I asked Claude whether it could help me with creating an Icefield Strip above Juneau.

it replied not yet but if it had access to the MSFS 2024 SDK it would be possibe.

It walked me through enabling Developer Mode showing me a step by step picking the coordinates then writing the XML code how to place objects and testing them within the terrain model.. It gave me certain Taku Glacier coordinates that should not have any conflicts. it checked how Runway and scenery objects are placed..it talks about crevasse free spots it talks about working through SDK Console errors any snags...

this was a one minute question and it delivered a PDF of how to create a runway and basic scenery...

is this going to be a game changer especially if it had access to the SDK or one thinks its all hot air?

Fascinating and frightening...

Hamish

So why aren't Microsoft/Asobo doing that? Or maybe they are, and the result you can see is the default scenery?

Possibly they are...

What I find interesting.. does anyone here remember Return to Misty Moorings?

They had for FSX CIRP and SIRP amazing ice field runways that had so much character, there was even a yeti you had to find!

I am surprised so many developers only focus on airports and not more interesting locations that involve weather short runways and a story...

Emma Field was part of that...

if Claude can help maybe I can have a go I have some great ideas for an adventure...

Recreating the set for John Carpenter's The Thing just outside Stewart would be an example..

Maybe I am too niche?

h

4 minutes ago, Hamish100 said:

Possibly they are...

What I find interesting.. does anyone here remember Return to Misty Moorings?

They had for FSX CIRP and SIRP amazing ice field runways that had so much character, there was even a yeti you had to find!

I am surprised so many developers only focus on airports and not more interesting locations that involve weather short runways and a story...

Emma Field was part of that...

if Claude can help maybe I can have a go I have some great ideas for an adventure...

Recreating the set for John Carpenter's The Thing just outside Stewart would be an example..

Maybe I am too niche?

h

In case you weren't aware, Return to Misty Moorings does exist for MSFS 2020 / 2024 too. It doesn't include all of the epic previous locations they had, but what's there is a lot of fun still.

I agree, I'm surprised this form of scenery design isn't more popular.

8 hours ago, Hamish100 said:

is this going to be a game changer especially if it had access to the SDK or one thinks its all hot air?

I'm sceptical that Claude (or another AI agent) will be able to generate scenery effectively, for several reasons:

• LLMs generally aren't good at spatial reasoning (which is important when generating scenery)

• The LLM isn't able to "see" the effects of its actions, i.e. what the resulting scenery looks like. Having created sceneries in both MSFS and X-Plane, I find this to be a critical part of the process. One of the reasons AI agents are good at programming tasks is because of the "feedback loop" where they can run the generated code, see what it does, and correct the code if it's not doing what it should do. WIthout a similar kind of feedback loop, I wouldn't expect an AI to do well at scenery generation.

• The LLM likely isn't able to accurately extract geolocations from aerial images (which is the way that humans doing scenery design usually place scenery objects).

It would likely be possible to develop/train a custom AI model specifically for scenery generation that overcomes these limitations, but that would take a lot of effort/money, and I'm not sure that'd be worthwhile even for Microsoft/Asobo.

If you do give this a go, I'd be curious to hear how it goes!

Claude is indeed insane. Two weeks ago I asked if it can help me generate a livery. It says yes as long as I provide some sort images and do some convertings

40 minutes ago, History said:

Claude is indeed insane. Two weeks ago I asked if it can help me generate a livery. It says yes as long as I provide some sort images and do some convertings

Did it work? You should post a result

I think we are definitely at the point where it makes sense to keep an eye on AI development for sim development. I recently asked AI (probably Gemini) for help with making more localised vegetation for an area, and it gave pages and pages exploring different options, based on the SDK. The only problem was that it didn't seem to recognise the 'not yet implemented' tags. However it did have a real insight into what these features might do without them actually existing yet.

