SMAA

Hello Crosire hope you are good.

Great work on the 4.2.1 update, the GUI is really smooth and everything loads no problem.

We have always used Boulotaur2024's DX11 SMAA injector on WIN7 and it was brilliant in all DX11 applications, it looked the same as the DX9 Masterfx render of the SMAA. Here are some screenshots of DX9.











Unfortunately were having some real issues with DX11 SMAA injection. It doesn't seem to be working correctly with WIN8.1/10, its a subtle difference and looks nothing compared to the DX9 render of Masterfx or Sweetfx.

We downloaded MasterFX from Martymcfly for DX9 applications on WIN8.1/10 and it works exactly the same as WIN7, it works correctly, you can clearly see the pixels crunched up and it looks brilliant.

So
MasterFX for WIN8.1/10 on DX9 applications works correctly.
Reshade 4.2.1 for WIN8.1/10 DX11 applications has render issues.

I honestly thought that rebuilding Boulotaur2024's Revision 47 would fix the issue and months later we have built the DLL's but they just don't want to activate? They do generate a log file but its empty, it refuses to log anything? Such a shame.

I did create a forum thread 3months ago to get as much information as possible before trying to contact you, its a mixed message.

Users are stating that WIN8.1/10 has increased LOD Bias which sharpens the image, and I believe this is true because I noticed when I moved from WIN7 to test out a new OS WIN8.1/10 that all games were sharper, which creates aliasing. Now that we have MasterFX working in WIN8.1/10 in DX9 I don't understand why Reshade cannot produce the same image? I could understand if MasterFX produced the same SMAA quality as Reshade 4 but its not the case, SMAA does indeed work on DX9 correctly.

Many users have stated that the SMAA isn't as it used to be, I downloaded Sweetfx version 2.0 and guess what, this old version does a better job at SMAA than Reshade 4, somethings not right with Reshade, it not crunching up the pixels like Boulotaur2024 or the MasterFX version.

Thankyou for your time and hope to speak soon.

This has nothing to do with ReShade and is simply because the SMAA shader distributed ( github.com/crosire/reshade-shaders/blob/master/Shaders/SMAA.fx ) is the official SMAA, whereas the SMAA shader that was part of SweetFX was a customized version by CeeJay. Updating that version would yield the same results.

Hello Crosire and thankyou for the reply much appreciated.

So from my understanding, Sweetfx has a completely different set of SMAA parameters? That Ceejay created a custom SMAA version for Sweetfx? Also you said that updating Ceejay's SMAA would make no difference? Users on this forum was stating that all the versions of Sweetfx or Reshade have the same SMAA parameters? Now were confused?

Isn't MasterFX based on the old Reshade? This version does a much better job at SMAA than Reshade 4.2.1 in DX9.
You said that using Ceejays' SMAA wouldn't make a difference? If I understand that right, using the old Sweetfx 2.0 K-PUTT does a better job at SMAA than Reshade 4.2.1.

But the best version of all is the legendary Boulotaur2024's SMAA, what an absolute beast of SMAA. If only we could speak with Boulotaur, the true god of SMAA. We was hoping and preying that rebuilding the parameters for Revision 47 would end all this, but I, a mere mortal cannot update Boulotaur2024's shaders, its way beyond my capabilities, that method of AA is outstanding.

So is MasterFX using a custom version of SMAA also? Seeing as its a completely different render from Reshade.

I'm sorry for asking so many questions but I'm so confused on how moving from WIN7 to either WIN8.1/10 that there is no decent SMAA anymore, this is how all games on WIN7 with SMAA used to look like.



This is taken from here www.iryoku.com/smaa/ one of the official SMAA pages, Reshade cannot render anything like that anymore, that picture is no custom version of SMAA either, so if Reshade is infact using the official parameters for SMAA, it should look like that screenshot, but it looks nothing like, there is absolutely no blending/crunching of pixels what so ever in the new Reshade I'm sorry to say, many other users have posted this same issue with all due respect, let me upload some screenshots of DX11 games on how they used to look.

