Photoreal ~ what makes it?

[Kadri]

...
... and it will definitely be very much appreciated by us.

Who is "us"?

Martin i wrote some things in a little angry mood but than changed my mind ...

This is the "Terragen 2 Discussion" board and i think some posts here are worth to read for others.
Speaking about photorealism or realism etc. is a nice thing and i would be only sorry about that if it stops.
But the problematic parts about the contest (questions general directly to the contest) would be better to be discussed in his own part on the forum as it seems .

And man smileys can be deathly as it seems. Frak !

[gregsandor]

LOL
Actually that photo brings up a very good point, the lighting is almost entirely indirect with a very overcast sky, the shadows are very soft (look under the station wagon).
I've never had much success with overcast scenes, others have.
Is this the actual scene you're trying to replicate?

Richard

Nah, at this point I'm still gathering ideas and thinking this through.  I probably will go for a modern look, but the architecture and even the way the roads are laid out come from an earlier time I want to work that old influence in somehow.  Maybe you're onto something with that flat look, perhaps simulating a modern camera but using old-style framing and lighting (overcast sky, etc) would lend a sense of age and speak subconsciously to the viewer.

[gregsandor]

Simple solution, just run your render through Instagram.  Then you get that "magical postprocessing" look that's so popular these days 

[Dune]

Participants; better read this again, then decide what to do with your renders before submitting: http://www.planetside.co.uk/forums/index.php/topic,15724.0.html

Interesting discussion, by the way.

[gregsandor]

Participants; better read this again, then decide what to do with your renders before submitting:

Yeah, that change was made as a direct slap at my part of the discussion of what constitutes photorealism in rendering in this thread, but the artistic question that Cyphyr asked when starting the thread still stands:  "what makes an image photo-real?"

[Oshyan]

Well, sad to see a bit of crossed wires going on in this thread, but I think we all have similar goals! So let's get back to constructive discussion.

Greg, I think you're trying just a bit too hard to get into the philosophy of this, like trying to define "art". It's a potentially interesting discussion in itself, but probably best for another time.

Basic aspects of landscape realism, quickly defined as per my specific perspective, in rough order of importance to my mind: atmosphere, lighting, practical/geological accuracy, randomness, detail, baseline render quality. Anyone else care to post their own ranking? It does not have to use the same terms as mine, can use more, fewer, totally different ones, whatever. This is just my view, briefly stated. I can argue my perspective if desired, but I'd rather see others post their particular personal summary first.

I do think the discussion about dramatic vs. pedestrian imagery is interesting and potentially relevant. I was going to say that a good approach in scene construction could be to ask yourself "Would I be surprised to see a photograph that shows this?" and if the answer is "Yes", then you have a greater challenge in front of you, if nothing else (i.e. it could still be "photorealistic", but would be harder to make it so). This does ignore the possibility of aiming for dramatic presentation, unusual depictions.

I'll give you a practical example from our user gallery, that I think both highlights the issue here, as well as my points above:
http://planetside.co.uk/images/rapidgallery/slides/f56e3c259ba20200452593a4c6195e88~sauravs-subedi_tgd5.jpg
This is an incredibly simple and, to my eye at least, rather realistic image. The biggest contributors to realism here, for me, are the atmospherics, lighting, and geological accuracy (from DEMs). There is little randomness, not much detail, and the render quality is not even outstanding (or perhaps the image compression is a bit too much, or whatever, hopefully you get my point). This is one particular approach you can take to realism, a somewhat minimalist one, but one that demonstrates the potential power of the "dramatic" presentation. Essentially Saurav used the strengths of the system - atmosphere and lighting - to make up for either TG's own weaknesses in other areas, or the simple challenges of actually building realistic depictions of more detailed elements (e.g. foreground rocks, vegetation, etc.). Of course he was probably also using some reference photos that looked quite like this. The argument can shift in another direction here too, that the choice of subject matter can heavily influence the path you must take to reach realism...

