Do low-poly props need LODs?

Usually not, and the honest version of this answer is one that most asset packs will not give you, because “includes LODs” is a bullet point that sells.

An LOD chain earns its place when the geometry it removes costs more than the machinery it adds. On a stylized prop of a few hundred triangles, it frequently does not.

What an LOD actually costs

It is not free. Each level is:

• more memory and disk — two extra meshes per asset, resident whether or not they are drawn;
• more import and build time, multiplied by every asset in the pack;
• a swap the engine has to evaluate every frame per instance, plus a pop the player may notice;
• another thing to get wrong — an LOD1 with a different pivot, a missing UV set, or a silhouette that changes at the transition is a bug that only appears at one specific camera distance.

Against that, the saving is triangles the GPU was not struggling with. Refer back to why materials matter more than triangles: if your scene is CPU-bound on draw calls, LODs save you nothing at all, because the draw call happens whichever level is selected.

When they do pay

• High instance count. Forty chairs in a hall, a hundred books on shelves, street furniture down a whole block. The saving multiplies; the cost does not.
• Above a few thousand triangles. A 4,000-triangle armchair at the far end of a room genuinely does not need 4,000 triangles. A 400-triangle tray does not care.
• Large screen-space variation. A prop the player sees both at arm’s length and from across the level.
• Anything with dense repeating detail — a bookcase full of individual books is a good LOD candidate even at a modest triangle count, because the detail is invisible at distance and the geometry is not.
• VR and mobile, where the budget is tighter and the win is proportionally larger.

When they do not

• Props under roughly a thousand triangles that appear a handful of times.
• Anything the player only ever sees close — desk clutter in a first-person scene.
• Scenes that are already CPU-bound. Fix your batching first; the LODs will not help.
• Props with a silhouette so simple that LOD1 is barely different from LOD0. If you cannot make a meaningfully cheaper version, there is no version to make.

What to check in a pack that advertises LODs

• The triangle counts per level. If LOD1 is 90% of LOD0, the chain is decoration. A useful LOD1 is roughly half; LOD2 is roughly a quarter.
• The pivot is identical across levels. A shifted pivot makes the prop jump at the transition.
• The silhouette holds. Decimation that eats a chair’s armrest produces a visible pop, and the fix is authoring the level rather than turning a slider.
• Whether the collision follows. Collision should stay on LOD0’s hull and not swap.

What we ship, plainly

Our pack ships an LOD1/LOD2 chain where the budget makes it worth having, following the standard _LOD suffix convention that Unreal and Unity both read.

For a good part of the set — the median asset is 672 triangles — that chain is a convenience rather than a saving, and we would rather say that than let the bullet point imply a performance win it does not deliver. Where it matters is the upholstered pieces and the bookcase: the armchair, the desk chair, the telephone and the book-filled shelves are the assets where LOD1 does real work.