Thankyou for your time and hope to speak soon.

Just wanted to say -- I have noticed this too. I think, though it's been a while, since the move to Reshade 3.X (meaning, from memory, Reshade 2.X was fine). SMAA just has never worked as well as it used to, often seemingly only partially anti-aliasing jaggies regardless of the settings used.

I've been meaning for some time to set up a comparison using older versions but RealLife(tm) constantly saps my time. However now XRJ44 appears to have done that! Considering injecting SMAA is one of the core benefits of Reshade it would be really good to get to the bottom of this. I'm not a programmer so can't help with the shader side of things, but will happily test builds for anyone who takes this on.

Can also confirm something is fishy with SMAA with latest reshade compared to bolotaur version.

Looks like the shader got updated, from what little I can tell it's now using the ultra preset settings meaning it's toning down the extreme values for detection and there's a 25% corner rounding which should improve efficiency but will soften up text and UI elements a bit from how I remember when it was discussed.

Seems to be changes to the way it gets the depth buffer too guessing that fixes a earlier workaround for getting it to recognize the preprocessor settings for reversed and flipped in ReShade and as a part of what looks like a standardization of features for ReShadeUI.fxh it should again be fully compatible with ReShade 3.0 and 4.0 and then that UI file will probably be updated instead of the shader files themselves if required for later functionality in ReShade itself.
(Though in addition to ReShade.fxh that means most shaders now also require ReShadeUI.fxh for this inclusion and it's functions.)

Not sure if it fixes aliasing coverage compared to earlier versions though, rise in specular surfaces, shader effects and all that so it's not just about geometry edges and luma or color detection which I guess is also one reason for the increase in temporal AA solutions in games now even if the softening side effect can be a downside in terms of overall image quality and perceived sharpness.

github.com/crosire/reshade-shaders/commi...e3c8b1764e99d66fe123

Hello.

Looks like the forum mods have moved my posts here to a new thread.

Let me explain.

Morphological Antialiasing is already in the AMD driver control panel, which is an old method of AA, you can activate this on all DX11 games but it looks really dated and effects fonts and GUI elements. You can clearly see where SMAA took its parameters from, they look quite similar in that SMAA fixes the fonts and other elements.

SMAA is an advanced method of Morphological Antialiasing, which Reshade isn't activating any ''Morphological Antialiasing'' which is the foundation of SMAA, the ''crunching up of pixels'' what I explained earlier.

Also exactly what other users have stated? How can you not see this Crosire? There is no Morphological Antialiasing for SMAA to even work off it looks like.

Crosire please if you can just create a test version of Reshade with ''Morphological Antialiasing'' only, then update the parameters with the SMAA from Boulotaur2024's dev site.

There is no SMAA working in Reshade 4 its obvious by now. Even I saw the problem since Reshade 3 and its still the same now.

Thankyou for your time and hope to speak soon..

Not denying things may look different. But SMAA is working just fine. Check the debug textures if you don't believe me (the "View edges" and "View weights" options). Comparison:

Original image:

Official SMAA via SMAA demo from www.iryoku.com/smaa/ :

ReShade SMAA:


There are a few slight pixel differences in the comparion above which I can't explain (I'm guessing it's because of HDR vs LDR), but nothing groundbreaking.

Also, there is nothing special about Boulotaur2024's SMAA. It's a more or less straight copy of the SMAA code from the official SMAA demo. SweetFX SMAA was edited by CeeJay with a few performance improvements. Somebody would need to port them over to the current ReShade SMAA shader. But they shouldn't introduce big quality differences.

Hello Crosire and thankyou for the reply.

I have to respectfully disagree here, Reshade SMAA is not working. With all due respect you don't seem to be listening to anybody here, users are clearly stating that SMAA is broken, chop off my head if you have to but its true, Reshade 2 seems to work I guess ok but its not as good as Boulotaur2024's shader which you can clearly see Morphological Antialiasing.

