I was studying TF2’s walls the other day
and discovered something weird. And by that, I don’t mean how you can shoot through, taunt
kill, or even stick to them– you can check out my other videos to learn about that–
here, I’m just talking about how they look. Take this brick wall found in
Granary and Well. At a first glance, it looks like a simple texture on
a flat wall because, well, it is, but there’s something more to it that isn’t
easily distinguishable and which we don’t see at all in Source 2. It has something called
an “ssbump” map, but before I even say what “ssbump” stands for, let’s see how this brick
texture was put together in the first place. On the technical side of things, there’s
a file called “brickwall001.vmt,” which is basically a text file that contains all of
the instructions for the material. At the top, it says “LightmappedGeneric,” which tells the
game to calculate the material’s lighting per pixel. Models, on the other hand,
like this barrel here in Gorge, have “VertexLitGeneric,” which calculates
the lighting for each vertex instead, and, the pixels in between are interpolated. This
isn’t as accurate as per-pixel lighting, which is why models like this barrel
look weird in high-contrast areas; you can literally see the interpolation happening
across the triangles that make up the mesh. Anyway, going back to the brick wall, the next
thing we see is a parameter called “$basetexture” alongside a directory and file name. The directory
is relative to the TF2 materials folder within the “tf2_textures_dir” pak file, which I can access
using a program called GCFScape. So doing that, I’m able to locate and open up that file,
which looks like this, and as you can see, it’s a texture file. In fact, the file name
extension “.vtf” literally stands for “Valve Texture Format.” It might seem silly considering
we typically use things like PNGs and JPGs, and I think Valve realized this
because you can now use them in Source 2. But since TF2 still runs
on Source 1, we’re stuck with VTFs. Moving onto the next line, we have another
parameter called “$bumpmap,” and a couple of lines farther, “$ssbump,” which is
where things start to get interesting. If I open the corresponding texture file,
it looks like this. What the heck is it? If you’re already familiar with how
materials work, you might think that this is a bump or normal map. If you’re not,
don’t worry, because this is neither. Instead, it’s an “ssbump” map. What does it
stand for, and what does it do? Well, I’ll get to that in a moment. First, I’d
like everybody to understand what bump and normal maps are, since that should
help in understanding what an ssbump is. All three maps control how light is reflected
off of the surface they’re applied to. Bump maps do this by using light and dark values. The
lighter values cause the area they coincide with to reflect more light back to you whereas the
darker values do the opposite. This is done by skewing the normals– which are lines perpendicular
to the surface– towards and away from you. You can think of this like a person using one of
those tanning reflectors at the beach. When they point the reflector towards themselves to
get “more sun,” they’re also angling the normals towards themselves like a lighter value in a
bump map. In 3D applications, this simulates curvature across parts of the texture where the
bump map transitions from one shade to another. Normal maps serve the same function, but take
things one step further by controlling the skew of the normals in all directions– not
just towards or away from you. This is done using RGB color values which, if you don’t
know, are red, green, and blue. By mixing these three colors together, the surface
normals can be skewed in any direction, allowing for more control over how
light reflects off of the surface. Both maps are well-known
to 3D and material artists, and are used in practically every
modern game, but not TF2. Well, it is, and every community mapper I’ve seen has
used them, but many of the game’s stock assets, like this brick wall I keep coming back to, use
something else called “ssbump” maps– where the “ss” stands for “self-shadowing–” and believe
it or not, it was actually created by Valve. In fact, Valve employee Chris Green wrote
an entire paper about it back in 2007, where he dives deep into the inner workings of
ssbump. I’ll only be covering a small bit of it, so feel free to read the rest of the paper on
your own if you find this interesting. And, you can also follow Chris on Twitter where
he posts stuff like this on a regular basis. So what does this paper say about ssbump?
