Yes, NVIDIA has great tools for that, I won't deny that by any means. In the meantime, a lot can be solved with AI, also as an aid to understanding processes.
Here, however, we have a completely different problem: 

What you are proposing is more or less a physically based world model that represents all processes in the real world without exception.

This is necessary, because it is not possible to determine all parameters, which have an influence on the appearance, in real not world-wide (or at all?). Hence the "world model".
By the way, this is also what climate science is trying to do - but much slimmed down, because the computing times are too long. And here "only" the climate is calculated - with uncertainty factors. And we are talking about supercomputers calculating for weeks and months.
Now we would have to increase this to the extent that we do not only want to output simple 2D/3D arrays, raster data with a resolut of 50 km/pixel, but a 3-dimensional image, in which every ray of light is simulated correctly, with an insane resolution. And not only one picture, we want to have 60 pictures per second or even more!
Here we are: Even a cluster of the most powerful computers in the world would give up.

Therefore my statement: This is not possible (yet).

Edited September 15, 20232 yr by MatthiasKNU

We'll see, AI is for initial data translation/recognition to the world being generated, it's not predictive in the same sense as climate processing which does indeed consume considerable processing resources.  We're also starting to see usable quantum computing services that do all the work extremely quickly and provide the results.

Much like BlackShark.Ai, except you don't wrap textures from an image onto low polycount objects.  As for textures, use AI to determine materials from the texture, object type (window) and from a predefined set of materials match up and build the world ... OR if no such material/objects match you create a new one based on similar objects AI can use as training to create a new object.  All you feed the simulator is the already constructed world.  What you get over time as the world changes are constructed updates.  You have complete control over what how you work photographic objects, determine if a road is made from concrete slaps or tarmac or stone/dirt, you ignore cars on that road (lets face it who wants to see traffic driving over static flat car images on a road?) ... all AI processed using native materials on the destination polys not UV mapping of photographic image.  What is required are very high quality initial photographic texture from which the AI can build materials/objects.

All the parameters don't need to be calculated, just the more obvious impacting ones.  The world generation is done pre-processed on servers/quantum computers living elsewhere, the users PC is just a client to provide the lighting and flight simulation physics.  I guess you can think of it like Bing on steroids but without the melted buildings, baked shadows, flat cars, overpass/bridges one can fly under ... and a blended world because all the materials are color accurate so the lighting responds accurately when introduced in the render pipeline.   

 

22 minutes ago, CO2Neutral said:

We'll see, AI is for initial data translation/recognition to the world being generated, it's not predictive in the same sense as climate processing which does indeed consume considerable processing resources.  We're also starting to see usable quantum computing services that do all the work extremely quickly and provide the results.

Much like BlackShark.Ai, except you don't wrap textures from an image onto low polycount objects.  As for textures, use AI to determine materials from the texture, object type (window) and from a predefined set of materials match up and build the world ... OR if no such material/objects match you create a new one based on similar objects AI can use as training to create a new object.  All you feed the simulator is the already constructed world.  What you get over time as the world changes are constructed updates.  You have complete control over what how you work photographic objects, determine if a road is made from concrete slaps or tarmac or stone/dirt, you ignore cars on that road (lets face it who wants to see traffic driving over static flat car images on a road?) ... all AI processed using native materials on the destination polys not UV mapping of photographic image.  What is required are very high quality initial photographic texture from which the AI can build materials/objects.

All the parameters don't need to be calculated, just the more obvious impacting ones.  The world generation is done pre-processed on servers/quantum computers living elsewhere, the users PC is just a client to provide the lighting and flight simulation physics.  I guess you can think of it like Bing on steroids but without the melted buildings, baked shadows, flat cars, overpass/bridges one can fly under ... and a blended world because all the materials are color accurate so the lighting responds accurately when introduced in the render pipeline.   

Yes, I totally agree with you there!

However, the only prerequisite is that the influences of the materials and objects on the light representation and thus the appearance for the viewer in reality is known, so that it can be recreated in an engine. And that is not the case so far.

The "creation" of the scenery, terrain shape, vector data, buildings, trees, yes, even just traffic up to animated people, creatures, birds, all that can be done (ok, still with some cutbacks...) to some extent by AI. - Provided that the computing capacity is there for it.

But the visualization of the results has to be done by the PC in the end - and as long as the basics for a realistic visualization are not there, you can't really use all the advantages of this method. And this is the real problem:
Engines that are among the best for realistic rendering (for example, the Unreal Engine) have not yet managed to process such large amounts of data.
Engines that can handle such large amounts of data, however, don't look as good.

AI may be able to help here in the future, but we're not there yet.

