Consensus on bit depth in texture creation

CowKing

I'm currently working on a project of mine and curiosity has taken hold. I was wondering from other people in the industry, do you work exclusively in 8/bits, 16/bits, or 32/bits, or change between them depending the texture/step in your workflow/asset?

Personally, if it's a low-poly, small, and out of the way asset, I just do what's fastest for the workflow and use 8/bits so I don't need to fiddle with exporting. Although, for more high-profile assets, I use 16/bits because it can be super duper helpful when I have a lot of blending going on between layers. The only problem is when I'm working on a 2K or 4K res texture and the file size gets ginormous.

So, what are your thoughts?

passerby

for most modern work people use 24bit, 8 bits per channel, than if they need a alpha channel to store more information they will use 32bit.

edit:
are you talking about per channel or per image?
if your are talking about per channel sometimes i will bump things up to 16bit of working on heightmaps, and sometimes temporarily do a few things in the 16bit range if i have too much banding, but otherwise pretty much everything is done in 8bit per channel.

and 4k mapsize wtf, i would shy away from map sizes that big for game work, i ussualy go no larger than 2k at extreams.

gray

i paint everything in 16bit linear. even if i have to output to 8bit. when you are blending or using light strokes it makes a difference. there is no 32bit linear as far as i know and it is unnecessary. 16/32bit float is used for vector displacement maps normal etc.