MasterEffect ReBorn official thread

GTA 5 AO is also horizon based, HBAO. Its bilateral filter has a too wide interleaved sampling so if you zoom in on the 4k screenshots, you see a grid pattern.

Your AO looks pretty good, is the right picture color bouncing?
Is it something like ssgi and ssdo?

SSDO has a fixed direction, in many adaptations done with a fixed light direction. Mine is more like those regular SSGI implementations and Boris' SSIL, where IL is the better term, Indirect Illumination. It's one of the two things I'm experimenting on. Separable filter needs depth info for blurring so AO does not bleed over object edges. Now usually I would do AO in RGB channel and depth in alpha channel so it's only one texture lookup per blur sample but if I have IL also, I either have to mix AO and IL somehow and put it in RGB or put IL in RGB and AO in alpha channel. That means I need 2 texture lookups per blur offset, one AO/IL texture and one depth. No problem with low blur amount hence my other problem: trying to find a noise function that can be blurred to a homogenous surface with only a few samples. Currently I use an ordered dither pattern but this is problematic on lower resolutions (most common solution for AO performance improvement, rendering in lower res).

Hope you solve those problems.
Too bad I can't help you with those, I've never done any shader programming.
I've only worked in Gameplay and AI areas of game development, never in Graphics.

Almost no one will be able to help me with that, even if I could find someone who has experience writing AO shaders, these guys never take care on blur friendly stochastic sampling, it's just 1) add a random random generator 2) don't think about it anymore 3) ???? 4) profit

I read on this link that in Assassin's Creed IV Black Blag they used temporal supersampling to improve AO smoothing on consoles (they used Alchemy SAO). They used the temporal supersampling before the blur to avoid undersampling artifacts. Maybe you can try that.
Although I dont know if Reshade can give you previous frames info, I think this was one of the problems why CeeJay.dk couldn't implement SMAA T2X.

We have previous frame information but thats only about scene color, depth and alpha. Most temporal XX approaches (e.g. temporal SMAA) take velocity values into account. Of course we can still calculate estimate from the color and especially depth information but its never gonna be as accurate so probably not worth the effort/impact. The current motion blur algorithm uses those estimates but its not a shader that needs AA accuracy.

I was suggesting Marty that he could try temporal supersampling on his AO algorithm to solve a problem he said he had.
Maybe he could improve AO accuracy by using previous frames info before applying blur.
That's what was used in Assassin's Creed IV Black Flag .

ivanosky wrote: I was suggesting Marty that he could try temporal supersampling on his AO algorithm to solve a problem he said he had.
Maybe he could improve AO accuracy by using previous frames info before applying blur.
That's what was used in Assassin's Creed IV Black Flag .

keeping in mind ac4 on ps4 uses temporal supersampling in general not just on ao they probably have a trick where they hit two things at once here

Using info of previous frames is a very easy task and can be used to improve a multitude of effects in games. It's just it cannot be realized properly on ReShade. I have other solutions though for my AO, it is actually pretty fast. Maybe I keep using the dither texture or I go another way using frac based noise functions, these produce a certain kind of randomization, different frequencies like small and big waves. Need to try a few things to find a blur shader and a random function that work good together. So far the dithering seems most efficient, needing a smaller blur than true randomization but there is still room for optimization. It's problematic on small sample counts and as I said, on lower resolutions.

After all, if Rockstar Games is satisfied with such an AO, I shouldn't worry:

Hello, first of all thank you for your hard work.

I was hoping someone could help me with a crashing problem...
I believe it occurs due to MasterEffect.h and ReShade.fx, I don't know how or why... but ReShade.dll (d3d9.dll) and the unlocker launches fine, but this is the crash report I get:

gyazo.com/0e33b46b44cc8474b20c5ed127b9825c

This is in the character menu with a black screen and interface text.

I did all the steps, I even tried ReShade with Sweetfx and that's the same problem... I tried googling but the problem is just non existant to anyone else...

If you want to use the ReShade unlocker, fine, but don't expect support here.

Marty McFly wrote: II try to make it close to Boris Vorontsov's SSAO which is by far the best one ever,

No the best AO -and I just found out about it- is this one : )



Nvidia VXGI sample - DX11 only - needs depth buffer and projection matrices from the game to operate properly. In other words practically impossible to integrate in a universal injector like the ones we have (I don't think approximating matrices would do the trick).

But boy does it look awesome !

Boulotaur2024 wrote:

Marty McFly wrote: II try to make it close to Boris Vorontsov's SSAO which is by far the best one ever,

No the best AO -and I just found out about it- is this one : )

[img
But boy does it look awesome !

doesnt this only run on maxwell due to the hardware change to handle voxel data faster

Voxel based stuff, I must admit I don't know how one would integrate stuff like that into ReShade (if it's possible at all) and per pixel normals AO is always a million times better looking than simple normalfromdepth normals which contain polygon info,.
However, I'm pretty satisfied with mine atm, it's fast, good looking, view angle stable and versatile. Plus it can do IL so it's pretty much GI. This drop to pitch black at steep corners, well, let's just say, my falloff function is optimizeable.

Marty McFly wrote:

that looks really good as is atm i mean you can sit and nit pick at something forever but you gotta find a middle ground for this is acceptable heres my goal and find perspective of what is actually achievable reshade cant and shouldn't be used to make the most real effects i think if devs are lacking in that then projects like this showing how effects with basic info can do more then their half ass works do should be enough to start making a point. Also like my last post voxel based effects are super expensive on anything but maxwell atm average user is probably on fermi gen cards or equal amd even kepler cards are slow with anything using volume based/tessellation hardware isnt designed to work well with it considering nvidia states maxwell is 3x faster on average with those if you take in the numbers they throw out either way it in general for average user voxel based stuff is pointless

So I just checked the VXGI sample and - well implementing is impossible due to one very good reason: it's half in byte code, half in ASM.

With the new framework I have a problem where when using manual focus the dof_nearblurcurve value has no effect. Does anyone else have this problem, and is there a fix?

Do you have the focus distance set to anything significantly greater than 0?

it's at 0.01 currently.