Honestly, it's a very broad topic, but I think I am seeing the pragmatic goal here, and it's not necessarily what I have even been talking much about: the goal, as I understand it, is ultimately to get to some broadly applicable, but specific and useful advice that people can help to improve realism in their renders (improve = make them more realistic)! To which I say, good idea. With that in mind, the above example is perhaps useful if you can consider it as advice to potentially try to avoid or hide elements you are having problems with, but it may also be a rather limiting or defeatist strategy.

So, to be more useful... the question that remains to my mind is, how advanced do we want to start at in our consideration? A good example is the simple but powerful benefit of good lighting choice. It is, I believe, a relatively basic lesson that front-on lighting tends to be very flat, boring, and indeed generally unrealistic-looking. So to my mind that could be lesson #1 (and it even plays into the example I post above, hehe). But is that too basic?

- Oshyan

[cyphyr]

Continuing the discussion about "realism" (I've dropped the "photo" since it seem so contentious).

Randomness Well the world isn't really is it. There are lots of surfaces that appear random but in truth they follow rules, order etc, dirt tends to fill crevices, crevices tend to form along lines of structural weakness, the list goes on. We use random noise (Power Fractals) to simulate this apparent randomness but by necessity this can only go so far. We need to find a way to simulate the "ordered randomness" we see in nature.If I was working in a normal 3D package (Lightwave, Maya, Max etc) then among other things I could render out occlusion maps which could be re-projected (and inverted black for white) onto the scene as a mask for dirt. Something like this might be possible in TG. Texture the entire scene flat white (grey might be better) and light with Ambient Occlusion only (loose default the blue hue and replace with white). You would then have an image that could be re-projected back on to the scene or used in post as a mask to apply some "ordered randomness".
I'd be interested in other ways of "ordering randomness". ("Two to go easy on the mayo")

Variety One population of the same tree really won't cut it, Three is not much better. There was a thread a while back about adding a global scale texture to a population so there was a variety of textures across a single population. The same is true of "Fake Stones" and "Surface layers", basically the more layers the better. Even if an area is predominately one colour/texture there are subtle variations that may be hard to pick out specifically but their absence detracts from the scene.

Human/Natural interface We can use Terragens powerful fractal nature to create to create pseudo realistic worlds, couple that with using something like World Machine and very realistic terrains can be created (see above for how to surface them in an equally realistic manor). We can also use external programs to create road layouts, buildings, human activity evidence etc. How "realistic" this is is down to the skill of the artist, nothing more. But the interface between the entirely natural fractal world and the entirely human ordered world is tricky. The occlusion map trick or something like it might work although I suspect it may be too subtle. I don't have an answer for this one and I'd be interested in what the board has to say on the subject.

Ok time for some more coffee

Cheers

Richard

PS just saw your post Oshyan (aren't you supposed to be out enjoying your self... ) and I agree with all the points you make

[gregsandor]

There is no "Button 9," no magical formula or "Make Art" button.  Unless the artist has considered some of these things, at least subconsciously, his results will be hit or miss, and mostly misses at that.   There are guys who make beautiful paintings and photos and never consciously "know" this kind of stuff but they are rare.  Yesterday I ran across an old thread on cgsociety with some guy crowing about running a test of TG2, he said he cranked all the settings up for "godlike" (his words) quality, and in spite of several advices from experienced TG users telling him that it wouldn't produce "godlike" results, he went on.  You can give out whatever combinations of quality settings, heck, have Matt code in a preset default for great renders, and still you'll see what you see today in the galleries.

I propose that though it is as technically demanding to produce a pedestrian snapshot as an artistically composed work, the "ordinary" shot is more likely to more often fool the eye.   Second, in my experience, even with quality scale models on accurate digital elevation model-based terrains you can occasionally get lucky, but that in itself isn't enough to consistently produce the kind of image we are discussing.  Also you must consider the target audience and its expectations of what constitutes a photograph in creating a picture. 

