Worksflows, modularity, efficiency, tiling texture atlas with overlapping UVs & TIME
polycounter lvl 7
Hey all,
Maybe this can be a discussion.
A recent response of mine to another poster on this site has got me wondering about workflows, time and efficiency. In this case I'm talking environmental modeling, UVing and texturing in a modular/reusable/hero + tiling/atlas/overlapping UV sense. I've read as many articles and opinions as I can find - but almost all conclusions have been, well, non-conclusive. Here are some of the most helpful to me:
Joel Burgess & Nathan Purkeypile - Skyrim's Modular Level Design
Lee Perry - Modular Level Design
Scott Jones - Investigation into Modular Design Within Computer Games
I figure it's fair to say that there really is no 'best way' when it comes to completing large environments on a team and in a timely fashion - skill level/experience, communication, available time, budget, number of designers and artists - but I'm a firm believer that there is always a 'better way.' What do you think?
I guess what I'm looking for are opinions into (in this case, and modularity in general): is the time required to plan and UV then texture[atlas] a near modular environment WORTH it. So far my (first) personal attempts to stay in front of a tiling texture atlas with overlapping UVs environment has been daunting. I've done environments with modular assets and tiling textures and simple texture atlases, but nothing like this. Everything from modeling each asset with consideration to polygon placement and its' scale on a texture atlas to overlapping UVs onto tiling texture sections OF that evolving atlas has got my brain twisting, untwisting and twisting again. Power of two.... this model relative to that model... this side tiles into this and also into that... etc... etc...
...but this to my knowledge would be the most 'game' efficient method I can think of... ONE drawcall, smallest final texture resolution, relatively smallest polygon count (not vertex count, however)... but unless this can become something close to motor memory, I don't see this paying off in terms of time-efficiency... but it does in terms of real-time graphics, processing and loading times...and instead of going back and reducing polycounts and texture resolutions, why not do it the first time around?
If you're still wondering what the hell I'm talking about (lol), this is a section from the post to the other forum-user.
"What I mean is that you'll have to split your roof's UVs into separate (but identical, or close to it) overlapping shells. You can have a roof texture take up any percent of the entire texture as long as the left side continues into the right. It's a tiling texture within a texture atlas and it only works if your asset is modeled and UV'd in the correct way. Now you can have your trim texture share the roof texture as well AND it can tile or simply be the resolution you need. I'll attach a picture of something I'm working on to show what I mean.
The Mesh w/ WIP Texture
The UV's:
Right side X lines up with left side X and same with top to bottom X (a tile) - the difference is that you can 'cut' this texture out and put it into a larger texture atlas; just as long as the UVs stay overlapping where they need to
The drawback to this method of splitting a mesh into more UV shells means that there are going to be more vertices (two UV shells sharing the same vertex means one vertex on each shell). However, from discussions in the tech forum, games are hitting polycount caps way before they are hitting vertex or texture resolutions caps. The only place that you'd need to consider vertex count is in mobile gaming."
So yea..... pretty open-ended discussion here, but also, am I insane for trying this modeling/texturing method? Please post what you think about any of the various subjects throughout; the more insights, personal experiences and opinions the better....!!!
Maybe this can be a discussion.
A recent response of mine to another poster on this site has got me wondering about workflows, time and efficiency. In this case I'm talking environmental modeling, UVing and texturing in a modular/reusable/hero + tiling/atlas/overlapping UV sense. I've read as many articles and opinions as I can find - but almost all conclusions have been, well, non-conclusive. Here are some of the most helpful to me:
Joel Burgess & Nathan Purkeypile - Skyrim's Modular Level Design
Lee Perry - Modular Level Design
Scott Jones - Investigation into Modular Design Within Computer Games
I figure it's fair to say that there really is no 'best way' when it comes to completing large environments on a team and in a timely fashion - skill level/experience, communication, available time, budget, number of designers and artists - but I'm a firm believer that there is always a 'better way.' What do you think?
I guess what I'm looking for are opinions into (in this case, and modularity in general): is the time required to plan and UV then texture[atlas] a near modular environment WORTH it. So far my (first) personal attempts to stay in front of a tiling texture atlas with overlapping UVs environment has been daunting. I've done environments with modular assets and tiling textures and simple texture atlases, but nothing like this. Everything from modeling each asset with consideration to polygon placement and its' scale on a texture atlas to overlapping UVs onto tiling texture sections OF that evolving atlas has got my brain twisting, untwisting and twisting again. Power of two.... this model relative to that model... this side tiles into this and also into that... etc... etc...
...but this to my knowledge would be the most 'game' efficient method I can think of... ONE drawcall, smallest final texture resolution, relatively smallest polygon count (not vertex count, however)... but unless this can become something close to motor memory, I don't see this paying off in terms of time-efficiency... but it does in terms of real-time graphics, processing and loading times...and instead of going back and reducing polycounts and texture resolutions, why not do it the first time around?
If you're still wondering what the hell I'm talking about (lol), this is a section from the post to the other forum-user.
"What I mean is that you'll have to split your roof's UVs into separate (but identical, or close to it) overlapping shells. You can have a roof texture take up any percent of the entire texture as long as the left side continues into the right. It's a tiling texture within a texture atlas and it only works if your asset is modeled and UV'd in the correct way. Now you can have your trim texture share the roof texture as well AND it can tile or simply be the resolution you need. I'll attach a picture of something I'm working on to show what I mean.
The Mesh w/ WIP Texture
The UV's:
Right side X lines up with left side X and same with top to bottom X (a tile) - the difference is that you can 'cut' this texture out and put it into a larger texture atlas; just as long as the UVs stay overlapping where they need to
The drawback to this method of splitting a mesh into more UV shells means that there are going to be more vertices (two UV shells sharing the same vertex means one vertex on each shell). However, from discussions in the tech forum, games are hitting polycount caps way before they are hitting vertex or texture resolutions caps. The only place that you'd need to consider vertex count is in mobile gaming."
So yea..... pretty open-ended discussion here, but also, am I insane for trying this modeling/texturing method? Please post what you think about any of the various subjects throughout; the more insights, personal experiences and opinions the better....!!!
Replies
I also wonder that in modern engines how much more efficient to have several meshes mapped to the same texture, usually with heavily overlapping faces, instead of using substance painter to paint all of them individually...
Substance... yes, still valid to use with packed textures. You just need to adapt your outputs to fit.
I'm curious if there are nodes for packing tiled sub-textures into a larger atlas, and especially for adding gutters between them (to prevent mip bleeding). Otherwise, you need to take the output textures and pack them yourself in Photoshop.
Several meshes... yes, this is still valid. RAM is still limited, so re-using textures is essential. Unless you're using virtual textures, like used for Quake Wars and Rage: http://wiki.polycount.com/wiki/MultiTexture#Megatexture. Most games don't use this technique however, takes too much time to generate all that uniqueness.
I think I just need to be fix with measure and I can start some cool Modular asset be untill that I can't do something weld !
Thanks having interest !
We have a good section on modular design in our wiki. http://wiki.polycount.com/wiki/Modular_environments