Luma AI has introduced Imagine 3D, a text-to-3D AI generator. The current results still have the quality of a mediocre 3D scan, but that will undoubtedly be refined.
Demo video:
I wonder what this will mean to stock 3D sites like Turbo Squid, CG Trader and Sketchfab. If I was in charge there, I’d immediately invest in similar AI tech, and offer a subscription or pay-per-generated-model option to generate your own 3D models.
1 Like
Imagine fingers?
I still have not seen great AI results for fingers at all.
This looks very useful though…
2 Likes
Yeah, correct fingers and eyes are often a bottleneck for current-gen AI. But it will all be refined the next few years.
2 Likes
I have often wondered why? as its not like theres any shortage of hand or feet images.
Even if they wrote a cornercase for them?
Yeah, I guess hand + finger gestures are too complex to interpret correctly by the current AI networks. The ‘guessing’ doesn’t suffice, more actual understanding of how hands and fingers work is necessary.
Don’t forget that even seasoned artists still have trouble with two things: hands and eyes. 
2 Likes
I am slightly more excited about NERFS at the moment, very interesting watching that tech
2 Likes
I wonder what the next iterations of these text-to-3D engines will enable. Before you know it you’ll be able to enter a prompt like “Cute character by Pixar, fully rigged, Blender armature.”
2 Likes
Yes maybe nerf assisted if they can extract models from it, which I believe is not yet possible…
2 Likes
Things are moving so fast now, even I can’t believe it sometimes…
What a time to be alive!
2 Likes
When it comes to 3d models, it is clearly more challenging than images. First and foremost, there is way less data for this than for images.
They most likely have to go through some heavy preprocessing steps to get some sort of standardized mesh and texture format.
It is very impressive to see that this works at all!
Extracting models from nerfs works, but you get a horrible topology.
3 Likes
I will be out of business once ai can create portraits/3d models in a similar quality as i can. Well it’s already happening for still images. I don’t feel much love for this kind of tech to be honest.
This threatens my financial existence. 
Software devs and companies are the big winners here, as always.
Thanks for sharing though.
4 Likes
Co-pilot is a thing, so software devs will be automated too at some point…
2 Likes
Yeah, we’re living in weird times. I guess we’re on the verge of a radical change of the creative landscape. In my humble opinion a universal basic income is the way to go, as more and more professions will become automated.
7 Likes
Yep, i agree with you.
3 Likes
There are also a lot more variables to consider. It is one thing to generate a model, it is quite another to generate a model with consistent attributes that is not a polygon soup. Model AI that can generate animation-ready meshes (which are already rigged) will be even more difficult.
At the least though, there could be a minimal work requirement attached so it feels like it is being earned, as well as dissuade people from staying home and playing video games all day (even if it is community service jobs like tree trimming and litter cleanup). A lot of jobs will be automated by AI, but everything being automated to the point of what is found in Star Trek is still a long way off.
2 Likes