How can you say there's nothing special about Boulotaur2024's SMAA? The only shader which actually worked with x64 DX11 correctly, how can you say its a straight copy of the official SMAA parameters when that is x86 DX9 only, Boulotaur2024's made SMAA x64 DX11, that is world class bloody hell, have some respect when you talk about gods.

We cannot produce anything like the screenshot you have provided, it looks no where near, we have tried many different settings but its basically the same pixelated image quality. We can clearly 100% see that there is no Morphological Antialiasing active, you can clearly see.

You have an Enhanced Subpixel AA but no Morphological Antialiasing AA, there not combined, so you don't have SMAA or any form of it I'm afraid which is a shame because you did. Was it removed because AMD own that AA? Something's gone off here. How could we apply only the Morphological Antialiasing to Reshade if I may ask?

I have wrote a detailed explanation of version comparisons for you to maybe analyze and find the issue but you are adamant that your SMAA works, I can tell you for free its not the case here.

You stated earlier that Reshade uses the official SMAA parameters so may I ask why are the screenshots different and why is reshade2 producing Morphological Antialiasing and Reshade 4 isn't?

You know aswell as I that there is no Morphological Antialiasing active in Reshade so there is no SMAA.

The only SMAA injector which works 50% is Reshade 2.0 K-PUTT version, so if anyone is reading this and wants half decent SMAA than I would recommend that version for Windows 8.1/10. Also please reply here your view on Reshade 4.2.1 SMAA compared to K-PUTT. We need to get Crosire to analyze these methods to fix the issue, without more posts I doubt anything will be fixed here, what a god damn shame.

Makes me want to just move back to Windows 7 bloody hell.

Thankyou for your time and hope to speak soon.

The amount of misinformation and halfknowledge flying around here is concerning. Let's clear that up a bit.

It's not "my" SMAA. The SMAA effect in the ReShade shaders repository is 1to1 copy of the original one published by its creators. I encourage you to do a text comparison:
ReShade: github.com/crosire/reshade-shaders/blob/master/Shaders/SMAA.fxh
Original: github.com/iryoku/smaa/blob/master/SMAA.hlsl
I have no rights or major involvement to it.
As such I'm not defending anything. Just presenting facts that I gathered via my own test. If there is a bug somewhere, then that needs to be fixed. But so far I was not able to reproduce any wrong behavior.

How can I say there is nothing special about Boulotaur2024's SMAA? 1) I have source code to it. 2) No, he did NOT "make" SMAA support DX11. You seem to have a great misconception about what SMAA is and where it came from. SMAA was developed by various people at Universidad de Zaragoza and Crytek. The shader they provide supports DX9, DX10, DX11, OpenGL and basically any other graphics API as long as a couple of interfaces are defined. It is mainly a concept for game developers to implement in their engines. But they also provide a demo application which shows how such an implementation of the entire SMAA pipeline would look like in DX10 (it is on GitHub: github.com/iryoku/smaa/tree/master/Demo/DX10 ). From an API point of view DX10 and DX11 are almost identical. And there was another injector for SMAA which supported DX11 way before Boulotaur2024's albeit great project: it was called InjectSMAA, was developed by mrhaandi and was the initial injector that SweetFX ran on. Unfortunately it was missing a completely unrelated DLL export that was introduced in Windows 8, so it caused issues there. Functionality-wise nothing would have changed. There are no gods. Just people who have done great work to make an existing AA solution which was developed by another set of people available to a bigger audience. Neither Boulotaur2024 nor mrhaandi nor I did reinvent SMAA.