Randomness Well the world isn't really is it. There are lots of surfaces that appear random but in truth they follow rules, order etc, dirt tends to fill crevices, crevices tend to form along lines of structural weakness, the list goes on. ... render out occlusion maps which could be re-projected (and inverted black for white) onto the scene as a mask for dirt. Something like this might be possible in TG.

You can do exactly this for models, create AO maps (rather something like them) to apply fractal-mixed dust and dirt and grime.  It helps "seat" the model in the local earth.  The same principle works on terrain features too of course.

[attach=1][attach=2][attach=3]

Variety One population of the same tree really won't cut it, Three is not much better. There was a thread a while back about adding a global scale texture to a population so there was a variety of textures across a single population. The same is true of "Fake Stones" and "Surface layers", basically the more layers the better. Even if an area is predominately one colour/texture there are subtle variations that may be hard to pick out specifically but their absence detracts from the scene.

You can texture populations so the color varies by location.

[attach=4][attach=5][attach=6]

[Oshyan]

Richard, you bring up some very good points and potentially useful ideas. The "Ambient occlusion" approach (i.e. "dirt map", I think) is a particularly interesting one, and I wonder if you could not accomplish something like it with some fancy node work. I believe, for example, that you might be able to create a map of how "rough" the displacement is across the scene, and use this to control texturing. That might be a similar idea. It wouldn't work for objects, nor for the intersection of object and terrain, however.

Greg, "target audience" is mainly what I am arguing against being a particularly important part of this discussion. It feels fairly important to me to narrow down the potentially very broad and diffuse conversation to target the key, most impactful ideas, because otherwise it will be easy to get "lost" in discussion and I think the momentum of the thread quickly diminishes (as I would say has already been seen here at times).

I say that "target audience" is less relevant because we know that at the least the people judging the contest will be in this forum, and I would hope we can agree that we have a *general* sense of people's taste and perspective on "realistic" here, obviously with some possible exceptions (people with unique ideas of realism I guess), people with different taste, etc. But let's keep in mind this is less about "I like that as an artistic piece" (on which I would say we'd have a wider variety of opinions) and more about "I would believe that is a photo of a real place", the latter of which I think many more people (here and elsewhere) would agree on. I think even outside this forum the general perception of "realistic" is not hugely controversial, it usually comes down more to how discerning, picky, and critical a person is, and how much knowledge they have about 3D rendering or at least special effects techniques.

So, to use a practical example, it seems quite obvious to me that your postcard image, with the hand-colored, sketched look, is not what the goal is. Even if it might have been considered realistic or "photographic" at one time (it's arguable if that word even meant the same thing at that time, and then we just get into literal semantics, etymology), it's certainly not now, it is quite clearly more of a "retro" look. Perhaps you brought it up merely as an extreme example (in fact I consider this likely), but just because there can be an extreme example does not mean there is not in fact an already existing - even if not specifically definable - consensus. Even if not, I don't think we need to start at the broadest ends of possible definition in order to get to common understanding, rather, I would be more interested in comparing 2 potentially realistic, "modern photographic style" images, but with different subjects or artistic approaches, and getting some perspectives from various people on whether they think each is realistic. If you want to take the "target audience" approach, perhaps starting with something like the Autodesk "Fake or Foto" tests would be useful:
http://area.autodesk.com/fakeorfoto/

In any case I haven't said "you can't think what you think", only "I don't think a discussion of some aspects is the point of this thread". Maybe I'm wrong. I like the direction Richard began taking above, practical, functional ideas and advice. If you feel a discussion of target audience is critical to that, I won't argue the point further.

- Oshyan

[cyphyr]

You can texture populations so the color varies by location.

