by ModelFunShipyard » Wed Jul 15, 2026 6:04 am
That is a really good question, one I've been asking myself for some time already. From what little I could gather in a quick research - I do not understand much about AI nor am interested in doing so - but it looks like so far it can generate seemingly decent quality models from 2D drawings, or a small prompt. Though these seem to mostly revolve around the world of gaming miniatures and the like. Personally, having been creating model kits for the better part of a decade, I believe the current obstacle that won't be easily overcome by AI is very simple: AI just can't think for itself.
To create a proper kit, not just some World of Warships model you loosely cut into pieces and stick in a printer, you need to analyze and create submodels of your subjects in your head while you are working on it, and think about how they interact with one another (how they align and how they can be put together, pins, or tabs, etc.). The second problem is references: not everything is digitized. AI can't use material that hasn't been digitized, but you or I, as human beings, can pick from a few lines in a book or an actual, archival drawing that AI doesn't have access to. With that comes the implicit understanding that you know what you're looking at, and what that item is supposed to do or move in real life. AI can access to all the knowledge on the internet, but whether it can correctly analyze certain bits and pieces and surmise what their purpose is, is another matter. Not all plans are labelled, like the Booklet of General Plans of the USN actually are. Third, and this is directly correlated to the previous point, a ship is a massive conglomerate of various systems put together, and not just that. The sheer amount of data you'd need to know and process as an AI to produce an accurate model kit, let alone a specific one (say Essex in 1965 exactly), is unfathomable right now. We also get back to point 1, which is do you have the data to surmise the correct appearance of Essex in that year or not? That is why we modellers have to do so much research beforehand, and because some items cost so much. If you don't account the amount of hours you spend trying to find info on the item you're designing, that's a good 20-50% (depending on subject) of the time you spend on a model just down the drain, because you're not getting return on investment on it. Fourth, sometimes there just isn't data available for some parts or ships, that have either been lost to time, or were never properly documented in the first place. How do you program an AI to behave in that instance? We can make a best guess based on our experience and knowledge, but with AI it's a shot in the dark, because while it may have ALL the knowledge at its fingertips, it doesn't have neither experience nor interpretation of knowledge. And I mean that in the most human sense of the concept. AI doesn't "learn" nor "think ahead" in the strictest sense of them, it just tries things and excludes what the algorythm tells it doesn't work because, ultimately, it isn't capable of critical thinking.
Also, does AI know and take into accounts the printer settings for a specific material? Because with a miniature you can get away with many things, with a ship that has so many fine and delicate details, you really cannot, at the risk of making them grossly out of scale. That's just a flaw inherent of modelling things with massive scale gaps (just compare a 1/700 model ship with a 1/72 tank. The tank will be much closer to its original counterpart than the ship).
That's my, admittedly, very limited and shallow take on AI in 3D modelling as it currently stands. I, too, would like to know your takes on it, if you have any, but I don't have any illusions, in that it will get there at some point, and when it does, it's going to put a lot more people out of a job than it's doing now. But for the moment, we soldier on.
That is a really good question, one I've been asking myself for some time already. From what little I could gather in a quick research - I do not understand much about AI nor am interested in doing so - but it looks like so far it can generate seemingly decent quality models from 2D drawings, or a small prompt. Though these seem to mostly revolve around the world of gaming miniatures and the like. Personally, having been creating model kits for the better part of a decade, I believe the current obstacle that won't be easily overcome by AI is very simple: AI just can't think for itself.
To create a proper kit, not just some World of Warships model you loosely cut into pieces and stick in a printer, you need to analyze and create submodels of your subjects in your head while you are working on it, and think about how they interact with one another (how they align and how they can be put together, pins, or tabs, etc.). The second problem is references: not everything is digitized. AI can't use material that hasn't been digitized, but you or I, as human beings, can pick from a few lines in a book or an actual, archival drawing that AI doesn't have access to. With that comes the implicit understanding that you know what you're looking at, and what that item is supposed to do or move in real life. AI can access to all the knowledge on the internet, but whether it can correctly analyze certain bits and pieces and surmise what their purpose is, is another matter. Not all plans are labelled, like the Booklet of General Plans of the USN actually are. Third, and this is directly correlated to the previous point, a ship is a massive conglomerate of various systems put together, and not just that. The sheer amount of data you'd need to know and process as an AI to produce an accurate model kit, let alone a specific one (say Essex in 1965 exactly), is unfathomable right now. We also get back to point 1, which is do you have the data to surmise the correct appearance of Essex in that year or not? That is why we modellers have to do so much research beforehand, and because some items cost so much. If you don't account the amount of hours you spend trying to find info on the item you're designing, that's a good 20-50% (depending on subject) of the time you spend on a model just down the drain, because you're not getting return on investment on it. Fourth, sometimes there just isn't data available for some parts or ships, that have either been lost to time, or were never properly documented in the first place. How do you program an AI to behave in that instance? We can make a best guess based on our experience and knowledge, but with AI it's a shot in the dark, because while it may have ALL the knowledge at its fingertips, it doesn't have neither experience nor interpretation of knowledge. And I mean that in the most human sense of the concept. AI doesn't "learn" nor "think ahead" in the strictest sense of them, it just tries things and excludes what the algorythm tells it doesn't work because, ultimately, it isn't capable of critical thinking.
Also, does AI know and take into accounts the printer settings for a specific material? Because with a miniature you can get away with many things, with a ship that has so many fine and delicate details, you really cannot, at the risk of making them grossly out of scale. That's just a flaw inherent of modelling things with massive scale gaps (just compare a 1/700 model ship with a 1/72 tank. The tank will be much closer to its original counterpart than the ship).
That's my, admittedly, very limited and shallow take on AI in 3D modelling as it currently stands. I, too, would like to know your takes on it, if you have any, but I don't have any illusions, in that it will get there at some point, and when it does, it's going to put [i]a lot[/i] more people out of a job than it's doing now. But for the moment, we soldier on.