I’m sure that there are tools to automate some of the work, but my understanding is that in most cases modelling artists want some kind of control over the generated LODs to ensure they don’t look like shit. Removing vertices on a 3d textured object is not nearly as simple as scaling a 2d picture as far as I understand it. You need to avoid mismapped textures, clipping vertices, the wrong missing details causing obvious pop-in, etc. A triangle in one place can be redundant but another triangle elsewhere may be a critical detail whose removal will be obviously missing from a distance (for example if you model the white house, you really want to keep the small flagpole up top at ALL levels of detail, but automated systems might remove it).
TBF part of the problem is that modern graphics cards mostly can shrug off insane amounts of geometry and badly optimized models, so management must have heard “high prio but not strictly blocking for release” and said “put it in the backlog” (aka “lmao whatever nerd I don’t care then, please focus on Marketing’s feature list happy please and thank you”).
Yeah it’s a super hard problem, but my point is that in the specific case of a city building game most of the models are going to be extremely small on the screen 90% of the time, so you can get away with pretty terrible automated LOD in a way you can’t in most other genres; you just don’t swap to low-poly versions until stuff is extremely far away.
I was actually just thinking they should do this as an interim solution and then make optimized LODs to replace the automatic ones with. In fact, perhaps that would have been the thing to do instead of making advance apologies. Slightly suboptimal image quality at release would probably have been more tolerable than heinously bad performance.
I’m sure that there are tools to automate some of the work, but my understanding is that in most cases modelling artists want some kind of control over the generated LODs to ensure they don’t look like shit. Removing vertices on a 3d textured object is not nearly as simple as scaling a 2d picture as far as I understand it. You need to avoid mismapped textures, clipping vertices, the wrong missing details causing obvious pop-in, etc. A triangle in one place can be redundant but another triangle elsewhere may be a critical detail whose removal will be obviously missing from a distance (for example if you model the white house, you really want to keep the small flagpole up top at ALL levels of detail, but automated systems might remove it).
TBF part of the problem is that modern graphics cards mostly can shrug off insane amounts of geometry and badly optimized models, so management must have heard “high prio but not strictly blocking for release” and said “put it in the backlog” (aka “lmao whatever nerd I don’t care then, please focus on Marketing’s feature list happy please and thank you”).
Yeah it’s a super hard problem, but my point is that in the specific case of a city building game most of the models are going to be extremely small on the screen 90% of the time, so you can get away with pretty terrible automated LOD in a way you can’t in most other genres; you just don’t swap to low-poly versions until stuff is extremely far away.
I was actually just thinking they should do this as an interim solution and then make optimized LODs to replace the automatic ones with. In fact, perhaps that would have been the thing to do instead of making advance apologies. Slightly suboptimal image quality at release would probably have been more tolerable than heinously bad performance.