio, but compared to this UV, it's absolutely different level. Hunyuan was extremely unstable, and it was taking like up to 10 minutes. Not to say it's still in beta for already like six months or more. And as I said, Tripo also released P2, so the quality of low poly meshes also improved nicely.

Objectively, Tripo becomes top one for game development. Back to Philipp. Next up we then also have WorldSkype. This is a new tool that will generate basically objects from your 3D Gaussian splats. And this is especially useful of course for interiors but also for different locations where you shoot on set for example or you take your camera with you, create a Gaussian splat you know or generate this Gaussian splat and then you can extract objects from it and then rebuild your scene.

And this is really great though because you can see that you know it's able to extract like a lot of objects from your scene. So you will basically have this presentation where you can you know have your Gaussian splat and then you can extract like couches tables and so on from your Gaussian splat and then use that. They also state here that it's complete and watertight so you don't have any issues then you know when using it in your 3D software.

They also put here some examples which I found really interesting. So here we have the office example that was reconstructed. We have 678 objects here in the office. And you see that the detail we get here in those close -ups is also really not that bad I would say and it's all also extracted per object. So you can see when I go here to the explode view you will see how it's kind of scaling here. And same thing we have you know with other examples here so. also a city example as well even though here it's showing the interior.

Don't ask me why but anyways I think you get the idea and the hang of it you have like this use case where you can easily extract objects now from your Gaussian splats and then rearrange them or place them in your 3D shot and you know work from there. So really cool stuff to see. Here we also have them by the way in Unreal Engine rendered. So that's of course also really cool so you can see what cameras they placed like the point load itself and so on.

Yeah some really cool examples we will link you as always everything down below and then you can check it out yourself. You will also see that this is based on trellis 2 and then used to create those objects. So really cool to see that also integrated here and they're using a mix of Pixar for the trellis 2 and Dino V3 you know to create those shots and use them with the new technique. So really cool to see that and yeah it was released like just a few days ago so definitely make sure to check it out and let me know your thoughts down below in the comments.

It's been already more than a week since GPT -6 Astra released and we see more and more amazing cases. One of the things that was rocking last week is that GPT -6 Actually can do the animations and I was so impressed that I also made a video on my channel showing like how I animated a creature with multiple legs a scorpion. And guys I was same surprised as you when you saw that video because I did a test and it works.

The difference between fabul 5 .1 and GPT -6 on such task is huge and we also see more examples on X how people create animations for humanoids etc. I personally test it in various way and can confirm that works but I really recommend you to use it in a smart way providing with some Api's combining with the right libraries etc. That way you're gonna get some quality results. And tomorrow most likely I'm gonna drop a video about my AI workspace and how I make GPT -6 work at max.

Next up we have Kai Ninja and really it's a really interesting name though I have to say but it does a really great job in you know taking Trellis 2 as a base and here's another example where actually you know people are taking Trellis 2 now and build on top of that which I think is really exciting though to see how much at the base of Trellis 2 is pushing people to take it now to the next step and you can see that you know Kai Ninja extends basically our Trellis 2 to generate separate part meshes from only one image.

And this is really cool. Like you can see for example here, I mean this is a really basic example of a chair. But you will see that you know it extracts all parts correctly and we then have them directly in in our shot. And you know it consists of 15 split parts that are all you know like for themselves. And this helps of course for splitting your objects because of course we stumble always about the issue that we have one combined mesh when we work with trellis 2 for example.

So this is really cool to see you know that they build on top of that and that you can use this then to you know split your objects. Here we also have it with you know like for example this car which is a bit more complex. So we have then here the weird split as well as the same here for this microwave where we have then the doors and the handle itself separate as well as on this one here. So yeah really really handy and useful stuff I think.

And definitely something to look out for. You can see that the model itself is not released yet but it hopefully will come soon. And we are not sure yet about the license unfortunately if they will monetize it or not. But this is for notice statues I will definitely check it out once it's available to see what we can do with. Kai. Ninja you might not notice but Hunyuan Tencent they actually dropped another 3d AI model but.

Since it's not available on Hunyuan website or there was no loud release many actually didn't even notice and I'm talking. About Miura 3d I hope I spell it right I actually did a post a while ago showing that like Tencent cooking some new model and. I was not able to say the name but now it's and it's released for their new platform called Miura Tencent design so it looks. Like casual 3d platform with some nodes etc nothing really special and yeah and like agent I do not.

Advertise it I'm not really used it but here you can see that there is like Miura 3d beta available and same model. Is also available in Tencent cloud API called Miura 3d I did some tests. And I don't really see big improvements in quality yes it sort of does better textures for mesh I don't. See like big difference and compared to any other frontier model such as Tripo high 3d or Meshy. It doesn't really show it and to be honest that's really sad I was really hoping that Tencent cooking something big and.

Show us Hunyuan 3d version 4 that will outstand everyone else but it seems it's not happening soon and then. Last but not least we have snap 3d here also just a really quick example it's kind of doing the same. Stuff as Kai Ninja. That we can take it and it will assemble all scene and then you know split parts of it so we have then you know the 3d parts from a single image. here Here's also technique behind that you know that it's taking our geometry and then you know doing the 3d part generation.

You can also see here some results just stumbled upon that and for it's might worth sharing still because you know it's also using this really special technique to split your models. So yeah definitely interesting to see. You can give it a read if you want here also to read their paper to see in case you're interested about that how they you know work on this technique and what it can do for you. And they also released this video here on Youtube where you can also see basically you know where they take it to the image and then they use it in the image to 3d here it was only with Hunyuan 3d 2 .1 but then they use their technique you know to split their models even building your connectors for example that you can use for 3d printing which is also interesting though.

And yeah really cool to see this is also being worked on because obviously this is great to use and yeah very handy to have split parts in the end. And one more interesting news from my side is about Hk Tex paper which stands for heat. kernel. Textures and these guys basically introducing new concepts of how we can. Store texture data without needing a UV so basically it will define textures directly on this surface and.

You don't really need to unwrap models etc i mean it goes on top it's quite big paper to read and you can also find. Some examples how it's gonna look and to be honest like right now there is not not much big difference like basically. They say that we can achieve same results with texturing but we don't really need UV and that's actually cool. Concept especially in view how fast 3d AI evolves yeah i mean the texturing that way just not gonna make.

Sense so we're just gonna go work on surface and preserve that information on top of the surface with smarter better ways. Of optimization because again uve etc all that things is like first generation of 3d like i mean many things. Gonna change very interesting research let's see if it will take any place in game developments anytime soon so we really hope. You have enjoyed this video as always make sure to subscribe to chefhans channel as well as my channel and see you in the next one.