Usage of AI in the Hobby

[JJ1892]

Hello All,

 

I am writing this post with the hope that it will serve as a place holder to discuss AI (Generative like MidJourney and LLM based like ChatGPT ) usage in the hobby. 

 

I’m personally busy with:

+ Using ChatGPT to help me write Arduino script for my Messaging platform.

+ Using AI to generate Platform messages in Japanese with the benefit of not infringing on JR properties. 

+ Trying to decipher how I can use AI to make track plans that will serve a jumpstart for further work. I am also toying with Genetic Algorithms for that. 

+ Whenever I have the urge to open amazon.co.jp I try first to create imaginary Japanese trains with generative AI. My personal weakness for forced associations. Alas, usually it does not help.  

+ Researching the evolution of a certain train lineage. 

+ Help me find unique trains that are in my blind spot (namely, do not appear in a KATO or TOMIX catalogs, nor appear in forums or in the pages of the big vendors).  

  

So, some ideas how AI could be used or discussed in the forum:

1. Generating Historical Context: If the model is based on a particular era or location, AI can provide historical context, industry details, and realistic scenarios from that time.

2. Theme and Design Ideas: we can brainstorm unique themes, such as a futuristic sci-fi railroad or a historical recreation of a specific train route. 

3. Where models are integrated with JMRI or Arduino-based automation, ChatGPT and the likes can help write or debug control scripts.

4. Generative AI can help generate station statements, background voices and the likes I any languages. 

5. Generative AI can be used to create unique textures for backdrops or ground cover. 

6. AI could suggest design ideas for custom scenery, props, and buildings which you could then create using a 3D printer.

7. We could use ChatGPT to generate stories or histories for the towns and characters within our model train world for added depth and context, making the layout more immersive.

8. Generating initial track layouts under specific constraints.

9. Generating fictional trains and then printing them in 3D printers. 

 

Will be happy to hear how others are using AI in this hobby. 

 

To make sure, the following is _strictly not_ under the focus of this post:

1) Using AI to generate messages for this forum.

2) Making anything with AI while infringing the intellectual properties of others.  

 

Just a (human) thought. 

 

Best regards,

 

JJ1892

Edited March 2 by JJ1892 formerly Abushoni

[cteno4]

Just a clarification note on the forum guidelines on AI for everyone. Last year we added the prohibition to the forum guidelines of no post content can be AI generated. We did this in response to a user using an AI system to generate their posts here and it was a mess. We decided we did not want forum postings to be come an AI event, but keep it a human to human interaction. Use of language translation system [many of these are now AI based] is fine as we are a worldwide forum but predominantly converse in english and we want folks to feel like they can use those system to translate their own words in their language to english [or other languages as needed]. But we don’t want the AI system writing the text from a few commands. We also just want to steer clear of AI generated content posted here that may contain parts of copyrighted or potentially copyrighted content of others as this is a currently hotly debated topics and rules and laws vary around the world and we just don’t want to step in anything. We also don’t want the forum to become an AI training ground as well.

 

We do allow discussion of the use of AI in the hobby though, like JJ’s post here. But please be careful of our guidelines of this slipping into politics which we do not want.

 

It a great JJ has brought the topic up for discussion and they were great to check with staff first to make sure of the guidelines, thanks!

 

cheers

 

jeff

[Junech]

For me the only appearance of AI is translations since the outputs are way more detailed then normal translation software. For everything else I don't need it.

 

For anything with research it can be a starting point but it has so many mistakes in the answers that I won't use it beyond that. If I need to double check everything anyways I can go to the source from the start.

 

For themes and such I like to go with prototypical things that's why I don't need ideas.

 

I can see AI being helpful with programming things although I didn't need that so far.

 

For the voices things: I don't need like a lot of people talking as background noice and for station announcements you can find a lot of voice actors that will do small things for just a few dollars in any language. So if I ever need something like that I will most likely go that route.

 

Images for backdrops I would need the prototypical pictures and with AI not getting close enough to them I would need to take them myself or a professional to take them or an artist to draw them. For textures there should already be everything imaginable as seamless textures for 3D modelling available online.

 

Station layouts follow a simple enough logic to create them fast and easy yourself with a bit of reading into them. And I like doing them since I find it pretty relaxing... So no AI as well.

