Well, I guess it's no mystery to most of you that alpha transparency doesn't work with the Z-Buffer. While this seems to be a widley known issue, I seem to have missed this somewhat horrifying fact. So, what are some methods / formulas to have your particles (smoke, fire, or even just cut-out sprites) blend in with the scene so it LOOKS like it works with the Z-buffer? So far, I have transparent sprites that are carrying the background image, but as soon as they overlap another sprite... well... you know what happens ~_~ PS. oh oh, idea here, is there a way to have the Z-buffer / object verticies tell the program what order to draw them in?

Quote:Original post by TheKrust
Well, I guess it's no mystery to most of you that alpha transparency doesn't work with the Z-Buffer.

Actually, that's quite a mystery because alpha transparency and the Z buffer work just fine.

I think what you really mean to say is that alpha compositing is order dependent since transparency means you need to see the stuff behind the thing that is transparent. So if you draw the stuff in the wrong order, you don't end up with the stuff that is "on top" being... well, on top.

The problem isn't the Z buffer, but the order in which you draw things since Direct3D doesn't perform visibility globally like a raytracer, but locally using the Z buffer.

The standard approach is:

1. Draw all completely opaque geometry.
2. Don't draw any completely transparent geometry :-)
3. Sort partially transparent geometry by depth and draw in back to front order.

There are a few more tweaks you can do based on the transparency pattern in your geometry. If you have geometry that has opaque centers and transparent/fuzzy edges, you can use alpha test to reject the partially transparent parts and just draw the opaque parts (in any order). Then switch the test to reject opaque parts and draw the transparent parts, again sorted back to front.

To get pixel perfect accuracy, you will need to sort individual triangles back-to-front, but in reality noone does this, they just sort by the bounding box of the object.

Quote:Original post by legalize
3. Sort partially transparent geometry by depth and draw in back to front order.

Well, that's what my PS. part was about... but how would I go about doing that? Either by standard math or some kind of pipline function.

I understand that a Z-depth test would be very easy to do as in (out of my ass example):

if (z1> z2)
draw object1
else
draw object2

but that kind of thing would become very messy once view transformations come into the equasion (maybe world transformations, but I don't honestly know).

Quote:Original post by TheKrust

Quote:Original post by legalize
3. Sort partially transparent geometry by depth and draw in back to front order.

Well, that's what my PS. part was about... but how would I go about doing that?

Making a bounding box for each object.
Transform the bounding box by the current World*View matrix.
Sort the bounding boxes by their farthest away Z value.
Draw the objects in the order given by their bounding boxes.

Quote:Original post by TheKrust
oh oh, idea here, is there a way to have the Z-buffer / object verticies tell the program what order to draw them in

If you've got your geometry stored in a space partitioning tree, such as a BSP- or Octree, it's a trivial matter to traverse it back-to-front. This article explains it rather well. Specifically, the back-to-front traversal is in the section titled "Alright, I've Got A Tree. What Do I Do With It?".

Quote:Original post by Mr X
5. Enable Z-Buffer BUT DISABLE Z-Buffer-Write

That's awesome! I wouldn't have figured that. Yeah, it doesn't cover everything, but it sure helps for now. I'm looking into some other methods as posted above, not many making sense to me... but we're all here to learn.


EDIT: after putting that into the code, accuracy wise, that is wrong on sooooo many levels, but I guess if no one can tell, who cares?

Quote:Original post by legalize

Transform the bounding box by the current World*View matrix.
Sort the bounding boxes by their farthest away Z value.

so your saying I just multiply the world and view matrcies together and I should get the absolute Z value? Hmmm... past experience would suggest that's a little too easy... am I missing something or is that really it?