From the look of the preview screen you have multiple populations each with it's own texture. Or is one population with a texture that varies over it. It's the second option I'm interested in. As memory serves (not very well atm) there was a problem getting the global texture to work with opacity channels whicvh becomes important with complex objects but is less of an isue with simple models like grasses etc.

Oshyan - Blue nodes might do it but I don't think it's just an issue of "roughness". Think of a cliff face meeting a flat horizontal surface, not very rough but definitely the sort of area that would need some sort of dirt map. I do see however that my suggestion of a camera projected inverted AO map is something of a blunt instrument, pretty sure there will be a better and more effective way.


Cheers
Richard

[gregsandor]

The left hand image (red white and blue corn) is one population, masked (thats why you see the spaces between the fields) with one image to color it.  In practice I blend a colormap or orthophoto with the natural leaf textures to produce leaves that have local color variation.  I use the method for grasses, crops, and trees.

As for your AO map, if you want to generate one procedurally you can begin with finding sharp angles in terrain, and probably could do it with objects.  I'm sure I tried it at some point, but the method I use now has worked best and gives me the best control over the result.

Here's the map I used when I was developing and testing the corn shading I think (its dated 3/28/10, so its been a while).
[attach=1]

And here's the crop mask

[attach=2]

[gregtee]

Wow, I missed this thread until today. 

I've spent my whole career trying to achieve photo-real imagery.  It can be incredibly hard to do because ironically the closer you get to photo-real the more unreal it starts to look right up until you cross over the barrier and achieve the result.  This is especially true with CG people.  There's even a phrase for it, called Uncanny Valley.  That would be a cool titled TG render, -one that's almost photoreal.  "Uncanny Valley".   I might have to do that. 

Anyway, to my eye, the paramount thing is lighting.  Everything else is secondary.  If the lighting isn't working, nothing else will save it.  You can have a plain white box in a white room and if lit properly it'll be photoreal.  Things like chromatic aberration, barrel distortion, fringing, DOF and such in my experience are things people do to essentially put lipstick on a pig, assuming they haven't already achieved the photoreal result before adding these secondary filters.   I never let artists pile on that stuff until it's agreed we've taken the image to as close as possible to photoreal in the time we have, and only then can these subtle camera lens artifacts be utilized. 

Once the lighting is working, I'd say that having good models with the appropriate level of surface detail added is next in the chain.  This includes things like beveled edges to catch light, displacements, and enough subdivision to negate faceting, etc.  This is something that bumps me sometimes with the Fake Stones.  I find at times I can see large facets on them in areas that look unnatural, but because the rest of the image often looks so good, I can overlook it to a degree.  It's still there and bugs me though.

Next is good textures and shader work.  Textures that don't reveal pixels, don't repeat, are in the correct color space, and aren't unnatural in their range of tone are important.  Shaders should have the appropriate diffuse, specular/reflection levels with proper fresnel angles set, bump or normal maps to add further detailing, and something that's lacking in TG at this time which would be a good sub surface scattering option.  Certain plants, like succulents for example would be very difficult to do at this time in TG, though most stand alone renderers out there such as Vray support it.  This is a more esoteric point though and doesn't hold TG back from doing photoreal.  It's more of a want than a need. 

After that I'd just say common sense when putting an image together.  Make sure things in your scene are layed out and arranged in a natural way.  Paying attention to detail and logic are important.  This is where artistic flair comes in to play and separates so so renders from truly great ones.  I'd say to people struggling to achieve something photoreal that duplicating something you see in a photograph you like would be a great way to approach the problem.  Frank B did this with great success on his cumulous cloud render.  This approach, using photographic reference and copying it is used all the time in the CG industry and is a great way to force yourself to critically look at things.

Once all this stuff is done and your image is looking like a photograph can one start putting in the lens distortion effects, like flares, DOF, fringing, light wrap, grain, etc.  These are all things that degrade the image to varying degrees, which is why artists tend to want to use them right away when their image isn't looking photoreal yet and they start to lose patience or run out of ideas to otherwise take it the image where it needs to be before putting on those final touches.  I had an artist once who used to put a heavy diffusion on all his renders because in his mind, that made it photoreal, and it wasn't helping when other artists who were easily persuaded by that kind of hocus pocus would ooohh and aaah when they saw it. 

