To model or not to model small details for baking

Hello everyone, thanks for attention
I’m trying to do my very first model, and straight off it is supposed to be lowpoly for using in realtime game engines etc. It is cartoon-styled WWII plane, inspired by BF Heroes models, but based on different real-world prototype (Yak-1).

I’m making highpoly right now. Everything was going smoothly, when I noticed that there are that small joints between metal panels which hull are made of. The question is: should I model them for baking normal maps and everything else, or it is more convenient to just draw them in an image editor after baking? I’m not quite sure that I can draw them in normal maps because of lack in experience, and I’m not sure that I can model them either, because it will make wireframe waaaaay more complicated. Picture:
Can anyone take a hint which way is better, and maybe there are some lessons about that?
By the way, sorry if there are grammar mistakes in the post.

In my experience, it’s far easier to concentrate on modeling things which determine the silhouette of your subject, and do the rest in the texture. Panel breaks are an excellent example. If you concentrate on modeling the form of the plane, you’ll be able to paint in the fine details later, rather than fussing with them in 3D. Much faster and easier, in my experience.

Now, that does leave you with the question of normal maps, which lucky for you, many people have come before and answered. Generating normal maps from B/W images is trivial to do for most serious 2D apps. I use PS, and that’s all I know, so I would steer you to the NVIDIA filter which works without a hitch (unless you’re trying to use it within an action). If you’re not using PS, google for a normal map solution in whichever app(s) you’re using. Also, Keep your panel breaks on their own layer, so it’s very easy to convert to normals when you’re wrapping things up.

As for combining your detail nmap with your baked nmap, there are multiple choices and methods on how to do this. Some are arguably better than others, and there’s a lot of math to back it all up, but for the most part the end results are hard to distinguish from one another, outside of extreme examples. If you’re not picky (and using PS) it might be fine to just google for a PS action and call it good. However, there’s a nice, mathematically correct, option you can find for blender, here.

Good luck

Thank you a lot for good tips, friend.

If actual geo is wanted then: