If you want to make photo-real images with TG2 then you have to take TG2's strengths and weaknesses into consideration.
Then it's matter of trying to make the best use of the strengths and to avoid situations which TG2 can't cope very well with.
In regard to that, in my opinion, TG2's strengths are:
1) atmospheric model, especially the atmosphere node/shader itself.
Picking the 'right' atmosphere density and haze settings can make or break the sense of scale and will determine how saturation of shading and shadow detail in terrain will look like.
2) procedural displacements for adding detail and shading/texturing to your terrain.
This requires no explanation of course, because this is why we all love TG
There are some aspects though which you should be careful with, later more about that in the TG's weaknesses part.
All in all these strengths offer a lot, especially if you want to make barren scenes without vegetation, like deserts and the like.
Also imho, TG2's weaknesses are:
1) object rendering -> objects nearby look too CG. Mostly due to lack of some advanced kind of shading.
A simple example is Silva3D plants' example renders on their site which are rendered with Vray.
If you reproduce such a close-up in TG2, it's always quite inferior.
Mostly textures don't come out as crisp and clean and also (self)shadows aren't that good looking.
You can go through quite some lengths of getting them to look better by hugely increasing AA together with cubic b-spline filter and also upping the resolution. If you don't up the resolution then the self-shadowing is too dominant and you can't get nicely detailed lighting.
Which leads to the second weak point...
2) GI -> to continue with the objects. In the distance GI is uncapable of correctly sampling the vegetation. This leads to too dark vegetation as if the renderer decides to treat the whole object as one shading result, thereby destroying all the detail.
Similar results and issues are with displaced terrain where GI gives unexpected results over distance.
These 2 phenomena I describe can be quite clearly observed in my DeviantArt gallery:
http://tangled-universe.deviantart.com/gallery/It's difficult to say though whether it's GI which is undersampling or atmosphere or some other process in the renderer which pre-calculates occlusion/surfaces.
There's no documentation about the renderer which gives a detailed explanation how it step by step treats the scene so that you can devise a detailed and especially direct approach on tackling these issues.
So far the only solution is to increase rendersize considerably to ensure that the renderer 'sees' all the geometry of the vegetation and thus applies correct lighting and shading to it.
Increasing GI to 6 along with 512 atmo samples didn't improve it, so my suspicions is that there's some pre-filtering/calculations of surfaces which is limiting.
However, like I said before it's difficult to say without knowing more about the renderer.
Other issues are in-terrain-shadow objects which tend to look plain wrong as soon as you're dealing with either quite displaced surfaces or denser atmospheres than default. Not to speak about a combination of the two. Avoid that at all cost, you'll never get it to look photo-real.
3) noise sampling -> especially in surface shaders the renderer has parameterizations skewed towards high frequency noises which it at AA-stage can't eliminate.
If you design a scene with crop renders of your foreground surface and add subtle brighter speckles to your surface's base then they will look nice.
However, over distance these subtle speckles become dominant and overrule the darker surface base.
So all in all a lot of aspects are involved in considering something photoreal and a lot of that is esthetic and highly specific for each situation and scene, as Oshyan layed out.
However, I think that besides that there are also some technical aspects involved, which I summed up above, which you can keep in mind to stay on the 'photo real road'