Lately I've been looking at AI 3D models, and settled on Tripo as it specialises more in 'game-ready' low poly assets, whereas most of the others give a huge number of vertices, which don't suit the sim. Initially I wanted to model some carvings adjacent to an airport I'm working on, and remembered the old days where you could upload hundreds of photos and get back a 3D model. With Tripo, I can get a good result from a single photo, although front/back/left/right is better.

It's great for things like vehicles to populate airports etc, I can upload a photo, or just give a text prompt with the make, model, year and colour. I still need to create LODs, but the starting point is a fully textured LOD0 model ready to use in the sim.

I always imagined an app which let me visit an airfield, take photos with my phone, and using the inbuilt GPS coordinates it would build me an almost finished airport by the time I got home.

Note that the Tripo link above may give me credits via an affiliate link, but I don't make any money from this, and only create freeware scenery these days.

I hope not. AI should not replace artists

31 minutes ago, Tuskin38 said:

I hope not. AI should not replace artists

I have thought a lot about this, and while I agree with you completely, I'm not too sure that this applies to my own experience with scenery development. I've been doing this for 25 years, and for half that time it was my sole source of income. What I do is not art, in most senses, the only way it would qualify is that I have always had a vision* of what I wanted to represent, but I am really a technician. I follow a technical specification, built around taking my own photos and using these as the foundation of my textured models.

I work around a few limitations, mainly aphantasia, the inability to visualise in my head, (so the *vision I mention earlier is not actually any image, more an idea of what makes good scenery), and autism, which gives me a specific goal which 'normal' folk may not have.

When I did this for a living, I paid for most of the tools I used, but I don't recall paying anyone to produce something which could be considered art. In my case I'm not switching from paying a person to paying an AI, as I was the only 'artist' who ever got paid by my business. AI just speeds up some of the work, and allows me to look at things which I couldn't include through my own efforts.

I've modelled a LOT of hangars in my lifetime, and I'll aways do these myself, as there's not a lot of artistic ability required -- they are all just a box with a photo stuck on it, after all. Some things, though, I wouldn't do justice to using my 'mid-level' skills. I've modelled cars before, for example, and I could build a fleet to populate carparks in MSFS 2024, but AI does it a lot quicker and more consistently than I could, so I'll use my time for something which makes better use of my skills.

Sure, most AI tools have plundered someone's hard-won talent to the extent that some artists will lose out badly, but this is one part of AI, which will need to be dealt with at some point, although this may not be a quick and easy thing to fix. It doesn't take away from what AI does well within the bounds of ethics. AI is not in itself unethical, it's those who 'own' it who seem to want to proceed without limitations, but that goes for most technological development these days.

What you’re describing here is AI-assisted work—something that has, I believe, already become a firmly established part of software development. However, I would point out that all the major providers rely on LLMs; while these can ingest vast amounts of data, they frequently get things wrong. Furthermore, if you point out errors to these models, they have a tendency to get stuck in a feedback loop where they keep trying to "correct" their incorrect information with even more incorrect information. It is also incredibly easy to trick LLMs by presenting actually correct information as if it were wrong. So, there is nothing wrong with offloading the tedious, exhausting, and time-consuming parts of scenery development—tasks that are easily automated—to AI. But handing over the entire development process to AI isn't the way to go. You could try it, of course, but then you’d end up with results like Microsoft’s Windows localization in my native language; it contains so many appalling textual and grammatical errors that it was almost certainly generated by AI without anyone ever reviewing it afterward.

I also generally consider the development of scenery to be a less creative process than creating video games themselves. When you recreate an airport, the real-world model already exists; your task isn't to creatively design a new airport, but rather to replicate the existing one as precisely as possible.

This would pose a real problem for the current Marketplace revenue model if taken far enough.

On 6/22/2026 at 12:37 AM, Hamish100 said:

I asked Claude whether it could help me with creating an Icefield Strip above Juneau.

That was the actual prompt? I'm interested in some corrections to default airports, but (unlike FS9/FSX) I don't know where to start.