Blended normals hack for fake SSS
polycounter lvl 19
Hey guys,
Tried the "Blended normals" shader hack today in Mental Mill. Found info about this from Naughty Dog's Siggraph powerpoint.
Works quite alright for the rendering cost if you ask me.
Head and textures is from Infinite Realities, as posted on PC News page a week or two back.
Here's my results.
Gif doesn't do much justice, but it's nice to see them on top of each other for comparison reasons.
EDIT: Oh, and a big thanks for my colleague Niklas for helping out with the blended normal effect.
Tried the "Blended normals" shader hack today in Mental Mill. Found info about this from Naughty Dog's Siggraph powerpoint.
Works quite alright for the rendering cost if you ask me.
Head and textures is from Infinite Realities, as posted on PC News page a week or two back.
Here's my results.
Gif doesn't do much justice, but it's nice to see them on top of each other for comparison reasons.
EDIT: Oh, and a big thanks for my colleague Niklas for helping out with the blended normal effect.
Replies
I wondered if the same could not be done with this method, simulating scatter with the weight rather than the blur radius.. not as fancy but there would be no need to do the light blurring or use memory for multiple normal maps.
Regarding this method it does not use multiple normals maps. From what I can tell the extra cost compared to not using it is one extra diffuse light calculation and one float3 lerp.
What you essentially do is calculate light for diffuse light twice, once with normal map and once without. Then you lerp between these results with separate weighting values for each channel (RGB) in the lerp. You let the red color stay closer to diffuse light without normal map and the green and blue closer to the normal mapped result. This gives the illusion of the red scattering.
I definitely want to find out if I can force a lower mip through mental mill.
Do you know how to force a certain mip level with HLSL/CGFX code?
Thanks.
@kodde: i think what your looking for is called LOD bias, or mip bias, a negitive value will force higher mips a positive value will force lower mips.
// suppress the blue and green components
blendedDiffuse.g = min(blendedDiffuse.r + SuppressCyan, blendedDiffuse.g);
blendedDiffuse.b = min(blendedDiffuse.r + SuppressCyan, blendedDiffuse.b);
lower the value, the more cyan is suppressed.
Ah thanks. It was along these lines I was thinking originally. But I figured there might be more to it. Shouldn't suppressing this also mean the red "bleed" effect would be less apparent? I mean, isn't there some correlation to how having much of the desired effect also means more of the not that good parts such as increased cyan?
I'm used to editing my shader in Mental Mill but I'm not completely lost when editing shaders by hand. I don't have access to MM here but I went ahead an edited it by hand. Here's a video of my suppression atm. I'm quite content with it's functionality. Thank you for sharing!
[ame]
Cyan seems a bit suppressed with the video compression as well, not as vivid anyway. Should be 1080p available ^^
Chronic> Yes thank you I have seen your skinshader
Its akin to using env mapping and calling it "specular". If your engine is really so poor that it has to use such a low effect like this, it shouldn't be using normal maps anyway.
i disagree completely, when you use proper maps and good settings it looks great. go play uncharted 2, or read the white paper, i would say currently for its cost is the best way to attempt SSS in "real time"
this is using the same method, just took this from maya...
****see one post down*****
In addition, Image-based reflections (env maps) are technically closer to 'specular' and more physically correct than what most games use. "Specular" is a fake concept used to make up for our inability to diffuse light and reflections in real-time.
Cubemaps are also more expensive than specular. So your analogy really doesn't make any sense.
A much clearer example of how powerful this can be.
I have to disagree with Odium, it's a great 'hack' and a relatively simple and cheap one as well.
EDIT:
Does the lighting model differ anything else besides the Blended Normals? No wrap around half-lambert stuff going on? Differing Ambient lighting?
My skin shader does use a modified diffuse shading function with an adjustable edge, but if you pull it too far across the surface, approaching half-lambert, things stop looking realistic, but a little softening can be nice.
Edit: taking care to be linear in your shading really helps things look better and more realistic too.
EDIT: Do you incorporate any kind of gamma correction?
Spent a few minutes making a basic spec map with gloss. Upped to 2048 maps.
The difference between the two images is that the shot on the right has:
-Blended normals with Cyan suppress
-Red color tint in darker diffuse light spectrum
-Light terminator edge pushed back slightly and tweaked a bit
I have an idea on how to get pass through light, but without shadows it will look strange.
EDIT: Just noticed that with blended normals you can increase normal map intensity/amount a bit above usual since this feature reduces the nasty black areas the normal map usually generates. Around 1.5x level still looked pretty good to me. The image above is still 1.0x normal map intensity.
If you look at the first posts in this thread you can see the difference between not using blended normal and using blended normal, no other really fancy things going on there.
