What triangle count should a stylized low-poly prop have?
For stylized props in a modern engine, most pieces land between roughly 500 and 5,000 triangles, and the number matters far less than the material count. A 2,800-triangle desk with one material is cheaper to render than a 400-triangle desk with four.
There is no single correct number, and anybody who gives you one without asking what the prop is has told you something about their confidence rather than about your budget. But the useful range is narrow enough to be worth stating: for a stylized prop in Unity or Unreal on desktop or current-generation console, roughly 500 to 5,000 triangles covers almost everything, with hero furniture at the top of that band and desk clutter at the bottom.
What actually costs you performance is a different number.
Draw calls, not triangles
A modern GPU eats triangles for breakfast. What it does not like is being interrupted. Every material switch is a draw call, and every draw call is CPU work that happens before the GPU does anything at all. Fill a room with sixty props at 2,000 triangles each and one material each, and you have 120,000 triangles and sixty draw calls — which batches down to far fewer if the props share a material or an atlas. Fill the same room with sixty props at 500 triangles and four materials each, and you have 30,000 triangles and 240 draw calls. The second room is a quarter of the geometry and four times the cost.
This is why our pack ships one material per asset and why we would rather spend triangles than materials. It is also the first question to ask about any pack you are evaluating: not “how low-poly is it” but “how many materials per prop, and do they share”.
Real numbers from a real set
Here is the distribution from the first zone of our 1960s Office pack — the 29 props with a measured web export — taken at export rather than estimated:
| Triangles | |
|---|---|
| The 29 measured props | 43,869 |
| Median prop | 764 |
| Smallest (book ends) | 72 |
| Largest (desk chair) | 5,074 |
The median is low because most of a themed set is small stuff — trays, book stacks, pen sets, wall art — and small stuff should be small. The pieces near the top are the ones with curvature you cannot fake: an upholstered armchair, a desk chair with a five-star base, a telephone with a moulded handset. Those are where triangles buy you something.
Notice that the largest pieces are a desk chair, a bench sofa and an armchair — not the bookcase, which is the biggest object in the room. The bookcase is 764 triangles because it is boxes. Curvature is what costs, not size.
How to set your own budget
Work backwards from the shot, not from a rule.
- How close does the player get? A prop the camera passes at two metres can be a quarter the density of one that sits on the desk the player leans over. Silhouette is what reads at distance; surface detail is what reads up close.
- How many will be on screen at once? Fifteen identical chairs in a conference room is an instancing problem, not a triangle problem — but it does mean the chair’s budget should be the tightest in the set.
- Is the shape convex or does it have holes? Repeating gaps — a slatted bin, a bookshelf full of books, a woven basket — will blow past your estimate every time. Budget those separately.
- Where does the silhouette break? Spend your triangles on the outline. Nobody has ever noticed a flat interior panel; everybody notices a faceted curve on the edge of a chair back.
The check that matters more than the count
A prop can be inside budget and still be unusable, because triangle count says nothing about whether the geometry is clean. Non-manifold edges, flipped normals, interpenetrating shells, a pivot in the middle of the mesh instead of on the floor, or a hole in the back panel that only shows from an angle the author never rendered — none of those move the triangle count and all of them cost you an afternoon.
That is the reason we audit every asset from five angles rather than checking a number. The count is trivially easy to hit. The geometry behind it is the part that goes wrong, and it goes wrong on the angle you were not going to look at.
Updated 17 August 2026 · LeftPunch Studio
Related
We write these while building. The pack they come from — a 1960s office set of 110 props plus a 30-module architecture kit — is in production and every prop's audit verdict is published, including the ones that failed.
See the pack