 

For the world building for fictional layouts I'm way to deep into world building then ever going to give that to an AI as well. Running TTRPG like DnD for years got me going at those things as a free time activity for a long time already.

 

Same for 3D modelling... I enjoy it, so why give it to AI?

 

A lot of it boils down to the way to reach the finish line being my personal goal. I really enjoy the steps it takes to get to the finished layout, so I'm not going to give those to AI. And for the parts I can't do myself I would go for humans and commission them over AI too.

 

For me a hobby is a luxury and a luxury is never cheap xD if I need a backdrop and someone to take a some professional pictures I will get some trains less and spend the money there...

[JJ1892]

Junech, thank you for the detailed response. 

 

3D modeling, programming, layouts - I totally enjoy them as well. But as the known quote by a Silicon Valley executive said - "soon you would compete with AI". Many people interpreted this as AI replacing people  and stealing their jobs. But the executive meant that AI is on your team. Not against you, and certainly not replacing anyone. 

 

I was hoping to understand in this post, how members here enjoy exactly what you said, but _with_ AI.

 

I can ask a voice actor to record me a message in Japanese. It will take time of back and forth communication until I am happy. With AI I can do this in 2 minutes. AI amongst other things is about being more efficient. And being more efficient save me time where I can run more trains or try more setups.    

 

I also think that young people which I guess we want to attract to this hobby, will find AI natural as we see a word processor. Further, AI enables the "democratization of abilities" (this is a term taken from innovation theory not politics), in the sense that by lowering entry barriers, more people will be able to take part in the more complex facets of this hobby.  

 

As Mitch Horner said In Model Railroader, "The strides made in the field of LLMs in the past year is impressive. It now offers real potential value to model railroaders, particularly to those searching for inspiration or resources for a new layout, or those modelers beginning their first foray into the hobby." (source: https://www.trains.com/mrr/beginners/revisiting-ai-for-model-railroaders/).

 

AI is a rapidly evolving field, where I think there are many questions and people still try to find out the potential and usage of the technology. I personally think I will be a better railroader with AI on my side.  You said, " if I need a backdrop and someone to take a some professional pictures I will get some trains less and spend the money there...". I will use AI, and get more trains. 

 

I guess it will be interesting to re-visit this post in 1 or 2 years time, and see if reality changes. We will see if the hobby is going to be "only human" or "human + AI". As long as "human" is in the picture, I personally will be satisfied. 

Edited March 4 by JJ1892 formerly Abushoni

[Junech]

I don't see AI as anything different from a tool so I'm not against AI at all. It is more about how we create and use it.

 

The problem with AI is, as the article you linked mentioned, "the inaccuracies" it still has. Furthermore, AI currently doesn't actually understand what you want and (way oversimplified) gives you the statistically best estimate of an answer.

 

In comparison, the voice actor or artist/photographer (at least the good ones) have a lot of experience and knowledge gathered over the years and the back-and-forth you described is to give you not what you asked/wanted but what you needed (even if you didn't know that from the start).

 

The only way to counter that is to bring in the knowledge yourself and use it as what it is - a tool. So I still think AI isn't quite there yet and it still is better to seek help from people in forums or other social media instead of turning to AI in most cases.

[JJ1892]

I'm in total agreement with you that AI is just a tool, and even more so that if I have a problem I will approach the members of this forum.

 

I was hoping to hear exactly that - how people use AI as tools in model railroading. From the silence, I guess you represent the prevailing opinion.

 

JJ1892  

[Junech]

I think there are multiple reasons why AI isn't used in the field of model railroading at the moment.

 

Besides the general way being still to just "google" something instead of going to AI tools. People still want to talk to other people about things more than using AI. And all the things that fall in line with these things.

 

Going back to your list:

- Arduino and other programming is a relatively small group of the hobby, so finding someone who uses AI over sites like StackOverflow or just adapting someone else's script will be small.

- Platform announcements and other sounds are rare too besides the field of train sound. So it is the same as above.

