Hunyuan3D. A High-Resolution 3D Assets Generator

B by bitacovir on 10 Feb 2025, edited 11 Feb 2025 # +12 -1 votes

Hi. I just found references of this new AI technology in a youtube video and I wanted to share it here.

Hunyuan3D-2. It is a High-Resolution 3D Assets Generator for video games using AI. You can create the 3D mesh of an object with a text or an image. The project has some kind of freeware license for community people. So, I decided to run a short workflow thinking how this tool can be useful for early stages of architectural design. Something to full the gap between the ideas and concepts and the digital/3D work.

First, I draw a concept of my building. Simply, it is made by hand, a pen and paper...

Second, I uploaded a photo to the Hunyuan3D-2 demo webpage. Just run the generator and the result is... Wow! It is a full 3D mesh!!

Then, I downloaded the textured mesh to my computer to import it in my BIM project. Of course, it is a mesh in a glb format. You can post process it with Meshlab for a different format of less dense mesh, if you want. Once you imported it to your project, you can move or scale it.

Here, my design in a IFC file in FreeCAD...

And here, the same geometry inserted in Blender with Bonsai (the main building is a IFC project, the addition is a obj mesh)...

Also, Hunyuan3D-2 has an add-on for Blender, so you don't need to use the demo web. What do you think?

https://github.com/Tencent/Hunyuan3D-2

V by viktor on 10 Feb 2025 #

Nice, but in case you would like to change your concept, you draw another hand sketch? Because with the messy topology it produces, isn't it easier and quicker to do the concept model right in Blender? It is interesting though, thx!

B by bitacovir on 10 Feb 2025, edited 10 Feb 2025 #

@viktor said:

Because with the messy topology it produces, isn't it easier and quicker to do the concept model right in Blender?

The point is this machine interpreted the 2D sketch and created a 3D mesh fully closed. It completed the design... and It was right. It wasn't me... Of course, the mesh is not ideal for editing. But now you can use your hundred of drawings, photos, videos captures, etc... from reality and bring them to the 3D model for show. You can take photos of objects around you, with your phone, and insert them in the model with a single click...

P by paullee on 11 Feb 2025 #

Looks great!

Any options to import the texture of the mesh in FreeCAD? On the other hand, IFCexport or NativeIFC/Strict Mode ifc + Blender/Bonsai import is of good quality? Thanks :)

J by John on 11 Feb 2025 # +8 -1 votes

I've been experimenting with the similar https://huggingface.co/spaces/JeffreyXiang/TRELLIS

Drop a photo of a building onto it and it outputs a surprisingly good mesh of the elevation and also a good guess at the unseen/unknown sides. Export as a .gltf and import into Blender. Scale appropriately and you have a good draft for exploration/communication. The .gltf is also textured.

outputs:

S by steverugi on 11 Feb 2025 #

thanks @bitacovir

The point is this machine interpreted the 2D sketch and created a 3D mesh fully closed. It completed the design... and It was right. It wasn't me... Of course, the mesh is not ideal for editing. But now you can use your hundred of drawings, photos, videos captures, etc... from reality and bring them to the 3D model for show. You can take photos of objects around you, with your phone, and insert them in the model with a single click...

I think in Sketchup (or Blender/Bonsai with some training) you can do massing or sketching literally in minutes by hand, I'm very passionate when it comes to technology and this AI is amazing but I wonder how practical it can be in a real life working environment

..or probably it just depends on how you are used to doing something I'd guess