chippwalters
patreon
AI and Blender: use case
🕑 Added 2022-06-25 15:17:57 +0000 UTCSo for a while now, I have been thinking about how to combine these two. In particular, how to use AI to help Blender create better renders.
So, here's an example I'd like to share. The first image was created by GREEZYBEAR and uses his Sci-Fi in a Box kit Ops KPACK. I then used disco diffusion to create a number of variations. And, I used an online AI upscaler to resize the AI render. https://icons8.com/upscaler
I'm still playing around with different settings to see what works best, but I thought this was an interesting use case to share.
Comments
Chipp Walters
Yes, all very good points. I talk about some of this a bit at: https://discord.com/channels/703466172741582888/986067724864192532/990439137620394064 (If you can't login, then go to http://cw1.me/discord )
Marco Chacon
I think if you had an AI trained with 3d shapes with descriptions you could use the same basic method to have the AI generate shapes based on being trained on a large amount of sample data. I wonder if, like, Sketchfab's library would be big enough.
Alejandro Tejada
All these developments are very exciting! It feels like we are in the dawn of a new era in the production of computer graphics.
Justin Jacobs
Nice and subtle.. I wonder if the last 2 images could have been done with kit ops and k-packs just as easily though. Or perhaps, use Ai to generate new objects for k-packs or at least come up with some ideas. Kinda reverse engineer the Ai? Now that Ai generated art is here, it will only get better, and wonder if someday modeling something will become the “slow way”, when you have an Ai that can create a 3D mesh from a simple drawing or series of drawings and will generate the 3D model, place textures etc. kinda like photo scanning. Then the artist can light the scene, or by then Ai will light the scene with example images of lighting from 2-3 references. It’s still artwork and it still technically created. And I highly doubt that if you have 100 people input the same keywords and use the same exact settings that the output images will be the same. Similar perhaps, but never identical. Thus making each piece unique.
