Hyper quality but still efficient model LOD logic

[mcmonkey4eva]

So there's something I think some high end engines might do (I saw effects that might be this in GTAV)...

What if LOD'd objects, instead of or in addition to having some cheap one-off set of textures to apply,
had a sprite-ish prerender from the player's view, updated once per (given number of frames, or real-time duration).

This would involve:

• A texture for every model to placeholder it (can be fairly small) (Possibly in an array for mass-rendering?)
• A spriting system of some form (Possibly mass-rendered if possible)
• A significant update detector, probably with a minimum time duration (Player moves relative to object, or object moves relative to player...)
• A system to determine what LOD mode to use in any given situation (cross-trick LOD (what we already use), this new system, or no LOD effects)

[mcmonkey4eva]

Oh, this would probably work best with a diffuse+normal+whateverelse texture set, as opposed to pre-crunched lighting.
Would also require some weirdness to get shadows working right (Maybe an alpha texture from shadow viewpoints?)