Is there any good tool to convert 2d images into 3d models?

,

please follow the setup in the README.md file. this wil setup your wsl environment, python environment and other stuff. guessing this wil be your first local AI implementation. :wink: Good luck and fingers crossed!!.. Next big step will be your local image generator and the need for more Vram :wink:

1 Like

Thank you very much :), I will test it and show the results here in the topic when I create something.

But it’s not my first AI implementation ;). I already used (even in early development) comfyUI, stablediffusion, TTS models, trained my own voice models etc so I’m not a total newbie :slight_smile: . I’m new into generating 3d models, that’s where I’m totally new (and making games basically). Also I hope it will set up easy with fedora as host or I will just pass gpu into VM, but those will be my implementations. Thank you again for sharing your app! :smiley:

1 Like

Just behold that game-ready topology.
It’s almost million triangles.

6 Likes

Is blender really a viable option? Not sure if you tried it yourself, but the learning curve is MASSIVE. Unless you can devote months of your life trying to properly make models, that isn’t some random garbage, I think blender is a no go for beginners.

Take a look at this video to clearly see night and day difference, between an expert and a “beginner”

Oh wow, how did you do that? Did you use the sculpting tool?

I’m not sure what you mean I’m afraid. Blender is the main 3D software people suggest you learn if you want to create 3D stuff, and you don’t want to pay for Maya.

Or do you suggest just using AI to generate things instead of learning?

I’m just saying that I believe there are easier alternatives (Haven’t seen one though that isn’t Tinkercad), Blender for me is just too hard :sob:, and I feel like I am not the only one

I think something being hard is relative in this case. If you want to make realistic models, then of course, most 3D software will be very difficult. But if you start with making simple things, it’s actually really simple once you learn some of the basics. Then eventually if you want more complex shapes and models, you can progressively learn more and more.

3 Likes

I spent around 60 minutes in blender and made some shapes that were “close to” some letters so I think that the hardest part is know where to look for thing/keys, that you have in mind. Not particularly how to use blender, but where to look for something. In the long run, Blender sounds like much better option. In the short run, depends I would say.

2 Likes

Maybe you’re right, all of this gargon and features I was faced at was not a pleasant welcome (For example, I found over 10 color values in a single cube, and had no idea how to change it’s color), but I guess if you know what you are looking at, you probably can start building 3D designing experience

It’s the thing @redflare produced using “ai”.
I just imported it into Blender to check the topology and triangle count. It’s highly inadequate for game usage.

4 Likes

Yes. Go check out the classes on GameDev.tv. They take you from beginner to intermediate very quickly.

3 Likes

Not months … years. But it’s worth it.

3 Likes

Yep, Trellis is for HighPoly models. There are some tools available to use AI for retopo. But not 100% there yet, but close. Textures and UVs are a second problem. Take a look at the UVs of the generated model by @redflare . I predict, every face has it’s independent UV on the texture. There are some tools to generate better UVs with AI. You have to bake the textures.
An interesting comparison of AI and human artists on retopo:

I spent some time looking into doing this. It’s more work than I want to do right now. Because there’s a learning curve for me. But I could pull that material out and use it later. My graphics card had to think for a full 30 second for me to view the material.

My biggest problem was that I decimated it to 10% and it still had way too many triangles. When I decimated it to 1% it was missing huge chunks. I looked into a process of turning it into a low poly asset and it’s just learning a bunch of new things.

Just decimate is not the solution. You have to rebuild a nice low poly topology around the high poly mesh. And bake the normals and textures from the high poly to the low poly.

3 Likes

So I gathered.

Try Nanite instead in UE; it might do a better job than decimate in Blender.

You can’t really make a good production model without a human 3d artist. Re-topologizing an “ai” mesh would require almost the same time and skill as modeling it from scratch. So if your 3d skills are not up to par, “ai” likely won’t help you much. Especially for stuff that requires a lot of edge loop discipline, like skeletally rigged characters.

7 Likes

If you’re willing to spend the timing cleaning up an AI generated mesh with garbage topology for basic props, you may as well just grab a camera, go outside, and take some photos to use for photogrammetry. You’ll get better results and a bit of fresh air.

7 Likes