#random
March 10, 2022
posts-images

AI Jigsaw

There is a huge hole in the market for decent non-Caucasian stock photography (eg. Indian). I think in the near future, AI morph technology should be able to fix this by morphing an existing library of photos into multiple races. Wouldn't it be rad to search for a set on say Unsplash and then have a browser plugin that then recognizes the faces and race and morphs them into a race of liking? Has anyone come across any early open projects in this space, would love to dig in.

Update: I came across Generated. They use a GAN (generative adversarial network), StyleGAN, Karras, and Nvidia to build the model. They trained it on studio photos by Icons8. Pretty awesome stuff.