- There is so much to consider with track plans that I don't see AI being there soon. Too small of a market to train AI for that besides a passion project of someone with time on their hands. The amount of thought I put into planning of layouts isn't understood by AI at the moment. (a really small part is readable here: https://jnsforum.com/community/topic/20953-layout-ideas-for-busy-city-scape/?do=findComment&comment=262791)

- I skip AI for research since most parts I'm interested in are readable in a more curated form in print media or websites. The can be a lot of nuances that are important. I spend days translating Japanese texts using multiple tools including ChatGPT to get it right. There is a saying that "you need to talk to the author to make really good translations" for a reason.

- Fictional trains - besides trains from well-known fictional works or heavily inspired by "this follows general train design standards" are also rare.

- Fictional history about fictional train layouts is another rare thing, most cases are it is roughly in that real-world area so it follows that place's history.

 

 

Besides all that I know more people following the "the path of progress is the goal" than "driving trains is the main goal" people. As for me that accumulates to the Fremo meetup I had for 5 days over last weekend satisfied my desire to run trains for the next few months.

[Cat]

It's a tool that's redundant, as well as less reliable, to all the resources I have within arm's reach and not one that would not be as pleasurable as doing my own research and design.  Plus, I have no desire to contribute to the level of electrical power required to fuel it and the resulting heat output of the data centers.

[Little-Kinder]

Would be fun to use AI to generate layout plans.

 

Curious if it's something possible and reliable (or at least to give ideas)

[railsquid]

9 hours ago, Cat said:

It's a tool that's redundant, as well as less reliable, to all the resources I have within arm's reach and not one that would not be as pleasurable as doing my own research and design.  Plus, I have no desire to contribute to the level of electrical power required to fuel it and the resulting heat output of the data centers.

 

Ditto. I spend too much time in front of the computer for work (ironically part of which is now taken up with an ongoing battle against content theft by "AI" companies), and one attraction of model railways is that it's a creative thing which gets me away from the computer (mostly), I don't even want to mess about with DCC.

 

An additional aspect is that much of "content" use by "AI" to generate output was likely acquired with the consent of the content producer, and it would be hypocritical to be complicit in that process.

[gibet_b]

It's fun, I just finished to write an article about AI and model railroading. More precisely about the 3D generative AI (what it's good for, its limitations, etc.). 
2D generative AI could be used for the buidlings's interior amenagement.

[JJ1892]

gibert_b, if you feel comfortable, it will be great to have a link to the article once it is published and maybe some examples ?

 

Here is an article that follows your suggestion of generating internal interiors:

https://www.trains.com/mrr/how-to/expert-tips/designing-structure-interiors-with-artificial-intelligence/

 

Thank you,

 

JJ 1892

Edited March 5 by JJ1892 formerly Abushoni

[gibet_b]

It will be in french and in a magazine, but I will probably be able to put an english translation here, as few of our readers are fluents in english...

And yes, the article is exactly the kind of use I think about.

[JJ1892]

On 3/5/2025 at 12:49 AM, Cat said:

It's a tool that's redundant, as well as less reliable, to all the resources I have within arm's reach and not one that would not be as pleasurable as doing my own research and design.  Plus, I have no desire to contribute to the level of electrical power required to fuel it and the resulting heat output of the data centers.

Cat, I am more focused on the functional aspects of the issue of using AI for better or worse. Less on ethical, environmental and the rest of the facets that can quickly slip into politics and stuff we do not want here I guess. Not that these considerations are not important off course. Our trains are made of plastic and batteries which to say the least are not environmental friendly, and this web site is hosted on a cloud in one of those data centers. I suggest to focus on model trains.   

 

As for what you have said, I totally agree. I also like researching books and failing forward. It is not an either/or thing. AI, is just another tool. 

[JJ1892]

To sum up this issue, at least from my point of view, I was sure this thread will blow with ideas by the community. The silence and reactions of some of you, who have by far more experience than me, show otherwise. Maybe this yet another one of my baby steps mistakes in this hobby. 

 

In any case, I will be pouring AI related stuff into this thread as the technology matures and as I stumble on things, if only to see its evolution within the realm of this hobby. It may end as a hype, or we will be surprised.  