Well, to start, Chris states that bump maps lack “self-shadowing of surface details and
ambient occlusion,” and that by doing this– you can pause and read this if you want– a kind
of ambient occlusion can be added to the texture, which is when certain parts receive stronger
shadows in areas we’d expect them to be, like the crevices between these balls. A little further in, he adds the following: SSBump works with dynamic lights. It provides softer shadows and ambient occlusion. The shadowing information is generated
from height data– or bump maps. There is no increase in storage
space when compared with normal maps– I know it says bump, but trust me on this. And it preserves the use of the normal map
alpha channel for other visual effects. Okay, this all sounds nice, but what does it
actually look like compared to a regular bump or normal map? Well, there are two sets of
images in the paper, this one up at the top, and this other one toward the bottom. In the
bottom one, we see that Chris refers to what I’ve described as a normal map as a bump map, and
a bump map as a height map. This is why I swapped the terminology when going through the bullet
points. This paper was written back in 2007, and terminology conventions have
since changed, hence the confusion. But, I didn’t want to just take
Chris’s word for everything, so I created this test map where we
can compare things for ourselves. First, let’s see how this brick wall texture
looks under all of the different texture maps. Here, we have the base texture with nothing
else, so it looks flat. Notice how painterly it looks. That’s because Valve painted each brick
separately in Photoshop over a real-life photo. Next is the base texture combined with a
black-and-white bump map. You can start to see some of the curvature forming, which
becomes much more apparent when I disable colors. It looks a little weird, which I think
is because I had to enable the $ssbump parameter to get it to work in the first place, so I
don’t think it’s an accurate representation of how a bump map should really look, but I
couldn’t figure out a way to get around it. Now we have the base texture with the ssbump map,
which forms the version we see throughout TF2. Now we can see the curvature really take form,
especially around the crevices of the bricks. Finally, we have the normal map, which looks
similar to the ssbump map, but with a lot more curvature. You might even say that it looks
more realistic, but more on that in a second. So which one looks the best? You might
think that the normal map is the best for realism and that the ssbump is better
for a game like TF2 where the art style is relatively flat and painterly. And while
I wouldn’t say that that’s a bad takeaway, I do think it undermines a special feature
within the ssbump that isn’t in the normal. It might be hard to notice at first, especially
with colors turned on, so if I turn them off and point a flashlight at both of these walls,
we can see how the ssbump really shines; the crevices between the bricks are dark unlike
in the normal, which is the occlusion Chris was talking about. So in a way, the ssbump is more
realistic than the normal. But if that’s the case, why does it appear to produce less curvature than
the normal, and why was it abandoned in Source 2? Well, to answer the first question, my guess is
that the occlusion effect “softens” areas where sharp bits are poking out, resulting in better
anti-aliasing throughout the texture. So then, maybe this normal map setup isn’t more
realistic and is just missing lighting occlusion information present within ssbump
that ultimately makes it appear crispier. Regardless, I think I speak for everyone
when I say that ssbump works better for TF2 because you’d never want to have
such dramatic changes in a texture’s curvature when the entire game is meant
to look smooth and cartoony, and despite ssbump being potentially more realistic,
it just so happens to bolster this effect. So, if ssbump is more realistic, why
was it removed in Source 2? Well, as you may have heard, Source 2 uses something
called “Physically Based Rendering” that Source 1 does not. It’s the standard across
all modern game engines nowadays, and comes with new texture maps that aren’t
present in Source 1, which includes something called an “ambient occlusion map.” This nullifies
the need for ssbump since you can just make your own texture map to add occlusion instead of
trying to lump it together with the normal map. Before I wrap up, you might be wondering about
those other things I brought in the bullet points, like the “dynamic lights,” “storage size,” and
“other visual effects.” Well, I can confirm that ssbump does indeed work with dynamic lights,
even though we never see them used in TF2. And I can confirm that the file size isn’t any
higher than a normal map because the size of the ssbump and normal map PNG files– which I can
get by extracting them from their VTFs– are practically the same. In fact, in this
case, the ssbump is actually smaller. And finally, by “visual effects,” I can point to
this bullet point at the end of the paper where Chris writes, “Texture blending for operations
such as detail texturing, texture cross fading, etc. are much more straightforward.” To give an
example, I can point to these two rock ground textures I brought up in my video “Everything
Wrong with Dustbowl,” where I described how there’s a seam between them even though they use
the same base texture and have matching texture scale, shift, and UV coordinates. The texture
on the left called “rockground002” doesn’t use a bump map, but the one on the right called
“blendrockground004” does, and it uses them in conjunction with this “$blendmodulatetexture”
parameter, which I think is what Chris was referring to when he said “texture blending”
like “cross fading.” I might be wrong, though, so feel free to correct me about anything
I may have misinterpreted in the comments. So that’s why TF2 isn’t normal, both figuratively
and literally. It uses an interesting development by Valve that wasn’t very well known about
to begin with, and which is now obsolete. Thank you for watching,
this is LED, switching off.