Anyway, my .02 on the issue. 

-greg

[chris_x422]

Hi guys, this is a great topic and I'll try to chip in with a few tips and hope they might be a little useful.

Oshyan's tips on lighting and atmospherics are absolutely spot on, and also highlight a lot of the core strengths of the terragen renderer.
In most other 3d apps, depth and haze have almost always been achieved in post using the zdepth pass, this always frustrated me as it a very linear method of producing depth, and lacks realism as well as the lights relationship to the atmosphere.
The ability to layer atmospherics in terragen can produce far more realistic and natural results, it can even be used to simulate dust and random particulates in the air, so in that sense I often like to employ several isolated cloud layers to create different effects at different depths in my scenes.

It's also good to know the limits of your render engine and discover ways of circumventing the issues, for instance, the lack of shadow and colour detail in gi areas. Before the advent of better light simulation in apps like arnold and vray, we always used carefully placed bounce and fill lights to compensate, and while it can take some time, trial end error to get good results, it's still a very valid technique. I'd say that in most situations it's very tough to get away with simply one light and gi, and it's worth taking the time to experiment with lighting techniques, placement and even switches on the lights, ie using lights that only provide one or two forms of contribution like specularity/ diffuse/ atmospheric contribution. One thing we almost always employ in lighting is the use of our shaded spheres, 50% diffuse, fully reflective, and glossy etc as well as macbeth charts, these can be really helpful to have in your scene when lighting, as they allow you easily cross reference and check against any kind of reference plate, as to weather you are providing the right kind of values to your lights.

In terms of realism, shading is also crucial, and requires much study of your subject, how does the light react to the surface in terms of levels of reflectivity, glossiness and diffuse reaction to light, and how does it break up and scatter? going the extra few yards, say with road textures for example, to use maps that break up the specularity / reflections and diffuse contribution, can produce much more realistic results. Another tip here regarding textures, is that we almost always employ a linear workflow (which I'll avoid going into too much detail about here), terragen automatically converts your textures to linear space, which is great as it helps produce a more accurate response to the additional values of light, but you also have to look out for the colour saturation levels and not just the gamma. A lot of colour is reproduced by the addition of light, and more realism can be achieved by using slightly desaturated tonal values that work in conjunction with your light sources.
In all these situations, while trying to achieve specific realistic results, referencing your subject and type of image can't be underestimated, doesn't matter how experienced you are, without close study of relevant reference, it can be very hit and miss if you're looking for realism.

Richard makes a good point about bringing in maps such as AO to help create natural effects, and it's always worth exploring how we can create any kind of rule that lets us get into a specific part of a surface, I often use zbrush for it's multitude of different masking types, which are great for isolating different types of areas.

In terms of photoreal finishing, and matching plates in film, the end result is always done at the composting stage, where all the lens artifacts and specific attributes regarding the cameras used are matched as closely as possible. The matching of the camera types and lenses is very difficult to achieve in pure rendering terms and when looking for those types of results from your renders it is worth looking into how a lot of those effects are recreated at the post stage. Plenty of info out there.

Hope it helps, if just a little.

Chris

Also, +1 to everything Greg just said

[gregtee]

Where do you work Chris if you don't mind my asking.  You obviously know exactly what you're talking about.

-Greg

[cyphyr]

Wow, two great and very informative posts, thanks guys. A lot to take in here but this is exactly the kind in info and discussion I was hoping for in this thread.
By the way I had a quick go at creating camera projected AO maps of a TG scene and using them to apply a grunge texture into crevices and hollows. Not great results as yet, maybe working at higher res and detail will help.
Cheers
Richard