🧵 Generating faces?
Anonymous at Sun, 2 Oct 2022 09:13:19 UTC No. 920579
How do you people tackle the issue of making a face? I mean, making it from scratch may be the best option in some cases but then you will have to remember your custom topology every time, not to mention the whole material stuff, rigging, etc.
Face generators like FaceGen and Face Builder are hella expensive, there seems to be a web application which can give you something to work with but that one means you have to upload pictures with varying degrees of failure, MakeHuman's sliders may be of some help but sometimes they are quite limited in some areas.
So, any corporate secrets you wanna share for a more efficient way of creating faces from pictures?
Anonymous at Sun, 2 Oct 2022 15:30:54 UTC No. 920655
If you're going to use the same topology all the time just sculpt the topology into the shape of the new face.
Anonymous at Sun, 2 Oct 2022 18:12:55 UTC No. 920692
>>920655
Technically speaking that's the optimal solution but sometimes is not easy to do it consistently every time, it also means I need to keep an eye on head proportions too.
Anonymous at Sun, 2 Oct 2022 20:24:29 UTC No. 920708
Making a likeness of someone I take my head topology that I've iterated upon over the course of many years as a starting point
and put a 'PureRef' canvas containing numerous images of the person I wanna capture from a multitude of angles.
I then remodel my face to fit it as closely as I possibly can, then I walk away from the computer for a while.
When I regain fresh eyes I make a copy of my head and start making corrections.
I then morph between any corrections I made and the 'base', often times what sits at 70-90% of my re-edit and the base
is a closer overall match to what I actually see in the reference and I iterate like this til I can no longer spot any differences.
Process takes a couple of sessions over 2-3 days and I work on other things in-between sessions as I've found I grow blind for detail
if I keep staring on the same thing for too long and start getting locked in to my preconceptions what the surface looks like rather than what it actually looks like.
Anonymous at Sun, 2 Oct 2022 21:47:57 UTC No. 920718
yo just buy a head scan off sketchfab, they aren't expensive, and then use it as a mold to create a realistic base mesh.
then realise that humans are so finely tuned to perceiving human faces that its literally a case of shifting literal MM's here and there to create a completely new individual, xyz axis.
look at research papers, the average IPD range for example is like 7mm, 80% of people fall within a 7mm range. ludicrous.
so you only need one really detailed mesh to get you anything, fuck the wrinkles and pores, thats superficial shit.
shape, shape is not vairied, shape is uniform its the proportion that changes the shape, like stretching an elastic band, its not fundamentally different.
once you've learnt the average shape you just pull it in,out,z, x, up, left, y and boom.
don't believe me? superimpose an composite average face upon another and see for yourself, when seperate they will look like completely different individuals but then you superimpose and boom, you realise something insane, they have nearly EXACTLY the same quantifiable dimensions + shape give or take tiny tiny little changes all over the face.
you only need one base mesh, you do not need slider shit and learn to sculpt.