 

@MatthiasKNU, I know you have done exceptional work in bringing free photosceneries of high quality to the P3D user base. Instead of us trying to recreate the equivalent of BlackShark.AI, is there an alternate route where the community could help (i.e.funding, assistance) to make more projects come to light?

Thanks,

Vincent

3 hours ago, gyzmonfishball said:

@MatthiasKNU, I know you have done exceptional work in bringing free photosceneries of high quality to the P3D user base. Instead of us trying to recreate the equivalent of BlackShark.AI, is there an alternate route where the community could help (i.e.funding, assistance) to make more projects come to light?

Thanks,

Vincent

Well, that's really an excellent question.
And I'll also say this quite honestly: I don't know.

I think it just depends on the user numbers:
Many users = many sceneries, few users = few sceneries.

Therefore, one should actually ask the question the other way around:
Why does the P3D have significantly less users than XPlane and again much less than the MSFS?
(Well, why less than XPlane I personally can not understand, because I do not get along with XP at all... 😉)

With the MSFS, however, many reasons are obvious: The landscape rendering is very good "out-of-the-box", as is the light and weather rendering, and: It is easily available for PC and XBox via the MSFS Store. For 95% of users, this is the most important thing.

P3D has other strengths - but these are of interest to fewer users.
Therefore, to come back to your question, what would have to change:

The default P3D would have to change.

In particular, in the following points:
- Landscape rendering (yes, with the help of AI's capabilities!).
- Weather representation (yes, also here there would be huge potential for the use of AI - e.g. live weather not from model calculations as in the MSFS but on the basis of live satellite and weather radar data).
- Engine and light rendering
- Distribution channel - not only via the LM homepage, which alone scares off many.

Therefore, these are all points that we as users simply can't really influence, but LM would have to take action here - but since LM has the strong military fixation, these points are probably not in the foreground.

As I said, this is all just my subjective opinion - nothing that I can back up in any way with facts.

Edited September 16, 20232 yr by MatthiasKNU

I suggest we concentrate on the subject of Asia. That is the continent that needs more attention. From January next year I am planning on spending 7 months in development of a lot of scenery for Asia. First, Indonesia, which will start with Jakarta and hopefully spreads thru the whole country, but I'll be spending prob. 12 hours a day developing the autogen geographical data so I can gain the best possible quality out of the place. 

Second I am concentrating on the neighbour country of Malaysia. Then Japan, if I have the gut to actually produce the whole country in a month. Then Hong Kong Area, which won't be a lot of work, because the terrain sceneries of Taxi2Gate and FlyTampa are covering most of the town. Then Emirates, where will be not much of an labor as well because FlyTampa covers pretty much of the town. 

If I manage for a few months to produce those I will share them in my topic at this forum for anyone that wants to use them. Also I am planning on a large photoreal imagery cover for the whole country of China which I'll continue develop in 2025. Also India. Also Kazakhstan. 

My plan is simple - I WANT TO COVER ASIA WITH PHOTOREALISTIC SCENERY. With exclusion of Russia. Sorry, I don't like Russia on this troubled times and also that will be a mega project, which I can't create on my own. 

If you guys are more curious of my future projects I suggest you visit the second link in my signature. I created polygons on that map to make visible the countries I am planning to develop from next year on. 

The reason why I start this projects is to make my World Trip 2024 project look ravishing anywhere I go on that map. I am going to produce mega video with it. For now I am going to invest time in landing practices of all airports in the world trip and export videos with them to have something entertaining for the YT channel I have and to show everything without my work on the scenery. 

This are my plans for scenery creation for now. And I'll use the old method, which is:

1. Creating polygon data in QGIS geographical software for autogen and trees. 

2. Importing the data into ScenProc, create my own scripts and export autogen. 

3. Downloading Ortho4XP for P3D tiles for the scenery as a background under the autogen. 

On 9/15/2023 at 11:22 PM, MatthiasKNU said:

- Weather representation (yes, also here there would be huge potential for the use of AI - e.g. live weather not from model calculations as in the MSFS but on the basis of live satellite and weather radar data).
- Engine and light rendering

This makes me believe you haven't tried the latest P3D V6 build and HiFi ASP3D for V6.  I don't want to go down the road of P3D vs. X vs. Y, but do you actually own P3D V6 and have used latest build?

9 hours ago, CO2Neutral said:

This makes me believe you haven't tried the latest P3D V6 build and HiFi ASP3D for V6.  I don't want to go down the road of P3D vs. X vs. Y, but do you actually own P3D V6 and have used latest build?

You are absolutely right, I do not have P3Dv6. But from what I've seen in all the many screenshots, this is still miles away from what I'm talking about here....

But please - if I am wrong here, I would be very happy to be proven wrong with pictures! 😉