To clear up what MLAA means, since that was casually thrown around: Morphological antialiasing is the process of finding specific patterns in the image and then blending pixels at these patterns to reduce the aliasing effect. And yes, this does in fact work in ReShade. The SMAA effect has a "Debug Output" setting. Set that to "View weights" and you'll loosly said get an image of all the pixels that were added at the edges to "fill" out the aliasing. If that is not the case, please show a screenshot with that debug setting active.

The screenshots I provided are easy to reproduce. I did nothing special. Just download the DX10 SMAA demo application from the link above (Precompiled Demo, here is a direct link for the lazy: www.iryoku.com/smaa/downloads/SMAA-DX10-v2.8.exe ), download latest ReShade and install it to that executable. Then launch the demo and you'll be able to toggle the integrated SMAA implementation on and off with a checkbox on the right side of the screen. Next open the ReShade UI and toggle the SMAA technique there (while the integrated SMAA of the demo is off obviously). You'll notice that it looks more or less the same. If that is not the case, please provide screenshots of that demo to show that.
As shown in above post it works for me and as such I have no clues to start investigating with.

So far the only thing I have seen here are a bunch of context-free screenshots of random games with distracting color schemes and no clear relation. If you want to bring across a problem please show before/after shots for image comparison, post your settings, verify you are indeed running the latest shaders from yesterday (use the setup tool to update). Anything that is not just subjective opinions and words, which is not useful to figure out a potential issue.

I want to help! But I can't do that with incomplete information.

XRJ44 calm down. Understand that Crosire wrote Reshade which compiles and injects shader effects, but the shaders themselves are written by various authors, and as he noted SMAA specifically has a long heritage and has been implemented at different times by different people.

@Crosire I'll do some tests this weekend with/without debug across a few titles and injector versions. I have all Reshades from 3-current, 0.18/0.19, 1.0/1.1, 2.03f, SweetFX 1.4/1.5.1 and even Boulotaur's injector I'll confirm if there is a difference, and if there is one where it happened (with what ever shader sources I have).

From my own testing the SMAA shader does have it's limits and felt like it covered aliasing better from before though games have also changed quite a bit and there's all sorts of different types of aliasing too so in newer titles this would probably hit certain limitations unlike less shader intensive and graphically complex titles though I have wanted to do a better comparison but never really gotten around to do it.

Still has a visible effect though, even with TAA and downsampling toggling on SMAA produces less micro-aliasing on edges or how to call it though TAA is best disabled or you'll get crawling and shimmering when these shaders are run over the screen space like this or well post-process effects I suppose.

There were also various adjustments for the default ReShade SMAA shader fixing and adjusting the depth buffer detection in some ways but I think this primarily affects when using predication allowing for both depth and color or luma to work together.
(On that note setting SMAA to depth buffer instead of color and luma certainly smooths out the depth buffer if using the depth shader but it doesn't seem to do anything in the game itself and it works much better to just use color or luma instead for this setting though with reversed, flipped and logarithmic into effect that might just be a consequence of these changes in newer games too.)

Interesting read-up too on how all this works, though there's not much I can do to actually contribute or help out. Not that good with shader and coding or programming but this has been really interesting to read about.

Hello Crosire and thankyou for the detailed reply.

I do understand, we have followed SMAA for over 5 years, I hand on heart honestly thought that Boulotaur created 'x64' DX11 compatible shaders? The reason is because back in around 2014 I couldn't for the life of me activate any SweetFX in x64 DX11 applications until we tested out Boulotaur2024's version. I understand Boulotaur did not create DX11 SMAA, but the shaders worked 100% in 64bit OS systems, which other shader injectors did not work at that time, well I couldn't get them to work. Many forum users said the same also.

This was the only x64 DX11 compatible shader at that time which worked flawless.

Boulotaur2024.Injector_CeeJay.DK's.1.5.Shader+ChromaticAberrationShader

sfx.thelazy.net/downloads/

I'm sorry Crosire for the last post please forgive me, you have created a world class shader application here, its top of the line, I am eternally grateful for Reshade I have to say this, thankyou for not chopping off my head. Months ago I did reformat the unit with Win7, 8.1 and 10 and went thought every Sweetfx, Reshade and Gemfx all the others. I should have wrote this down.

