Discussions about AI for and against content creations.
- Thread starter pdkoester
- Start date
zatovisualworks
Veteran but blossoming
Please, care for keeping it civil/visitable/unabusive/on-topic. Thank you. 
Ok. So, I'll start by saying thank you so much to all content creators both past and present. And to all creators both here in the U.S. and overseas. lot of good content from around the world. Trainz is my most preferred train sim over the other sims because of its freedom of play, which is why I support all content creation. AI, sketch up, whatever. Those that have mastered 3ds Max creation will always be the go-to choice for content, but. I don't think there's enough of those types of creators to fill the insatiable void that us gimme pigs have for the content we want, and nor should we expect them to deviate from what they normally create to satisfy that void. I think that other methods of creation should be encouraged or at least tested out and used as a learning tool for those who just want to contribute something. To be so triggered over a free digital download, that it stirs up half the community... Come on now... Is it really that serious? Maybe AI will get better over time, maybe it won't. You have the right to hate it all you want, I don't care, as long as "We the Piggies" and the void are ok with it for the time being. So, keep calm, back away from the computer, go outside and breathe in some of that fresh real world, cause at the end of the day, Trainz is just like any other video game. Mindless entertainment.
Last edited:
Dinorius_Redundicus
kuid 68213
I haven't delved into this very deeply, but my main concern would be about high-poly and/or texture counts without proper LOD reduction. By "proper" I mean LOD schemes that meet the technical requirements of >20% reduction per step and a final poly count under 500 triangles. There is also the aesthetic requirement that LOD transitions not be too visually obvious. High texture counts should be addressed by mapping onto texture atlases.
Having a zillion tri's out to 1000m (or whatever) then making it a 10-tri box or completely vanish is not an honest or effective LOD either.
In other words, use AI to do things you just don't have the talent to do, but find ways to make the assets performance-friendly in a Trainz environment before spamming them to the DLS. If you can't do that, be an honest Trainz citizen and prominently note the poly-count and texture count and lack of LOD in the Asset Description.
Having a zillion tri's out to 1000m (or whatever) then making it a 10-tri box or completely vanish is not an honest or effective LOD either.
In other words, use AI to do things you just don't have the talent to do, but find ways to make the assets performance-friendly in a Trainz environment before spamming them to the DLS. If you can't do that, be an honest Trainz citizen and prominently note the poly-count and texture count and lack of LOD in the Asset Description.
Last edited:
johnwhelan
Well-known member
I think my concern is people with unsupercharged computers thinking that trainz won't run on their machines because of suboptimal assets. The AI generated assets at the moment tend to use ploys rather than textures.
Cheerio John
Man, what you just said is straight rocket science to the regular consumer like me. But I understand that there are some performance issues with AI. (Love your content by the way, you are definitely a Legend) I'm not expecting an AI takeover which I think some of the critics seem to think so. But I see it having potential. I assume these modeling programs aren't thinking Trainz first, but maybe someone will figure out how to optimize it for trainz, and improve on its capabilities.
Therock
<kuid2:37581:420:69>
There are certainly easy ways to sloppily reduce tri counts of these behemoth models but half the problem is that arguably in most cases the (ai generated) models shouldn't be even close to their highest poly levels at lod 0. Due to lack of optimization, file sizes have inflated with larger texture sizes and increased model detail in the last 15 years but for any average joe without an FCT most assets have become less and less available if coming solely from the DLS, since the speed has remained capped at the mediocre 5kb/s for as far back as I can recall.
It is also incredibly easy to "comply" with the existing reduction/step and final lod, as you mention in your second point. I'm all for having those kinds of limitations on most types of assets in the first place, as an effort to put in a bare-minimum requirement, but I do feel like some responsibility is shouldered on the end-user to determine for themselves whether or not they can or should use the assets in question. Anecdotally there are several assets and even authors I've outright blacklisted in my game due to making "slop" that hogs more resources than I personally think it should on my system- meanwhile I've made some very resource intensive content for my own purposes that someone else may feel the same way about using themselves.
The last thing any one author needs is more arbitrary restrictions from n3v on what content passes error validation, especially if any of that content in question is never even intended for public release.
I can get away with uploading the biggest absolute trashware models or bismal [sic] content, depending on who you ask, to the DLS without breaking any "requirements" but at whose discretion should they be removed? at whose should they not be allowed in the first place? This is a can of worms that isn't so easily capped once it has been opened.
I enjoy finding something that hasn't been done before and creating it for Trainz. By that I mean, starting from scratch and making my own textures and all. I don't use A.I. but that's because I use GMax and there's no way to convert an A.I. file into something GMax can open. I'm probably heading down the 3DSMax track and I'll reach that station eventually. But right now, I'm sort of like Bob (MSGSapper) in that I mostly create for my own enjoyment and my own route(s). Sometimes I'm willing to help someone out with something if I, too, find that something is interesting and would be fun to do. But usually I don't take requests. As for A.I. there doesn't seem to be any sane reason why someone can't use it to make something for Trainz. Heck fire, we don't have enough content creators as it is! A lot of the old timers have either packed it in and gone on to something else or passed away. RIP those great content creators, Clem, Ben, and others I can't recall at the moment. I'm almost 79 so one of these days I'll be going home, too. But I don't see any problem with A.I. Now don't get me started on LOD. I don't provide it so just shoot me.
One thing I hear on these forums quite often is stuff like, "I don't know how to create content for Trainz." or "I don't have the skills to get into content creation." Listen folks, It ain't that hard!! If you can work a computer, you can learn Blender. I don't use it or know anything about it, but from what I've heard there are gobs of tutorials out there. Jump on the bull and ride! What have you got to lose?!
