Consider building a new open-world game. Approx. 20x20 km total area would work. No networking/multiplayer is ever planned for that project.
As far as I understand it, using the World Composition feature would allow designing in-editor all “sectors” as different levels/maps that stream in and out seamlessly both at runtime and at edit-time. Is that assumption roughly correct?
The big question for me here is Landscape. Right now I’m thinking the persistent level holds just exactly one single Landscape object 20x20 km off a 8k-by-8k heightmap. So there’s never any (manual) “stitching”/fitting and never more than one Landscape component consuming resources (rather than say up to 4 if the player is at a cross-section in between 4 neighboring sectors/levels)… is that the most reasonable approach?
Furthermore I’m inferring from the latest release notes that somehow Landscape support texture streaming, no idea how that would be worked out in practice. IIRC you assign a single material to the Landscape, but you might have 20+ ground textures (plus the same amount of normals) in total, however in practice no more than say 4 (plus normals) would need be sampled/shaded in any given frame. How to work it out which textures to stream in and out of VRAM seamlessly in a Landscape material context? Any tutorials on this that I may have missed?
Lastly, world position offsetting. Good idea to always enable? Or is that only for worlds that exceed MAX_WORLD_SIZE (previously 5km, now 20km)?