WINDOWS 7 64BIT.
WORKS CORRECTLY
(32BIT) - Reshade beta versions / Reshade 1 to 2 / Reshade 2.0 KPUTT / MasterFX
(64BIT) - Reshade beta versions / Reshade 1 to 2 / Reshade 2.0 KPUTT / Boulotaur2024 1.5.1 Chromatic Aberration Shader

WINDOWS 8.1
WORKS CORRECTLY
(32BIT) - MasterFX
(64BIT) - Reshade 2.0 KPUTT (50% SMAA)

WINDOWS 10
WORKS CORRECTLY
(32BIT) - MasterFX
(64BIT) - Reshade 2.0 KPUTT (50% SMAA)

Unfortunately every version of Reshade 3 to 4.2 at the time of testing did not seem to implement SMAA on any 3 of the operating systems.

I have the old screenshots of Dying Light on Windows 7 64bit with the Boulotaur2024 shader applied, let me start up Dying light with Reshade 4.2.1, I will apply the same settings and find the spots ingame again to take a screenshot in the same locations, I'll try my best anyway.

Crosire may I ask do you have a spare Windows 7 64bit unit? If you do have one, can I ask if you could download Boulotaur2024 1.5.1 shader and you can see for your self the differences between this and the current Reshade. Screenshots are good but mouse/camera movement makes a difference also I have noticed with different versions.

Thankyou very much for your time and hope to speak soon

First time I have seem Crosire write so much and the ending of his post made me feel a bit sad.

It's becoming so common to see people asking through rants more and more that I fear Crosire might get burned out of ReShade pretty soon saying that it is open source and all, with the shaders all there and the new GUI... ):

So I have been trying to get SMAA working in windows 10 In DX11 Game Vermintide 2. I realized it was not applying AA to the final image.

I have been trying to find what causing this issue in this shader. But, am not familiar with it.



I can only make a guess that it has some to do with the Weight Calculation not getting passed to Blend to the final output. But, to be honest, I am not sure.
I also tested in One other game A:MfP Same issue here. I am not sure what's going on.

Also using ReShade 4.0.2 the one EAC allows in this game. But, the same issue also shows up in the most modern version as well.

pastebin.com/fuDpGFSW

I cannot reproduce that. These are my blending weights right after starting a game in Warhammer: Vermintide 2 on Windows 10 with 4.0.2, latest shaders and default SMAA settings (SMAA is visibly applied to the image too):

Even depth edge detection works:

It may be my system then since I am using Windows Insider Build 18351.19h1_release.190301-1611

I will revert to the release version of Windows 10 and retest. Since it also causing issues with other applications on my system.

i dont understand why your still on about this after months, why dont you take the images between the shader versions and injectors and do a pixel count on different api's and on a select few games to test your theory rather then coming on here telling someone who has nothing to do with how the shader interacts with a game, he just made a program to inject it into the pipeline.

@Crosire.

Ok. SMAA. I said I'd do a comparative test two weeks ago and here it is.

Questions we want to answer: Is SMAA working as intended? Has it changed since earlier versions of SweetFX/Reshade?

---

1) Firstly -- thank you for updating sensible defaults in the latest SMAA on github. In my testing there is no benefit past 24 max search steps let alone 98, so 32 is plenty without incurring performance penalties at higher steps for no benefit.

I do think Ceejay's recommendation of Edge Detection Threshold of 0.10, or 0.08 at most is a good idea (rather than the 0.05 currently set as default) as again in my (brief) testing anything less than this (including 0.05) picks up details that shouldn't anti-aliased.

