Implementing the Tron 2.0 glow effect

You might want to check out the HDR demos that jollyjeffers and Muhammad Haggag have posted. thread 1 thread 2

Those are neat, but the most important part of the effect is the artist pipeline, where the artist can specify the glow amount through the alpha channel of the texture. The examples above don't implement the glow in quite the same way.

Hi,

I'd been meaning to implement this myself as I think the effect looks great and is pretty cheap (when compared to full blown hdr and bloom). So when I saw your post I figured i'd give it a shot [wink]

Anyway ...these are the steps I took (this is probably not the fastest way of doing it and if anyone else has some optimisations then please jump in [smile])

Step 1: Render your scene as normal to a texture (let's call it "originalScene") with the dimensions of your viewport. We keep this texture for later as it is used a couple of times.

Step 2: Render a screen aligned quad, with the texture "originalScene" mapped to it, into a second texture. This texture has half the width and half the height of the viewport (quarter the area). Also in the pixel shader you multiply the R, G and B by the A (just like it says in the gamasutra article). We will call this texture "glowTexture".

Step 3: Render a screen aligned quad, with the texture "glowTexture" mapped to it, into a third texture. Like before this texture's dimensions are half of the viewport dimensions. In the pixel shader we blur it along the horizontal and we'll call this texture "glowBlurH".

Step 4: Render a screen aligned quad, with the texture "glowBlurH" mapped to it, into a forth texture. As before its dimensions are half of the viewport dimensions. In the pixel shader we blur it along the vertical and we'll call this texture "glowBlurV".

At this stage you have all you need to render the final glow enhanced scene [smile] So....

Step 5: Render a screen aligned quad as normal (no render targets this time) and map "originalScene" and "glowBlurV" to it. Perform a simple additive blend and there you have it [grin]

Here is the result from the app I just knocked out (with MDX btw)...

My blur weights aren't the best and it could do with a lot of tweaking, but you get the idea [smile]

All the best,
ViLiO

Sweet!

Would you mind posting a few relevant parts of the MDX source code? Specifically, the syntax for rendering to a texture. Are you using multiple surfaces or the StretchRect method?

-Dave

I'd suggest you check out some of the resources on mdxinfo as I know a few of the intermediate samples use render targets (the Simple Bloom Sample may in fact be the Tron 2.0 technique but i'm not sure)

Anyway....



Regards,
ViLiO

Also check this and this. It shows (with code) how they modded Half-Life to add the glow effect.

One thing to bare in mind if you're doing this on "traditional" texture formats (that is, not floating point!) is you can get the filtering hardware to do a lot of the work for you [grin]

If you're careful about your placement of sampling points/offsets then you can get the filtering hardware to sample two texels and automagically return the average of both (which is very similar to what the bloom filters do anyway).

Thus you can either get "crazy bloom" for the same cost, or you can get the same bloom for a lower cost. With particular reference to the latter, you might even be able to combine the 2-pass approach into a single. YMMV.

This post (seemingly from an ATI person) suggests a slide deck I found very useful in this matter.

hth
Jack

Any way to do something similar using a fixed function pipeline? I'm trying to support old cards/ laptops that do not have shaders yet. I think the only way in this case is to write new textures to the polygons (switching between texture colors).

It's possible to do it with the FF, but the performance will likely be lower.

At the crudest level the blur can be done with just single texturing and lots of blending and offsets (using vertex alpha to set your weights). When I did my FF version I was using a symetrical kernal/filter, so I used two texture units at a time (you start with the center sample and then other passes do samples to the left and the right at the same time, as they have the same weight).

It's probably possible to use more texture units, but it's a right pain to get the weights setup right (especially in a cross-vendor way). If you really want to do it that way you might be better off using Cg.

Edit: I notice I've strayed into the DX forums. [grin] In that case setting the weights might not be as difficult as in GL.