New Local 3D AI Generator Beats Trellis2, Claude Blender Update, Unity AI & More! 3D AI News #11

Añadido: 2026-05-21

[00:00:00]We have a new 3D AI generator that beats Treasurus 2. But there's a catch. A new Unity AI implementation, an incredible claw to Blender update, and much more in today's news episode with Stefer.

[00:00:12]>> And guys, Pixel is already added to Top 3DI. So, if you want to test it out, go and check it there.

[00:00:18]>> All right guys, hold on because this might be the biggest news we have for the last several weeks in 3D AI. And this is Pixel 3D which was released by Tencent ARC lab from the Chingua University. I hope I spel that correctly. But basically, as you can see, it's a pixel aligned 3D generation for images, which was by the way also accepted at Sigra for the 2026 edition.

[00:00:41]And the special thing about it is that it's built on the trellis to backbone.

[00:00:45]So definitely acts more as an evolution rather than just a completely separate tool. And what you will see on their project page is basically that they have here some examples for what pixel 3D is basically doing and how it separates from any other 3D AI generator. It uses pixel back projection instead of generating canonical space what for example Trellis 2 is doing. So every pixel in your input image maps directly to a point in 3D which achieves near reconstruction level fidelity. There even comparisons like you see here in the beginning with this pizza even more accurate in some cases than Hanuan 3D3.1 which is of course also a big improvement. They also have a free hugging face demo that you can try out here where you can basically upload your own reference image and then try it yourself. Just be aware that here the generations are quite limited and you might only have one or two runs. So just keep that in mind before starting. also even have now a Confi Pixel 3D version already, which is also great and directly blends in with the Confi easy install that I showed you in previous videos where you can basically install it here and then set it up because it's built here on PyTorch 2.8 with CUDA 12.8 as well. But guys, there is a thing we have to talk about and this is the license with this one and unfortunately Pixel 3D is only for academic use. So there's no commercial or production use allowed basically. And it even gets worse because for the European Union, it's stating here that Pixel 3D is not intended for use within the European Union, which basically means that all of us that are based in the EU are not even allowed to run this model, for example, locally. And this is of course a big downside. The only way I found so far to run it is to run it on file.ai because here we have a commercial license basically to run the model. And you can check it out here. Of course, this will cost you then some money. So, it's like 30 cents per generation, but at least you can test the tool here and you will see that you get quite great and accurate results. I also tested this on different models as you can see here.

[00:02:47]Also, the sculpture head basically where we have some holes for some reason. I have to check out why that is, but basically this is what we get. And very special thing what I found in the examples which I found was really impressive is Pixar 3D seems to be able to generate cloth accurately as you can see here. So Pixel 3D is generally impressive but you can't really use it at the moment. So Trellis 2 still remains my biggest recommendation for further use for you guys. In case you would like to be informed upfront about it, I have my 3D AI newsletter where I covered this already on Friday on an exclusive newsletter issue where I covered everything in detail and you will be there the first to know about the latest 3D AI generators. And now having said that, back to Cheva. Okay.

[00:03:30]And we finally have Unity AI beta released. Well, this is amazing. And there's like several things actually inside. It's not just one AI inside Unity. No, it's actually bunch of tools.

[00:03:42]So first of all they introduced a Gentic in project assistance and you'll be actually having some credits inside and yeah I mean pretty much standard thing like as they say the subscription for it cost starting $10 per month which is I guess fine but the interesting thing is that not only you can use Unity AI but you can actually bring any other third-party agents directly through so-called AI gateway and together with that they also introduced MCP server. I mean, and that's basically means that you can connect cursor or cloud code using that MCP to an agent that you used to to an environment that you already used to. And I actually can't wait to see like cloud code doing something in Unity. And we already see on X some example people start creating games in one two days. It's not like AAA games, but it's still impressive. So if you're using Unity, you cannot miss this out.

[00:04:38]This is really really cool one. Next up, we have map to word which is also a really interesting paper as you can see because it's basically able to create those kind of city planes which I think is really cool and you can use here text to generate your world from it. So you can for example input here a forest with tall green trees and you will see that we get then this kind of visual representation which is of course really cool to see and we also have turntables here to explore where we see that we get quite some decent details. I would say this is the first time actually I see where we can then basically you know have scene level scale control. So we can also have like you know city like scans as you can see we also have like those castles and towers here and so really cool in case you also building a city or something like this and this is a quick way to then iterate on that. So pretty cool and as you can see the code will come soon. So this could be a very exciting opportunity to use. Another interesting thing that popped up on my radar was Nova 3D from a guy named Rare Sense. So it's actually a library that's building up shapes in a different way.

[00:05:44]It's not creating like 3D models but actually building shapes in functional way making them editable and it builds from primitive thing is that instead of like tripo rodin or meshy inside you actually put your API key to open AAI cla or gemini. So from the demo you can see actually what it is about and this is genuinely impressive and can be very very useful. I would say it's in early stage but it's really impressive. I probably gonna give it a test as well.

[00:06:15]Next up, we have a big claw to Blender update. And this is really impressive because we also saw the claw code implementation into Blender. And now this just came out. And it really blew my mind when I saw it the first time because World Labs developed like a workflow where you can basically take an image and you will then use a claw skill to then generate a 3D gshion spreading environment and it will also create all the 3D meshes and interactive physic objects for you from this one image. So when you see here this preview video, it will basically you know block out all the assets that we have on the right side image. As you can see, the environment will be then discion splitting environment basically and it will be combined with a 3D AR models basically that we need and this is all possible with cloth skill basically and the combination of world labs and a 3D air generator. But this is of course really cool for world building and scene building. So really cool to see that because it even understands physics. So definitely really exciting stuff and I will also link you this down below for you to play around with it. And another update for the cuscador version 2026.1 brings autoosing physics and we do have quadruplet workflows meaning that you can animate horses, cats, wolves. I mean basically quadruple animals and this is quite impressive. If you missed out I made a full tutorial about cuscador and it's quite nice tool to be honest. And then next up we have Modly. And this is also really cool because it's basically also a desktop type of app to generate 3D models from images using local AI and it's a bit like the 3D gen studio which we showed you last week but it uses here a base of Hanuan 3D 2.0 and 2.1 and you can then turn any photo into a 3D model using open source AI models and you can see this is was created by Lightning Pixel. So definitely cool to see this and you will see here that it's using Hanuon 3D mini tripo SG as well as Trellis to GGUF even which is of course really cool. So you can even run it low VRAM and then have your own 3D local generator which is of course really cool. And lastly we do have a sneak peek of upcoming feature from realusion which is called CC rap and this is what we actually need. You will will be able to drop any AI generated character as a full body and set up some points and it will be able to animate it overall which is actually amazing from the demo. I'm really hyped. So, let's see if that will work as smooth as they promise. And then last but not least, we have a new Comfy wrapper by Terry Ga. And this one is really cool because it's based on Comfy Mesh to motion which is a really impressive tool I have to say for rigging your characters and animating them. So you will see this here directly in Confy where you can basically have this interactive editor where you can create cameras but you can also you know import directly animations inside of it remapping them and so on which is really cool I think. As you can see we can animate the fox and stuff like this. You don't even have dependencies that you need to match. So really easy installation wise as well. And uh yeah, I think this is also really cool news to see this. As you can see here, some also some examples where you can explore the node and you have all those animations.

[00:09:41]So definitely something worth to check out and let me know down below in case you would like to see a video about this as well. That was it guys. Let us know what you think about it in the comments.

[00:09:50]Like and subscribe and see you in the next