Packing greyscale textures in color maps
polycounter lvl 12
I wanted to bring this up as a discussion because I heard of this method a while back I can't remember where it was mentioned, but they called them "loaded textures", but this doesn't seem to be a popular method of handling greyscale textures.
Why not take 3 grey textures and put them into the RGB channels of a color map, then reference them as needed. Perhaps my understanding of texture calls is flawed, but doesn't this mean you're saving quite a bit at run-time because of less calls? Regardless, it saves a bit of space small scale but can save megabytes in the long run. Couldn't this allow for a more content on the artist's end because of the additional space? When it's not necessary to have color info built into the texture then is this not a viable solution?
I recently used this in a couple props just for fun and it was great. I was able to put all the information that I needed for my model in one texture (+normal map and a color multiplied in)
This seems like a great idea for props and such where colors are added in later and don't need as much fidelity and certainly an area where hand-helds could benefit, but I haven't seen this in practice.
What are your thoughts, PC? Am I just misunderstanding this subject?
Why not take 3 grey textures and put them into the RGB channels of a color map, then reference them as needed. Perhaps my understanding of texture calls is flawed, but doesn't this mean you're saving quite a bit at run-time because of less calls? Regardless, it saves a bit of space small scale but can save megabytes in the long run. Couldn't this allow for a more content on the artist's end because of the additional space? When it's not necessary to have color info built into the texture then is this not a viable solution?
I recently used this in a couple props just for fun and it was great. I was able to put all the information that I needed for my model in one texture (+normal map and a color multiplied in)
This seems like a great idea for props and such where colors are added in later and don't need as much fidelity and certainly an area where hand-helds could benefit, but I haven't seen this in practice.
What are your thoughts, PC? Am I just misunderstanding this subject?
Replies
The process would be 1 big RGB luminosity texture + 3 little RGB color textures/vertex color ?
It adds complexity to the pipeline and shader though and 1 additional draw call in case of color textures?
or u were just talkin about "regular" greyscale maps (gloss, specular...) >> In this case, ofc it is useful.
Do you mean you mixed your diffuse and normal map together, or did I read that wrong?
Great for packing all your masks together.
I created a shader so the Gloss/Reflectivity/Emmissive were all thrown into one RGB texture.
It's something that's easily possible with node-based shader setups.
JacqueChoi's example is right on the money. Similarly I did a spec/emmissive/diffuse in one texture. All I needed was orange to be multiplied into the emmissive so I was able to get a nice looking candle done with one texture and saved a ton of space.
Also if there's a lot of overlap between all three channels it will also suffer from more compression artifacts
~mask - gloss <pure red layer set to multiply> [layers composing gloss map] - reflection <pure blue layer set to multiply> [layers composing reflection map] - dirt <pure green layer set to multiply> [layers composing dirt map] ~normal ~spec ~diffusethe channels compress differently, and DXT compression favors red over blue over green, as far as artifacting, so this is something to consider when packing these in. Green, in particular, is really, really shitty, and shouldn't be used for anything that requires fine material definition.
e:vv I meant DXT, specifically DXT1. AFAIK the DX normal format is 2-color only, but I'm not positive. I'm pretty sure a tech artist could weigh in better, I'm rusty as fuck on my DXT compression formats. DXT5 takes more memory if you're using alpha, as much as two DXT1 maps, and afaik still has shitty compression on green, so you effectively get 3 maps plus a shitty green for the price of 2 maps, instead of getting 2 maps plus a shitty green for the price of one map with DXT 1 (this is in terms of memory, not draw calls).
I always thought that was mostly a DXT1 (especially DXTnm, I think there's even one that gives you the option to completely pillage the green channel) thing, if you find a need for it, perhaps you can use another channel (RGBA) and use DXT5.
As for using the different layers for different things, I remember being shown a model of a Final Fantasy 13 character, the girl with long pink hair. Her hair had used 3 or 4 channels, for gloss, spec, weight and bendyness or something. How it calculated those in-game, I've no idea.
Um... apparently DXTnm sticks the red channel in the alpha, keeps the green channel, and wipes the red and blue channels with constant colours. (source)
Why does it stick non-constant information in the green channel if it's "really, really shitty"?
http://www.polycount.com/forum/showthread.php?t=87372