In this pic I think it's the pushing terminator line backwards that is also slightly increasing the light areas a bit. Shouldn't be that much though.
I tried using this curve earlier but as you can see some areas would actually get darker. This might not be a bad idea after all. I mean, a SSS material would absorb more light and therefor not be as lit as a non SSS material no? Or am I not seeing something? Does all the light eventually find it's way out again? Does some light deteriorate as it scatters around or how does light work?
Not really. The earlier pics really do look like trash and give off a white SSS glow rather than a proper sub dermal effect. The later pics don't exactly look perfect, but look much better, I'll admit to that.
But the key there was in what you said... You just called my "opinions" wrong... Thats like saying I'm wrong because I don't like Mac & Cheese. How can anybody be wrong for an opinion?
Regardless, the later shots in the thread have sold me much more than the original, hacky 1999 style normals they were showing with bad blocky lighting, so relax, its all gravy
I think Earthquakes comment was more directed at how little you seemed to realise the potential of this.
Yes the original image was not the greatest example of it but it was however a first pass attempt and therefore required us to look beyond the immediate flaws as Arshlevon did and think of its potential uses.
Calling a work in progress image trash just highlights how little you get it btw, not cool.
A quick read of any white-paper regarding this technique or some basic understanding of premise makes you look past the visual discrepancies in the initial example quite quickly.
Kodde:
Looking good, would be interested in seeing some examples with the normals pushed to 1.5 as you talked about.
Just wondering, but before i say something i have to make you understand that i have no technical background at all, so i shouldnt realy say anything :P
It's about the light passing trough thin surfaces, the new render to surface maps in 2011 have a sss option wich calculates how thin a area is, isnt it possible to use some of that?
This method needs to be combined with shadows to be more work more properly. But should be fully possible.
Even tough i'm not a character artist, i would like them to get a real good skin shader to showcase there models.
Knee-jerk reaction, no helpful advice on how it could look better. You claim it is nothing like how SSS works, but provide absolutely no technical information as to how it actually should work. This whole response is akin to saying "Shit sux", without backing up your argument with any sort of reason, logic, or adding anything positive to the thread. Basically the end result of this statement is showing polycount a new and exciting way for Odium to be a twat.
This is just factually incorrect on so many levels, that it makes any argument you possible could have thought about making above entirely irrelevant.
A. "Specular" is not a real thing, its a hack, a trick to approximate the effects of general reflection. "Specular" and "Reflection" are not two different things that exist in the world.
B. Using a proper Image Based Lighting environment in a game engine is not only more accurate, more physically realistic, it is also not a "cheap trick" that you would use in a low spec game, to suggest this is just silly.
Moreover, using cubemaps, with gloss maps to define the sharpness of the reflection is *very much* physically accurate to how the world actually works, and in an ideal system you would have these cube maps generated in various spots of your level. Much like what HL2 does, ideally, these could be generated per-frame or very frequently to have a wonderfully realistic, persistent IBL environment, again far from a "low spec" effect. This is the next best thing to *realtime raytraced reflections*, and certainly a far leap past pointlight only dynamic lighting systems.
C. Many games coming out today use IBL systems(cube maps!) for their games, and most games you will see come out in the future will use them. The days of Doom3 style point light only lighting are long gone, as they should be.
This sounds interesting, combining gloss functionality to the reflections. By sharpness do you mean like more/less blurred?
I'm guessing everyone is seeing how much of a twat Odium is.
haiddasalami> Good to hear, that's what internet forums are all about imo.
My Lead and I were talking about tinkering with the possibility of using pre-generated cubemaps for that exact purpose, do you know of any games that have or will be doing it?
I'd love to see some tests/results.
Maybe start a new thread if you do, so as to not high-jack Koddes awesomeness.
I can confirm it's definitely a sweet methodology though. and it let's you make badass metals!
In my shaders for this system the "gloss" is referred to as "roughness". It's stored in the alpha channel of our full-color specular maps.
It's used to do several things including:
controlling the cubemap mip-map levels (ie; "blurriness" of the cubemap)
controlling the exponent of the specular (we still use specular to reflect dynamic lighting)
mask a detail normal/specular (ie; "rough" areas can optionally get a detail normal that "smooth" areas won't get).
I think DICE used a system like this in BFBC2? check out this awesome whitepaper on their [url=http://www.slideshare.net/repii/frostbite-rendering-architecture-and-realtime-procedural-shading-texturing-techniques-presentation?from=ss_embed
]frostbite rendering system[/url]. (page 51)
As EQ said it's more physically correct than the standard cubemap/specular methods. I have spoken with several people who have compared it to mental ray's shader setup in terms of functions and controls.