I have definitely got the google collab working now! yay! It didn’t work this morn but after fiddling for aaages it started working. Seemed like signing in on a new google account worked.
So first you need images with the background removed so this site is perfect:
Then you upload them to the collab page:
Then you can download the file as:
- an image
- a video
- an obj file (CAD file)
It works really well with full body images, I assume this is what it was trained on. But this means if you use half bodies or close ups of heads then the forms it comes up with come out distorted. This means there are two types of forms I could create.
1 . attempting to be as realistic as possible (but some distortions naturally pop up):
2 . Attempting to make crazy creepy distorted forms:
So what have I made using this?
First I made full scale models of big-tech CEOs, so I can make slightly creepy faces/heads to add to busts or figures to use in some way:
Secondly I made creepier versions of these:
Then I made models from the AI generated sculpture heads I made which were very very distorted, some completely unusable:
Then I decided to try to make some less distorted versions of these. I tested out whether a larger sized image of a sculpture would create less distorted images. It worked reasonably well:
So I photoshopped the AI generated heads onto Michelangelo’s David :’) This worked surprisingly well:
Here is what they look like when passed through the AI: