IRAY Photorealism?

aaráribel caađo

I'd love to see that Mei Lin render with a less crazy camera lens setting. 

I'm sure eventually AI will make a lot of this irrelevant, although we'll still need to art direct it, and that might be by creating renders we hand over to an AI to "photorealize"

charles

Photorealism. 1st glance. double take. full on stare. Not sure how to get past the meaning of this except I feel there is different levels of breaching the valley. Some better than others. Sometimes I'm convinced at first glance but looking a second time something seems off. Taking then an experienced look start to find the flaws. Often in scenes I find the geometry objects are huge giveaways and can often be fixed with subdivision and smoothing. Usually the biggest offender to failing first glance is skin shine/gloss. Understanding and tweaking duel spec seems to be pretty much key these days. What are the things that first alert you to not being real?

charles

takezo_3001 said:

charles said:

This is kinda technical, so if anyone wants to know how I got topographic geometry from a pure single HDRI map I'll post a tutorial.

Yes, over here, I would love a text-based tutorial! 

I did a couplew of posts down but will make a step by step video when I get back from Maui.

lilweep

aaráribel caađo said:

blue said:At that scale, third one still looks most real to me because the pose is very natural, morph is natural, albeit super model-y, and most of all the level of detail in the hair simply cannot be matched.  

If we look at a photo taken on a potato camera, we still see it as photoreal because what the camera captured was real life which exists at like 30,490,305,343 pixels and is the culmination of very small subatomic detail.  Yes the camera will not capture every detail, but what it captures is still because of that detail.  So imo shouldnt your render immitate that detail even if you dont think you will notice it?  Anyone who tells you that you that you dont need to turn on normal maps (/HD morph) if you're zoomed out or use high-geometry hair is probably lying to you.  

Ha. I should have cropped out all of the hair, because none of the Daz hair I'm aware of get super close to photoreal. But I'd say the third one has all the details removed. I wonder if there's a bias that comes from persuing super detailed renders that leads your eye to look for things you might not otherwise. I just learned that when we learn to read, the reading-related shape recognition highjacks the parts of our left brain that go into facial recognition, so that literate people only use their right brain for facial recognition (and are worse at it as a result). It could be as you train your brain to see details, you start perceiving things differently than people who don't train themselves. Kind of like how an X-ray or ultasound professional can spot things non-pros can't. 

I'm going to quize some friends who don't do 3D art and see if they think which, if any, of the images are 3D. 

It's probably more the opposite.  People who are more used to looking at 3D images will be acclimated to 3D people and have a narrower uncanny valley while normal people will just instinctively know something isnt real and it falls into their giant chasm of an uncanny valley.

aaráribel caađo

Obvious gaffs like interpenetration, misplaced shadows, etc., often give things away, but I think it may be skin. That said, I got back a few responses from my friends on the trilogy of images I posted a few posts back (and right here, for convenience), and there wasn't a concensus, although 2 of 3 picked the Oh render of Vicky as a photo. 

(I told them 1 or 2 was CGI, and the other 1 or 2 was photo).

Friend 1 (a painter)

That’s hard to say, because there’s no clear focal point and everything is low grade pixels.   1 & 3 have have very unfinished look.  1 has no reflection in the eyes and the pupal is all black.  The seconds is the only one in focus but it is also very symmetrical   It has a look of graphics and focus/ or black ground blur has me guessing.    3 just comes across as middle of the road.   Not enough details to go off of.   Not overly pretty (2), or quirky realism of (1).   So my total guess is 3.  Who knows 

 

Friend 2 (librarian with no art background)

1&3 are CGI, 2 is photo
Although, 2 has that weird line below the hair line 
1 looks intuitively the realest 

 

Friend 3 (ex-professional photographer)

Left: 2
Middle 3 (more photo)
Right 1 (more fake)

There was also no any concensus as to how to reply to my question!

 

lilweep

davidtriune said:

Awhile ago I posted mei lin 8. I still think her textures are amazing. Here is a raw render in blender, the only thing i changed for her texture was the brightness and saturation.

Her nose in particular looks very good.

There are parts where it doesnt look so good though - that said, the information is there so you can probably make it look good.

Gordig

aaráribel caađo said:

Friend 2 (librarian with no art background)

1&3 are CGI, 2 is photo
Although, 2 has that weird line below the hair line 
1 looks intuitively the realest 

1 looks the realest, but 2 is a photo? That's an odd conclusion to reach. It seems like the one that looks like a photo would, by definition, be the most real.

aaráribel caađo

Gordig said:

aaráribel caađo said:

Friend 2 (librarian with no art background)

1&3 are CGI, 2 is photo
Although, 2 has that weird line below the hair line 
1 looks intuitively the realest 

1 looks the realest, but 2 is a photo? That's an odd conclusion to reach. It seems like the one that looks like a photo would, by definition, be the most real.

I suspect that's the way most people assess things—their impression, then some analys if they think it's called for. 

j cade

aaráribel caađo said:

blue said:

I think this is the problem with someone with super smooth, mathematically symmetric features, and clean textures, is how to insert detail into it to make look realistic.

I'm a little skeptical of the "too perfect for realism" school of thought. We're so used to manipulated images of people, where irregularities and "blemishes" are removed, are we really seeing 3D as fake because of that? Obviously, we're seeing something, and that something is probably subtle, so it could be there's a certain kind of 3D perfection (or simplification) that reliably clues our brain to the fakery.

This picture is obviously Photoshopped to the point where we don't see it as representing the woman. But I, at least, don't think it's a 3D render. It looks like a heavily manipulated photo. Masterstroke's image is better than this and has more skin detail, but looks 3D. My first thought is that there are subtle facial geometry issues at play, but given that since the dawn of photography, retouchers have been "fixing" facial features, I'm not convinced. Even with apps that cartoonize photos with huge eyes, it usually looks different than a 3D render given the same treatment.

To be honest, i'm baffled by it. 

Firtstly that is some terrifying photoshopping work. 

 

my personal theory is that its a prepronderance of elements that sells realness. So in that image, while the skin has been photoblasted into horrifying unreality, there are still a bunch of elements that do match reality so we still go "this is something real that has been photoshopped." The hair is an obvious element, but also the lighting, the eyes, and also the larger facial structure

 

 

with that said, If you told me this was a cg face blended with real hair, I'd probably believe it.

aaráribel caađo

j cade said:

my personal theory is that its a prepronderance of elements that sells realness. So in that image, while the skin has been photoblasted into horrifying unreality, there are still a bunch of elements that do match reality so we still go "this is something real that has been photoshopped." The hair is an obvious element, but also the lighting, the eyes, and also the larger facial structure

That's a compelling theory. 

 

Masterstroke

Gordig said:

aaráribel caađo said:

Friend 2 (librarian with no art background)

1&3 are CGI, 2 is photo
Although, 2 has that weird line below the hair line 
1 looks intuitively the realest 

1 looks the realest, but 2 is a photo? That's an odd conclusion to reach. It seems like the one that looks like a photo would, by definition, be the most real.

about shoulder position in nr.1, I just focused on the skin, environment and render setting. Her shoulders are default G8F pose. that's why ;-)

davidtriune

aaráribel caađo said:

Obvious gaffs like interpenetration, misplaced shadows, etc., often give things away, but I think it may be skin. That said, I got back a few responses from my friends on the trilogy of images I posted a few posts back (and right here, for convenience), and there wasn't a concensus, although 2 of 3 picked the Oh render of Vicky as a photo. 

(I told them 1 or 2 was CGI, and the other 1 or 2 was photo).

Friend 1 (a painter)

That’s hard to say, because there’s no clear focal point and everything is low grade pixels.   1 & 3 have have very unfinished look.  1 has no reflection in the eyes and the pupal is all black.  The seconds is the only one in focus but it is also very symmetrical   It has a look of graphics and focus/ or black ground blur has me guessing.    3 just comes across as middle of the road.   Not enough details to go off of.   Not overly pretty (2), or quirky realism of (1).   So my total guess is 3.  Who knows 

 

Friend 2 (librarian with no art background)

1&3 are CGI, 2 is photo
Although, 2 has that weird line below the hair line 
1 looks intuitively the realest 

 

Friend 3 (ex-professional photographer)

Left: 2
Middle 3 (more photo)
Right 1 (more fake)

There was also no any concensus as to how to reply to my question!

This research survey is very interesting. I think we should have forum polls. See how far spread out we are.

 I agree with the painter's opinions but I think 1&2 mainly look fake because they are staring straight forward, no one does that. 3 actually looks at you but she's missing some crucial anatomical elements, namely the nasolabial lines and maybe some infraorbital lines. If you photoshop those in there she might look real.

Masterstroke

blue said:

...

really? i dont think that's the biggest give away at all. shoulder strap isnt even hovering, and necklace could possibly be hovering because she's clearly shifting it's weight against gravity.  Yes, it wouldnt look like that, so is not realistic, but it's not the flaws my eyes are drawn to.

The eyes are the biggest offender in my opinion. In general the lines of her face feel too sharp to me.  I dont think the twiggy eye-liner and false top lashes really help with that.

I think this is the problem with someone with super smooth, mathematically symmetric features, and clean textures, is how to insert detail into it to make look realistic.

Also kinda feel like the hand pose is awkward when looking at that in isolation.

Thanks for editing

charles

Middle has the most convincing proportions and details minus the haircap. Maybe viewing it in smooth shaded to tweak it out and check if the subD is the same as the model would help. 1s proportions and gloss just seem off and 3rd as mentioned details are washed out and proportions exaggerated.

nonesuch00

j cade said:

aaráribel caađo said:

blue said:

I think this is the problem with someone with super smooth, mathematically symmetric features, and clean textures, is how to insert detail into it to make look realistic.

I'm a little skeptical of the "too perfect for realism" school of thought. We're so used to manipulated images of people, where irregularities and "blemishes" are removed, are we really seeing 3D as fake because of that? Obviously, we're seeing something, and that something is probably subtle, so it could be there's a certain kind of 3D perfection (or simplification) that reliably clues our brain to the fakery.

This picture is obviously Photoshopped to the point where we don't see it as representing the woman. But I, at least, don't think it's a 3D render. It looks like a heavily manipulated photo. Masterstroke's image is better than this and has more skin detail, but looks 3D. My first thought is that there are subtle facial geometry issues at play, but given that since the dawn of photography, retouchers have been "fixing" facial features, I'm not convinced. Even with apps that cartoonize photos with huge eyes, it usually looks different than a 3D render given the same treatment.

To be honest, i'm baffled by it. 

Firtstly that is some terrifying photoshopping work. 

 

my personal theory is that its a prepronderance of elements that sells realness. So in that image, while the skin has been photoblasted into horrifying unreality, there are still a bunch of elements that do match reality so we still go "this is something real that has been photoshopped." The hair is an obvious element, but also the lighting, the eyes, and also the larger facial structure

 

 

with that said, If you told me this was a cg face blended with real hair, I'd probably believe it.

What? I don't that's how we all looked when we just got out of bed in the morning?

UncannyValet

charles said:

Middle has the most convincing proportions and details minus the haircap. Maybe viewing it in smooth shaded to tweak it out and check if the subD is the same as the model would help. 1s proportions and gloss just seem off and 3rd as mentioned details are washed out and proportions exaggerated.

i can only guess what you are referring to.

aaráribel caađo

nonesuch00 said:

What? I don't that's how we all looked when we just got out of bed in the morning?

Dang, I was hoping nobody would figure out this was a selfie

charles

blue said:

charles said:

Middle has the most convincing proportions and details minus the haircap. Maybe viewing it in smooth shaded to tweak it out and check if the subD is the same as the model would help. 1s proportions and gloss just seem off and 3rd as mentioned details are washed out and proportions exaggerated.

i can only guess what you are referring to.

In smooth shaded view it is a lot easier to see how grafts like skull caps for hair intersect with the character model. When you get cracked lines in the render around face that is the cap intersecting too unevenly (as well as high displacements) and if you switch to smooth you will see the cracks appear almost exactly where the cap touches or crosses in and and out of the character. Your character subdivision should be around 3 to 5 but often hair caps default at 1 so there isn't enough poly density to match. but graft subd even when bumped up to match that of the character doesn't always line up properly and one needs to expand it a tad so it offsets outwards. some will say scale down to offset inwards but then you might as well just remove the cap...which is in option with some hair. but others it looks like the character needs Rogaine.

j cade

charles said:

blue said:

charles said:

Middle has the most convincing proportions and details minus the haircap. Maybe viewing it in smooth shaded to tweak it out and check if the subD is the same as the model would help. 1s proportions and gloss just seem off and 3rd as mentioned details are washed out and proportions exaggerated.

i can only guess what you are referring to.

In smooth shaded view it is a lot easier to see how grafts like skull caps for hair intersect with the character model. When you get cracked lines in the render around face that is the cap intersecting too unevenly (as well as high displacements) and if you switch to smooth you will see the cracks appear almost exactly where the cap touches or crosses in and and out of the character. Your character subdivision should be around 3 to 5 but often hair caps default at 1 so there isn't enough poly density to match. but graft subd even when bumped up to match that of the character doesn't always line up properly and one needs to expand it a tad so it offsets outwards. some will say scale down to offset inwards but then you might as well just remove the cap...which is in option with some hair. but others it looks like the character needs Rogaine.

Pretty sure the middle one didn't use a hair cap at all though. It was rendered in Maya with strand hair

charles

j cade said:

charles said:

blue said:

charles said:

Middle has the most convincing proportions and details minus the haircap. Maybe viewing it in smooth shaded to tweak it out and check if the subD is the same as the model would help. 1s proportions and gloss just seem off and 3rd as mentioned details are washed out and proportions exaggerated.

i can only guess what you are referring to.

In smooth shaded view it is a lot easier to see how grafts like skull caps for hair intersect with the character model. When you get cracked lines in the render around face that is the cap intersecting too unevenly (as well as high displacements) and if you switch to smooth you will see the cracks appear almost exactly where the cap touches or crosses in and and out of the character. Your character subdivision should be around 3 to 5 but often hair caps default at 1 so there isn't enough poly density to match. but graft subd even when bumped up to match that of the character doesn't always line up properly and one needs to expand it a tad so it offsets outwards. some will say scale down to offset inwards but then you might as well just remove the cap...which is in option with some hair. but others it looks like the character needs Rogaine.

Pretty sure the middle one didn't use a hair cap at all though. It was rendered in Maya with strand hair

Ahh well I agree with the librarian there is something funky going on just below the hairline.

j cade

A recent render,

also a product recommendation for any and all of alfan's expression sets - I generally don't use them at full strength, but in the 40-60 range and then further tweak with the pose controls. But I find alfan's expressions have a lot of character and asymmetry so I find with them as a base my expressions end up looking less stiff and more natural

 

also a test render with a happy accident with refraction.

ignore the skin which doesn't look great, but look at that green light being refracted and then illuminating the iris and catching the hd morph (most of the 8.1 core figures come with hd iris morphs but generally they're not all that noticeable)

Tugpsx

Wow J Cade, Your render look great. The eye shader you have worked on show the eyes nicely.

charles

j cade said:

A recent render,

also a product recommendation for any and all of alfan's expression sets - I generally don't use them at full strength, but in the 40-60 range and then further tweak with the pose controls. But I find alfan's expressions have a lot of character and asymmetry so I find with them as a base my expressions end up looking less stiff and more natural

 

also a test render with a happy accident with refraction.

ignore the skin which doesn't look great, but look at that green light being refracted and then illuminating the iris and catching the hd morph (most of the 8.1 core figures come with hd iris morphs but generally they're not all that noticeable)

Excellent work!

 

aaráribel caađo

This probably doesn't cover much that people following this thread don't know, but here's a nice mainstream review of how Hollywood is making lifelike CGI skin for film.

Masterstroke

This is as close as I can get to realistic skin.
finally found a good combination of maps and skin shaders.
https://www.daz3d.com/chiquita-hd-for-genesis-8-female for texture maps
https://www.daz3d.com/anae-hd-for-genesis-8-female for skin shaders.

Rendered with Spectral Rendering set to Natural 1964
SunSky only environment
Tone mapping defaul, except shutter closure value at 112

charles

Just a few sample closeup renders with 8k real skin scans in Daz on G8.1F, using Uber/IRay not PBR Skin. And some of my coctail of effects, which I scaled down because the skin is just sooo good on it's own.

 

 

 

 

 

aaráribel caađo

Those look amazing, charles. Are the 8k textures specifically for Genesis 8 or did you have to maniputlate them?

charles

aaráribel caađo said:

Those look amazing, charles. Are the 8k textures specifically for Genesis 8 or did you have to maniputlate them?

Thanks! I use 8.1F but use the base_female G8 UV set for body/head as I have my own micrososkin for uber and a layering and tone system developed for those maps that I am not ready to move to the 8.1 map sets. So I did convert some of them from 8.1 to 8. The base skins are done by a guy out of Hungry, I think he maybe partnered with 3D-SK (Who I have been using for micro detailing for a while and have a huge set of reference pics from) but not sure, I need to drop him a PM.  I also find I like the effect from Scatter&Transmit over PBRSkin, it just seems to pop more. I only did closeups because I've been using a male specular map that is creating seams and trying to clean it up now. Not all his stuff is 8k, I think he's just now moving into it with the newer skins and one needs remember to bump up the render compression max or it's lost.

To be honest I don't have the kind of eye you and Cade have with your posts and so am usually passive on those, but have very much been inspired.

charles

LOL I guess I didn't censor that waist image to the satisfaction of the moderator it was just removed.

 

 

 

 

 

charles

So here is a question for those working with the shader mixer. Is it possible to have a second Bump map?

Fore example with the hand and other parts I would like to have my main base map (I don't use normal) and then have another to do finer detail like exagerating the knuckle seams.