General Category > General CG Discussion
AI what are peoples thoughts ? I think we are in for big change.
3DInteriors:
They can facilitate our tasks and shorten the working hours.
Jpjapers:
I had wondered about the way AI could change rendering specifically and i wondered if we could end up with specialised models (the ai kind) that are hyper specialised packages that know how to render a specific object or surface style very very well. It could understand placement data from the 3d scene by using very basic placeholder objects for layout. Understands where the camera is, and lights are. And then uses lots and lots of specific stable diffusion models to generate the final image knitting it together mask by mask like an ID map.
Jens:
I'm currently using it as a tool for archviz in a couple of scenarios:
1. place 3D people in the final render and then using stable diff plugin in photoshop to generate "real" people in place of the 3D ones. Works very well and is wonderful in cases where client has very specific demands i.e. dockside workers in various genders, ethnicities wearing life vests and workwear (good luck finding cutouts or even enough images like that where posture, light etc is suited to then start cutting out). Also fine for just replacing with "normal" people in renders to avoid the 3D people look.
2. txt2image or img2img to produce references to discuss material/architecture on objects that are still just volumes/white boxes). Way faster than having to model and texture until client actually knows what they want.
3. for projects/proposals where the brief is more open the above combined with the inpaint function of SD is great to generate variations to the environment (in this example it was a big office complex and a new park was to be put in front, but the park hadn't been designed yet. I did a rough landscape and placed some trees/elements and light for a nice composition and could then generate more lush/detailed variations of the environment via inpaint to discuss which way we should go design wise before I start modelling the "real thing". If I had more ram on my GPU to make it higher res, I could have just used the generated results from SD (even taking some variations and photobash the best elements into each other). Now I might try to upscale and then rework the image by 512x512px frames and see if it can add more details.
TLDR: it is and will be an amazing tool for us as archviz artist. It might someday be the new "enscape" where clients themselves can get pretty good images with a couple of clicks/prompts, but in the meantime I still think there's a place for even higher end or more creative images from us :)
mferster:
--- Quote from: Jens on 2023-06-21, 13:09:35 ---I'm currently using it as a tool for archviz in a couple of scenarios:
1. place 3D people in the final render and then using stable diff plugin in photoshop to generate "real" people in place of the 3D ones. Works very well and is wonderful in cases where client has very specific demands i.e. dockside workers in various genders, ethnicities wearing life vests and workwear (good luck finding cutouts or even enough images like that where posture, light etc is suited to then start cutting out). Also fine for just replacing with "normal" people in renders to avoid the 3D people look.
--- End quote ---
Neat! can you show an example of before and after? Even cropped images are fine if they aren't public
James Vella:
I seen one recently on FB from an artist named Matt Hallett. Apparently using the technique 'in-painting'. Think you can find his page under: Matt Hallett Visual
This was his before Image
This was his AI replacement
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version