Krea 2 can make all sorts of interesting objects, including this one, and it did just that for my dataset.

Why produce a LoRA for that which a model can already generate?

Primarily for the efficient use of tokens and then for things that are not well known I can fabricate a goofy looking semblance of it, use i2i maybe and/or just send it through a first pass training to get rid of the artifacts and use that LoRA to generate a dataset.