@Flix @gogoblender
Yeah,the harder part is where could I collect all those artistic portraits.
The only place I know is deviant art.(I'm by no means an artist.)
If you could lead me to places with more of those headshots, I might as
well train one from scratch.
The only other option I have is to scrap from google or use bing's image api,
though those images are often not of very high quality.
Current generative nets aren't perfect, so you could expect some really
funny generations, actually I roll my own architecture, so much of
the flaws of stylegan2 or stargan2 could be avoided.
I guess some portion of those generated images are actually useable, so Flix
could use them in his MODs without concerns for 'copy left'
infringement. They are more 'original' based on your own interpretation.
There are some techniques to give image depth and make it look
3D, yet again, it's not like they always work so well, but I guess some pictures might
look really good.
There some one shot motion transfering tricks being used constantly in deepfakes,
so we could animate the portrait a bit and make it smile or cry, blinking eyes and the alike.
I'm not really sure how much time I could actually spend on this project, so no release date
promise. Data cleaning is notoriously time consuming, but hey I think this is a cool initiative!!