thank you for sharing the screenshots with us. For those of us that are not good at depicing the flight simulator in use just from a screenshot, do you mind editing your post to include the source of the screenshots? (P3D with scenery add-on xy). People that just briefly visit this thread will probably mistake one simulator for the other. Thanks.
And please also mention that this is not default P3D, is it?
Ok, done, understood.
If you look at screenshot S06a and S06b you can see the mentioned mismatch in tree coloring. The trees are so dark that my eyes can only focus on them. Their placement is pretty good though, I don't like the placement of the Aerofly FS 2 trees either.
The opposite is actually the case : it's the ground picture that is too bright in this case, the color of the trees is much more correct.
There are 2 facts to consider:
1) the difficulty of photorealistic scenery is that you have to do with available aerial pictures, the quality of which varies a lot depending on the place. But in the end of the day, you have to assemble a patchwork of photos and find a colorimetric compromise that fits for the whole region.
IPACS was able to achieve a very good ground texture on covered places, but it won't be possible everywhere I'm afraid. Screen S05a shows a burned texture and that's it with aerial photos, until a new release of aerial pictures gets available.
2) the add-on scenery was designed for FSX, and P3D definitely gives different colours and the match between autogen and ground texture is less good, but it's actually the ground texture that doesn't display anymore in P3D exactly the way it does in FSX. For technical reasons I had to switch to P3D and this difference is acceptable to me. The biggest drawback IMO is rather the hard coded LOD-RADIUS in P3D that causes the ground to blur at mid range, but that's another topic.
Regarding point 1) we all cook with water. Regarding point 2) I expect the so far excellent ground texture quality in AeroflyFS 2 to allow an optimal color match, while the brilliant graphical engine allows more distant sharp display.
S09 to me does not look as repetative to me. Sure if I wanted to I can find repeating textures but I don't think it pops into the eye as quickly. The colors of the textures blend in nicely with the colors of the missing buildings on the ground textures and there is no larger spot on the screen with the exact same colors or textures. Its randomized with enough variation, compared to S05b where there a large orange spot on the middle right of the screen, due to the same roof texture. On the screenshot S05a (plain, no auto-gen) you can also see some tiny trees on the ground texture near the wing tip and a lot of variation in tree density on the middle left in that small forest. With autogen on (S05b) all that detail information is replaced by a very boring looking patches of trees (all same color and height).
In S09 the colors blend in nicely at noon time because the ground picture is actually quite repetitive, bright white/light grey flat roofs. On the other hand, if you set morning or late afternoon local time the few sprayed buildings will pop out due to their direct lighted walls, drawing the attention to the missing buildings...
In S05b you actually have dozen of different roof and wall textures. If necessary you could improve the patchwork effect by increasing textures variety and contrast range, it doesn't change anything to the technique. It's only a question of fine tuning.
Regarding the trees in S05a/b, it is a typical but accurate effect with photos and database. The ground photo is several years older than the fresh trees database, and this is a patch of planted trees. On the photo these were very young trees, while they're now grown up in the database. They're all approximately the same size because they're all the same kind and age and must be more or less the same size in reel life. That's pretty accurate actually.
If we want to implement auto-gen it has to be better than that. We need trees in variying sizes and types (a lot more than on these screenshots), their placing should be at least as good as shown on these images. And they need to blend in a lot better. Detail information should be preserved and the result should look photo realistic, not cartoonish. And to me photorealism is only achieved when the entire screen looks good. Otherwise I drop out of the immersion right away.
I find very positive to have high expectations and aim at high immersion quality, that's what I'm (and many others are for sure) searching for.
But you need to start with something.
Manual placement is a dead end when it comes to representing a flight region.
It's ok for micro sceneries à la Orbx, it's very nice, but it definitely doesn't make a flight region and you cannot cover a flight zone with such techniques.
Most add-on airports available for FSX/P3D/XPlane don't even feature the necessary ground details for flying the published traffic pattern because they concentrate on a tiny range. It's lovely to make a few snapshots at ground, but doesn't bring much when it comes to flying if they're not integrated in a convincing, flyable region with homogeneous quality.
We used to do manual autogen placement back in the times of FS9 when there was no choice, but it's monk work for a result definitely worse than what we achieve today extracting databases. We could hardly display 10% of what is currently achieved in FSX, and God knows how limited we are by the old graphical engine...
Look at NYC you would spend years manually placing buildings and trees over the whole area, without any payback chance and a significant risk of overloading the graphical engine, without necessarily better result than procedural 3D. You would just waste your energy.
Yes we all want more detail and the 3D is very appealing to our eyes. Auto-gen can create that sense of highly detailed scenery around the globe at a low cost compared to modelling the whole world. All I am asking is that we should not stop in the uncanny valley but give the extra 15% to 20% of effort to get it looking photorealistic,
I fully agree, but the first step is to have the tools enabling to generate mesh, ground photo and autogen/procedural 3D in AeroflyFS and figure out how the graphical engine can cope with objects density, what are the winning combinations to achieve the best visual quality without loosing game smoothness / fluidity.
Textures variety is a matter of fine tuning afterwards, even if it's definitely not limited to 15-20% of the job.
As usual, 80% of the job is done with 20% of effort. The remaining 20% need 80% of effort.
Third party editors are willing to step in, because it is obvious AeroflyFS has a strong potential, but the current SDK doesn't allow to generate even a patch of ground and it's not yet obvious it will in the future.