 

JJ1892

Edited March 6 by JJ1892

[cteno4]

Like most more general AI LLM stuff its still a bit of a solution in search of a problem to find where it’s really good and useful. Hype and chaos around it as well has made many just step back and do a wait and see approach. It’s also a whole new tech thing to have a learning curve with and many may just not want to invest time in that learning curve. There will be more tech oriented folks like JJ that will dive into all this early im sure and find ways to use AI that will be of use and enjoyment to them and hopefully later to others not as into tech. Its also going to take a lot of experimentation by those that enjoy doing that sort of fiddling with technology to find those things that work and how to easily do the process as well as documenting it and getting it out to other hobbyists to play with and move forward.

 

I think there will be place that will pop up where AI will be a useful tool in the hobby. The link to creating the interior pictures is a good case in point where you can use current AI systems to make something good enough for that purpose without having to have graphic skills and utilize the vast reservoir of images that ai systems have sucked up for their use. Backdrops images may also become a place for ai to be able to stitch together or create wide aspect backdrops.

 

Programming for automation systems and arduino may also become a good place with simpler programming and command systems to create arduino C code for non programmers or for dyslexic folks like me that have issues with C programming. Same with more sophisticated DCC sytems, signaling, block control, tracking, and automation.

 

I think eventually some of these will be part of the toolbox folks can dive into to use in the hobby along with traditional ones. For the most part I would enjoy more using more traditional tools/techniques as a big joy in the hobby for me is making things myself with my own hands. But there are situations like above for me that AI could fill in where i just dont have the ability to do some of these things very well or as enjoyable to me. We all have our own mix of the things we like and dislike doing in the hobby so it will be different for all of us im sure.

 

As in any crowd there will be a spectrum of folks who would enjoy or feel comfortable with something like AI tech. It’s really all in where you get your joy. For me i tend to be a big pragmatist and always looping the ROI loop for myself of my enjoyment in the hobby and what i want to do against my skill set, resources, time and money. While ive worked on the bleeding edge of tech for decades in my work life and have a big tech aptitude, i still keep the pragmatic ROI loop rolling. This is why 3D printing is just not something i have jumped at. Spending tons of time on the computer doing 3D cad i dont find appealing to me anymore. Nor do i like just push the button and then the physical thing is all done for me, i do enjoy making things. There are places i think yes it would be grand to 3D print this part or that tool, but there is not enough of those for me to say its worth the time/effort invested to get those. I also love the mechanical design of things and building them so doing things in 2D cad and cutting parts flat and assembling into a 3D structure is my sweet spot of enjoyment and ROI. Others will have a very different mix.

 

I look forward to what folks present in this topic! Also fine for folks to start a new topic on an AI tool/technique and then just put a link to it in this more general topic to cross reference it to the more detailed/focused topic.

 

cheers

 

jeff

[cteno4]

So I just tried messing with the bing image create to make some japanese building interiors. I got some pretty nice looking ones that would work from use as interior backdrops inside small tomytec building’s first floors. There were always some odd details like wanting to always put a bunch of seats in any sort of interior you asked for even it there would not be seats/stools in that type of shop, sometimes multiple rows of stools lined up behind each other. I didn’t find a good way to try to fix the perspective from one position. It also wanted to make things multi story a lot even if you specified one floor. But with fiddling of ordering of the terms you can get some decent results

 

But thru windows on n scale some are plenty good enough. Only issue is the engine kept wanting to make very similar looking interiors in sort of the same 4 configurations even with different kinds of shops specified, so what you would actually be able to see from the exterior [and you would need to very close to see it] would all pretty much look the same. But again for this use it may be fine to just see something in there than being blank. But for something quick and dirty it would work.

 

Lighting them will need some work to make them better visible but will require some playing with to make them pop right, trickier than lighting an interior with actual items in it.

 

I took a bunch of attempts to try to make an office scene for a larger office building, but no luck there on anything usable. Usually these windows are much more visible and Im thinking the traditional route of simple bits of plastic and cardstock for office furniture and some figures will work much better.

 

but for small tomytec building interiors where you have a very limited view these may work fine and be faster than trying to make a 3d version of the interior. I’ll need to try a comparison some time here with each approach to see the work. I doubt i would do the tedium of doing all my tomytec buildings interiors with 3D bits but i could see popping in little backdrops quickly to help not make them look empty if you look closely.

 

cheers

 

jeff