Difference in output of save screenshot and viewport render

Hi Omniverse team,

As weird as it sounds , I started to notice there is a difference in the save screenshot(using F10) output and the viewport renders in the sharpness/details and colors/contrast/highlights. The viewport renders always somehow looks better than the screenshot output. in Please see the screenshot when i compare the renders side by side. This render is on path tracer, I don’t know if its noticeable when uploaded here but it is when I compare them on my screen.

Left is the save screenshot and right is the viewport

As far as I know they should be identical except for resolution. When you screenshot the rendering “in app”, its just whatever the resolution you have specified but scaled to fit your viewport (unless you are using viewport mode). If you hit F10, it will actually make a PNG at the exact resolution you specify. So for example 1920x1080. So if you are rendering at say UHD 3840x2160, you are going to get that exact resolution when you hit F10. However if you screenshot it directly if will be scaled down to fit inside the RTX window. Apart from that, they should be intentical.

The only other thing it could be is your RGB Monitor profile. If you screenshot a windows app, it will use your default monitor color profile. If you actually save a file like a PNG, it will use the default RGB color profile of the file format, or actually apply no color profile at all. There may be slightly different, which would affect gamma, and color contrast slightly.

Also if you are using an HDR monitor, and you have enabled HDR on your windows color profile, then taking a screen shot of anything in windows is going to look very different from saving a PNG with 8bit output.

1 Like

Here is a quick test I did using the realtime renderer. But I imagine its the same for path tracing

Hi @Richard3D

In real time the details in textures and sharpness are showing up much better than path tracer, which is puzzling. I do know that path tracer render takes more time to render but atm the path tracer render in omniverse are not showing the time to quality ratio, like at such high number of samples and time taken for a 4k render with denoiser tune down to a blending of 0.2 , it is still not showing the details/sharpness which is rather disappointing when compared to other rendering software like vray which would take like ard 2hrs with much better details/sharpness. Mirror reflection is also one area that need improvement, don’t always shows up well in the render.

Realtime (Translucency for fabrics and non glass material don’t show up well)

Path tracer

What is your GPU hardware ? A scene like this should be only 10 minutes max on a 3090.
The point of RTX is it is very fast. If you are spending more than about 20 mins a frame, you would need to breakdown your scene and workflow. I tried to limit my renders to 500 samples unless required to go more. Most 500 pass renders are 5 mins on my 3090.

So this scene is rendering at 4k. How many samples and how long ? Why are lowering the blend ratio of the desnoiser ? You want more noise / detail ?

I can relate to @DavidDPD when he talks about the PathTracer performance. I am long time V-Ray / Maya user and it is been about a year since I started to use Omniverse. If you actually compare the V-Ray progressive render vs PT without denoiser (because Omniverse denoiser is extremely aggressive), the results are not that impressive for PT, on the top of that it is true that often no matter how many samples you use, it never arrives to the level of detail that V-Ray does.

I will make entire post about it and some comparisons between the V-Ray and PT renders.

1 Like

I also was a very heavy Vray RT user in 3dsmax before OV. Whilst I can appreciate the comments, I do think the key issue is that you both feel that the denoiser is overly strong and reduces fine details ? Is that fair to say ? I would imagine that is why you are turning the blend down. Having said that, with the realtime optix denoiser ON, I have never seen anything as fast as PT for rendering. It can resolve 50% of detail in literal seconds, and 90% of detail in a minute. That last 10% is where I feel you both are talking about. The super fine, non denoised detail. For that I would love to see some examples where you say it is failing.

RTX is improving in leaps and bounds for each version we release. I will pass this along to the RTX team for discussion. I am sure any concerns you both have will be diminished with each update and improvment.

1 Like

I do love that omniverse rendering is fast and able to show like what the scene kinda looks like within seconds which is gd for iteration and which i guess why most of us are even interested in using this in the first place. Of course it isn’t realistic to expect a render to match vray if the time taken is much lesser but if we are comparing the same amount of time taken, most of us would still expect something much closer like 5% difference which sometimes have to do with rendering characteristics and not so much of details and sharpness, so currently there are lots of lost of details and sharpness that even crazy amount of samples and time spend can’t resolved this.

Although I think the sharpness and details in the rendering can be better but it still isn’t that unfixable using a highpass filter in photoshop provided the readability of text isn’t important and when the details are not behind a clear glass.

Now for the issues that I think are quite problematic:

  1. Glass

The main issue is in the details that is behind a clear glass material (of course it doesn’t matter much when its a fluted or frosted glass). Its very noticeable in the screenshot below, these details usually doesn’t even resolved when the samples are in the 3k range or above, it does get better when the opacity is tune down. The reflection/refraction doesn’t seems to accurately show in the glass nor the mirror too, so clearly something is off about the reflection/refraction be it mirror or glass.

  1. Antialiasing/ rendering not resolving

I tested out the AA and it seems at 1 radius the render doesn’t resolve in many spots, so I switch to box sample and reduce the radius(like 0.6-0.8 range) which helps but still some just doesn’t resolve during the render.

This issue is more pronounce on glass material but I have also encounter this on fluted wood panel, sofa etc. The 2560 resolution output fluted glass panel on the right side, , the first panel closest to camera display every lines as it should, on the second panel starts to shows lots of missing and unresolved details (Path tracer display a blur when it can’t resolved it seems). The third panel is just totally blur , no hint of fluted at all. The 4k resolution example shows the first two panels lines display correctly but still the third panel shows many missing lines at high resolution and high sample rates.

Previously the glass panel is a normal map fluted texture, so I tested out physically model 3D fluted glass panel in the last three screenshot. The first is box sample pattern at 1 radius, the second is box sample pattern at 0.68 radius and the last is triangle sample pattern at 1 radius. The panel that is facing the camera feels correct but the bottom parts still have minor unresolved details. While the panels on the right feel way worst than before , moiré and even more unresolved details from second panel onwards.

(Realtime)

(Path tracer - clear glass no opacity 1024 samples) (~20 mins 2560X1440, RTX 3080 10GB)

(Path tracer - Clear glass with 0.7 opacity 1024 samples) (~20 mins 2560X1440, RTX 3080 10GB)

(Path tracer - Fluted glass >4k samples) (>4hrs 3840X2160)

(Antialiasing error )

(box sample pattern radius 1, 1024 samples)

(box sample pattern radius 0.68, 1024 samples)

(triangle sample pattern radius 1. 1024 samples)

We are aware that are denoiser can occlude fine detail from a normal map. It is one of our roadmap improvements. So in this very specific case, yes, you are better off using a real geometric panel that a normal map for sure. The last results look much better. I am sure you are trying to match a specific look with the panel, but from a rendering point of view, what you are asking has always been extremely difficult in any renderer. You are expecting all of those tiny vertical lines to resolve perfectly, even in a fall away perspective. Eventually you will just lose details. If possible, can you not a) increase the size of the fluting to give the renderer more width to work with b) pull away from the right side view of the glass panels in this shot so you are not seeing the issues ?

I understand the fustration. I have done 25 years in arch viz myself using mental ray, iray, vray and octane. But one thing is true… every renderer has a pros and cons. Not one gets in all. Whilst we have the incredible speed and in most sitations, incredible results, there are fringe cases like this where our denoiser is a little agressive, or our anti-moire is not quite as good as a traditional renderer.

One more idea is to use IRAY (Accurate) in this case. This is considered this highest possible quality you can get. It is slow, but there in reserve for very tough rendering challenges. It is amazing with glass.

Having said this, we hope overall you are enjoying RTX. I know that exciting things are coming for RTX this year.

1 Like