That said, I noticed Marty's depth detection changes aren't currently included, which leads to:

---

2) So secondly -- depth detection for predication is broken in the current SMAA.fx. It doesn't read from Reshade's depth detection, which Marty fixed here:
reshade.me/forum/shader-presentation/187...h-il?start=600#26984

However that was on an older SMAA.fx, so I integrated Marty's changes to the current SMAA.fx on github. I don't have an account so I'll just paste it here for you to review/include:

pastebin.com/KRjDzFgG

This does now allow SMAA to see the depth buffer properly, however more on this below

---

3) And lastly, TESTING!

The below was done with current SMAA.fx on Github for Reshade 4.2.1, no predication enabled. All other SMAA versions are those included with the Reshade/SweetFX as at the time.

Save games were made where upon loading a static scene could be observed for comaprison.

Versions tested:

• SweetFX 1.5.1 with Boulotaur injector
• Reshade 0.19.1
• Reshade 2.03f1
• Reshade 4.2.1

Games tested:

• Firewatch
• Elex
• Transmission 120
• Metro 2033

Some notes here: I wanted to test games old and new. Especially a Source game -- Transmission 120 is Source 2. Elex and Firewatch a relatively recent, and Metro 2033 can be considered a little old now.

SMAA settings:

• Threshold: 0.10
• Max Steps: 64
• Diagonal: 16
• Rounding: 0

So what did I find?

TL;DR: SMAA's implementation hasn't changed throughout these versions as speculated. Well, mostly anyway.

There -are- in fact some differences when you look closely at the pixel level. In fact, even the compressed PNG file sizes differed between some of the versions above, indicating different volumes of data being stored. But they are so fine you have to zoom to see, and you wouldn't see them during gameplay. Though I captured 24 images (I included no AA and FXAA in comparisons), I'll include just one zoomed sample here from Transmissions 120 with clear lines on a box:

EDIT: *** apparently I have too many links in the message... will post following.

Other observations:

• Almost always FXAA provided superior anti-aliasing over SMAA at any settings. It does, of course, also tend to blur the image.
• SMAA, quite often, actually introduces artifacts making some edges *worse*. This is easily seen in Firewatch and Elex for example. This is regardless of settings.
• Predication appears broken, even with the depth buffer working (more on this below).

EDIT: I didn't test the MasterFX version that XRJ44 mentioned, and that he notes a difference on. Anyone have a link?

---

4) Lets talk predication
In theory this should improve edge detection for anti-aliasing by finding edges using the depth buffer and combining with luma or color detection modes.

After the above SMAA testing I ran a round of tests with predication enabled, first checking that SMAA was indeed seeing the depth buffer and comparing scenes with predication enabled and disabled.

Firstly, even if no depth buffer is available, just enabling Predication will alter the edges that SMAA sees -- causing the edge detection buffer to light up like a christmas tree, until you alter the scale. But It's altering SMAA's detection even if a depth buffer isn't available, which it probably shouldn't be doing.

If a depth buffer is available and you do set it up and select 'Depth' as the edge detection method, you get clear edges picked up via the depth buffer. And out of all the methods I tested, this actually delivers the best anti-aliasing for solid lines -- as good as FXAA without introducing artifacts. Of course, it's very limited in that the depth buffer only picks up edges defined by the depth buffer, hence luma and color are usually better options -- though I'm unclear why depth edge detection can produce such smooth anti-aliased lines and color edge detection can't in the same image.

In theory predication is supposed to work -with- the luma and color methods to improve detection and find more edges, but in practice I can't see data from the depth buffer being used in conjunction with luma or color when predication is enabled -- you have to select Depth mode for any depth data to be used. So Predication, at the moment, is I think quite broken unfortunately.

Transmission 120 comparisons:
(to really compare these you'll want to download and compare while zoomed)
No AA:

Reshade 4.2.1 FXAA qual=39

[Reshade 4.2.1 SMAA (settings as above)

Reshade 0.19.1 SMAA

Reshade 2.03f1 SMAA

SweetFX 1.5.1 with Boulotaur