But let's not put A.I. or people who use it down. Give them the right to chose how they create content, and if A.I. is a part of it so what? How in the blue moon is it hurting you? If you're p.o.ed because someone is using A.I. to make content, then don't use their content. Simple as that.
Cheers,
Dave
One thing I hear on these forums quite often is stuff like, "I don't know how to create content for Trainz." or "I don't have the skills to get into content creation." Listen folks, It ain't that hard!! If you can work a computer, you can learn Blender. I don't use it or know anything about it, but from what I've heard there are gobs of tutorials out there. Jump on the bull and ride! What have you got to lose?!
But let's not put A.I. or people who use it down. Give them the right to chose how they create content, and if A.I. is a part of it so what? How in the blue moon is it hurting you? If you're p.o.ed because someone is using A.I. to make content, then don't use their content. Simple as that.
Cheers,
Dave
In many cases you can use those ultra poly models to bake normals to a much lower poly mesh that will be your "starting" mesh - i.e. a LOD0 mesh. It's a technique I've been trying recently. The problem is getting a good UV map in the first place.
Wow, that's a mellow post for you Dave and, oddly enough, I agree with all of it. Well said.
I would use AI for a model but only if I thought the result was good enough. My current view is that it doesn't seem to give good results for hard surface modelling (locos, traincars, and most anything for Trainz except maybe trees, bushes and the like. Does a decent job on humans and animals. Some of the recent stuff from Bob I saw looked mostly OK to me although the old general store looked like it was melted to me.
TBH, I quite like making models in Blender from scratch. The biggest thrill I ever had in Trainz was making my own steam loco starting from a single vertex and watching it run around a track.
TBH, I quite like making models in Blender from scratch. The biggest thrill I ever had in Trainz was making my own steam loco starting from a single vertex and watching it run around a track.
The Mid-20c people, the AH people and of course MSGSapper people and animals, are all I ever refer to about this AI topic. Everything else AI... Yeah... It's on that struggle bus. Now that I'm a recently retired state government employee, maybe I'll "jump on the bull and ride" like Mr. Snow said. My biggest thrill for now is that I'm starting to get more comfortable using surveyor 2.0, where I become the next level of content creation in my own little Trainz world. Twisting and tilting content that I couldn't do in classic surveyor is making me look through old content that I didn't like before. I love mix matching and kit bashing and I always wanted to create building detail parts for trainz.
norfolksouthern37
what's a normal map?
Just as a technicality, this isnt true at all. but anyway.
Not directly addressed to Dave here, but there seems to be a lot of people that apparently still do not understand the argument. I don't think anyone is raising a point that someone is using AI. We are upset because AI creates content trash and don't want the available assets we have to be even more polluted with it. Same as when we didn't want to see loads of auto generated ruby/google objects flooding the DLS created mostly the same way yet not called AI then. It isn't a matter of attacking someone over use of AI it is a protest of the AI slop assets that come from it.
Saying "I don't care what you think don't use it" is pretty irresponsible and is why we have so many current arbitrary limits on creation. Make no mistake, when you see an error in content manager that your model is too high poly or that some small material defect is causing you to spend time chasing it and you cant figure out why in the world these silly limits exist THIS is the reason, because people can't or don't want to put in the effort to follow content creation practices that would benefit everyone that either chooses to use the item or is stuck using them because someone else chose to. When they have to use the program to limit the content because people wouldn't listen, we all lose.
I am all for new creators, but if you want to purposely create low effort garbage with no intent to improve or listen to critique and simply have the attitude that if you don't like it don't use it then no you can just keep right on going. That isn't content creation and you are no content creator.
Recently I have been playing around with photogrammetry (AI) and have been quite impressed with its capabilities and ease of use. Like all new technologies I think it has a place in Trainz when applied with care. My initial approach has been to use the photogrammetry model as a reference for creating the model in Blender. But looking at the end result I left to wonder if the photogrammetry model with some tweaking might be the better of the two.
With photogrammetry one can catch the actual texture on the subject. In the case of the Red River Cart shown here this has some issues, in particular, the shadows cast by the lights in the museum where it was photographed. But that could be corrected. The original photogrammetry image was 29,841 polys but with Blender’s Decimate Modifier and a Ratio of 0.1 applied as shown here, the poly count was reduced to 2,984, comparable to the Blender model. When viewed at a distance, as it almost always would be in Trainz, that seems quite acceptable for a LOD0.
I have been writing a series of YouTube videos on the subject. Given the interest here, I have posted the first two on YouTube.
A note of caution. SnapMesh Pro is in beta at the moment and I have had issues with it not running on my Mac with its M1 chip. So, I would hold off on purchasing it. I am currently working with the developer with a pre-release version to sort out this problem.
However, the photogrammetry model I have created here was made with the free version which works well on my Mac running either Blender 3.6.5, 4.4.3 or 4.5.1.
Cayden
With photogrammetry one can catch the actual texture on the subject. In the case of the Red River Cart shown here this has some issues, in particular, the shadows cast by the lights in the museum where it was photographed. But that could be corrected. The original photogrammetry image was 29,841 polys but with Blender’s Decimate Modifier and a Ratio of 0.1 applied as shown here, the poly count was reduced to 2,984, comparable to the Blender model. When viewed at a distance, as it almost always would be in Trainz, that seems quite acceptable for a LOD0.
I have been writing a series of YouTube videos on the subject. Given the interest here, I have posted the first two on YouTube.
A note of caution. SnapMesh Pro is in beta at the moment and I have had issues with it not running on my Mac with its M1 chip. So, I would hold off on purchasing it. I am currently working with the developer with a pre-release version to sort out this problem.
However, the photogrammetry model I have created here was made with the free version which works well on my Mac running either Blender 3.6.5, 4.4.3 or 4.5.1.
Cayden