Hello. I got to a point where next step is loading and using meshes. I read on the formats and what kind of data do meshes actually have (and how many details there are), but i cant wrap my head around it (especially the animated ones). Ultimately i want to prepare my own format, but i would like to be sure that I clearly understand the static ones first.

So to load static mesh i obviously need xyz of each vertex, then face information (its equivalent of index buffer right?) and uv of texture. Vertex normals are used for light effects only (is it better to have vertex or face normals?). Subsets are faces having the same texture, and rendering process is: load the vertex/index buffer and in loop for each subset load the texture and draw the whole subset, and thats basically it to load the mesh?

I want to implement skeletons in future too. I would like to have a format where i dont have to do ANY calculations at runtime - the exported file should have all the info. I dont need anything fancy - just animations in future and basic light. The thing that shocks me is why do people use ascii formats? Loading such is way slower than binary.

And to existing formats - i went through some but they are poorly documented, or ascii, or too "heavy". I am fully aware that i do not know anything about skeletal meshes right now but im catching up. But the format choice is the point where i am stuck

I'm on my way to work, so I don't really have time to reply and I may forget to reply later. So, what I can say quickly is you may want to check out my tutorial project on my website. The last example there is a bare bones game engine that can load models for rigid animation. Included in this DX11 Visual Studio Project (2013 I think it was), is the custom model exporter I wrote in Python to extract the model data from Blender.

You may also want to check out my HLSL series on my YouTube channel because that basically explains what data your model file actually needs to contain and why in at least an indirect way. Ultimately, your model data is just feeding the shader. So, if the shader doesn't use it, it's wasted at least in terms of what gets drawn to the screen. The HLSL tutorial doesn't do normal mapping, but it gets you right up to that point. It may be a bit more intermediate than beginner though. So, it may be one of those deals where you can watch it and learn what you can, go program for 6 months, and then come back and watch it again to get the most out of it.

You didn't say what platform you are on. The code is DX11. Except the HLSL stuff was written for XNA/MonoGame. Still, I have almost the identical shader to the last one in the series in my DX11 "engine". So, HLSL is pretty much the same on any platform and GLSL has totally different syntax but all the math is identical. And the math is like 90% of it, which is probably why you may want to watch the Matrix and Vector video before delving into the HLSL stuff.

Anyway, you pretty much need to master everything I've just given you before you even begin to worry about skinned shading, if you are going to do it yourself rather than just using an engine like Unity that does it for you. What I have there will do rigid animation (car wheels and simple stuff animated from code like that). Skinned animation uses everything you learn there and then makes it about 10 times more complicated.

Oh, I might also mention that my current (coding) project is to rewrite my whole engine in OpenGL 4.5 and GLSL. I'm about 20% done. Hopefully, I'll have code to post, if not a video explaining it, by this fall. I'm in an art program to learn to do game art right now, and that's consuming the vast majority of my time though.

And the only reason you need a face/triangle normal is to figure out what your vertex normals are. After that, you can throw them away unless you have a special use case. My custom model format is called PUNC (Position, UV, Normal, Color) because that's the vertex data it includes. UV for textures which means you may not use or need the color (it tends to be one or the other). And a vertex normal. That's enough vertex data to do pretty much anything starting out.

How do you write an exporter? I suppose there is some API which gives you access to all the data of model in the editor program, but can i do it in c++ ?

Decided to create standalone c++ converted from .obj to my format. The only concern now are skeletons now :)

Thats exactly how i see these things. Already finished converter from obj files to my format :)

Thats exactly how i see these things. Already finished converter from obj files to my format :)

I have to wonder, if this is your approach why use a custom converter at all and not just stick with obj format and write an importer for your game/engine? Obj might not support everything you want but at that point you would have to drop out the obj step altogether and write an exporter for your desired modelling application. Blender uses python so you could easily write an exporter using that (if Blender is your thing).

Because of load time. My format is fully binary, the reader just goes through and instantly loads the mesh, got all the info ready. If i were using obj i would have to go through the ascii text, then count vertex normals everytime i load the mesh etc etc