Marty McFly's Ambient Obscurance (MXAO) with IL

Are you perhaps using a lens shader before MXAO?

Game: Accel World VS Sword Art Online
Reshade version 3.0.8
Shaders in use: SMAA, MXAO, LumaSharpen, AmbientLight
After disabling every other shader MXAO it still looks out of place.
AO on




AO of

Video is up

Could someone, please, list some games where this effect works relatively well, without major issues (doesn't have to work perfectly)?

Works on every game that has flawless depth buffer support, that means all games with "Perfect" compatibility rating in Games section.

@DoomedSoul: I cannot figure out any rule by which this depth buffer is distorted, it's neither shifted or zoomed or both. It almost looks as if the camera position for the depth buffer is different. I can't help you fix that. You can't use any depth depending effect on this game then, sorry.

Hi all.the image is upside down.Any suggestions?

https://ibb.co/dDv8zQ

Pretty easy to fix go to Settings tab and change RESHADE_DEPTH_INPUT_IS_UPSIDE_DOWN=0 to RESHADE_DEPTH_INPUT_IS_UPSIDE_DOWN=1 than go to Home tab and press Reload

Marty McFly wrote: Works on every game that has flawless depth buffer support, that means all games with "Perfect" compatibility rating in Games section.

Black Desert is listed as perfect, but the MXAO effect renders over the UI, making it a hassle to try and read text.

Off: i.imgur.com/5iYVQ6h.jpg
On: i.imgur.com/XwA7w7H.jpg

As it's an MMO, the UI is pretty of important, so it's kind of unusable, which is a shame because it looks so damn good and makes all the foliage look so much better.

Is there anything that can be done about this?

Yeah well, perfect compatibility rating doesn't mean that it filters out the HUD, it merely means that it also has depth buffer support, which isn't given for all games. I've never played black desert, can the ui elements be moved? I'm asking because one could apply a static mask that prevents MXAO from being applied on these areas but that'd require the ui rlements to be at the same positions, otherwise it'd look very very odd.

The UI can be freely moved anywhere, yes.

So this issue happens in some other games too? It's quite a big killer outside of just taking screenshots..

Yes, it happens in every game, there's no generic way to detect when the HUD is drawn. ENB is specifically made for a single game, GeDoSaTo has specific hashes für such a thing so the user has to control when the shaders are injected.

Splendid news for all MXAO users: just got Temporal Supersampling (TSS) working. This technique varies the SSAO samples every frame a bit and combines the current frame with older ones for the final output. As you can see on the attached picture, the quality of the AO increases a LOT - this is raw AO with no blurring and only 8 samples. You can now use a lot less samples and still get a good looking AO - major performance saver!

That is really awesome. I really look forward to try the public consumption version. Not really sure how it will work but it won't look weird when every thing is in motion? Thinking about mix match between old frame and new. Not that is a problem for example temporal aa.

I really like your let's do more with less approach

There is ghosting - obviously. But it's only visible in debug mode or if you really try hard to look for it. It's an option for users who want that bit more fps. And as a side effect, the AO noise in movement for high frequency details such as grass - is gone the image looks much calmer now.

EDIT: I experimented a bit and for small scale shadowing like in many games, as few as 3 (!) samples are enough. This is on par with ASSAO, Intel's latest SSAO technique.

Next step would be sampling not all pixels not every single frame, if such a thing is possible. In the lines of estimating the work we don't do this frame based on the work it's done this frame and the work that was done last frame.

For example, divide the image pixels as if they were a 1920x1080 chessboard.
Frame 1, do work for the white squares. Frame 2, do work for the black squares. Avoid work which is not done this frame by estimating from previous frame (which worked on the pixel we're skipping right now), together with the surrounding pixels, which got updated just now.

I know this is a wishy-washy idea. It'd be nice if it worked though. Haven't had much success myself.

I did that earlier but it doesn't really work out plus it's more difficult in code. It's technically no difference if you, let's say do 10 samples for 1/2 the amount of pixels each frame or 5 samples for the full pixel amount each frame. As there is a constant cost for the code outside the AO loop, the latter is faster, even.

TSS like I do it is actually common for many games, except they don't use the exponential moving average like I do but copy multiple separate buffers and combine them as output. I interpolate between current frame and last frame (which is an interpolation between the last frame and the frame before...and so on). This means I had to choose an intelligent jitter algorithm so the total contribution amount for each sample jitter variation is same. But it saves me the cost of multiple additional textures, thus reducing the memory footprint.

For me personally everything calmer picture in motion is very nice improvement. I think grass in motion always looked wierd with ao but looked even weirder without ao. Heh, I guess everything is always a compromise between two things. I really look forward to try this out.

Really cool stuff. This could be really useful for people with little gpu power.

Marty McFly wrote: I did that earlier but it doesn't really work out plus it's more difficult in code. It's technically no difference if you, let's say do 10 samples for 1/2 the amount of pixels each frame or 5 samples for the full pixel amount each frame. As there is a constant cost for the code outside the AO loop, the latter is faster, even.

TSS like I do it is actually common for many games, except they don't use the exponential moving average like I do but copy multiple separate buffers and combine them as output. I interpolate between current frame and last frame (which is an interpolation between the last frame and the frame before...and so on). This means I had to choose an intelligent jitter algorithm so the total contribution amount for each sample jitter variation is same. But it saves me the cost of multiple additional textures, thus reducing the memory footprint.

I'm guessing that since this is a Temporal technique, it won't work with SLI activated?

Zireael wrote:
I'm guessing that since this is a Temporal technique, it won't work with SLI activated?

I would guess that this is a post processing effect of the frame it will work. It really just takes frame after it has been rendered. Other temporal stuff is happening while it puts together frame to be rendered. But reshade is always last, like a layer in photoshop. So I will think that will work with SLI.

Though I am completely layman I could be wrong.

Sunesha wrote:

Zireael wrote:
I'm guessing that since this is a Temporal technique, it won't work with SLI activated?

I would guess that this is a post processing effect of the frame it will work. It really just takes frame after it has been rendered. Other temporal stuff is happening while it puts together frame to be rendered. But reshade is always last, like a layer in photoshop. So I will think that will work with SLI.

Though I am completely layman I could be wrong.

Yes I was thinking myself the same, but I rather not speculate and ask the source