Your first retopo prompt and the triangle budget
Tell the agent to "make it low-poly" and it will either decimate the model into a crumpled silhouette or shave off two hundred triangles and call it optimised — "low" is a direction, not a target, and the agent has no idea which one you meant.
Choose your track
Read this module through your lens
“Low-poly” is a direction, not a target
A product model from Course 1 might be hundreds of thousands of triangles — fine for a still render, fatal in a game where dozens of objects must draw sixty times a second. So the first job of the game stage is to cut polygons. But “make it low-poly” tells the agent only which way to move, not how far, and not what to protect on the way down.
Given no target, the agent guesses. Sometimes it runs an aggressive decimate and the silhouette collapses into a melted lump. Sometimes it trims a few hundred faces and reports success while the model is still far too heavy. Both are “lower.” Neither is what you needed.
Optimisation becomes a real instruction the moment you supply two things: a triangle budget (the number) and the silhouette (the thing that must survive the cut).
The case study: an office chair to 3,000 tris
”Make this office chair low-poly for a game.”
The agent ran a 0.1 decimate. The chair dropped to 1,400 triangles — but the seat edge is now jagged, the cylinder base is a hexagon, and the armrests have caved in. The outline, the only thing a player sees at distance, is wrecked. Meanwhile the hidden underside of the seat kept its dense geometry.
”Retopologise to no more than 3,000 triangles. Preserve the silhouette: keep the seat and back outline smooth and the base cylinder round. Spend the budget on visible curves; delete interior and underside faces a player never sees. Report the final tri count.”
2,840 triangles, round base, clean outline. The report shows it deleted 600 hidden underside faces — exactly the polygons that should have gone first.
Where polygons earn their place
A triangle is worth keeping only if a player can see what it does. The budget should flow to silhouette-defining curves — the round base, the seat edge, anything that reads as an outline against the background. It should flow away from flat faces (a single quad needs no subdivision) and from anything never seen: undersides, interiors, contact points hidden by the floor.
So the inspection is two-part. First, read the final triangle count and confirm it is at or under budget — a number, not a claim. Second, compare the low-poly silhouette to the high- poly source from the camera angle players will actually use; if the outline still reads, the cut succeeded regardless of what happened to faces you cannot see.
“Low-poly done well” is invisible. That is the goal.
Hands-on exercise
Take a Course 1 model. Pick a triangle budget from a plausible use case (a hero prop vs. a background filler will differ by an order of magnitude). Prompt once with just “make it low-poly” and record the damage. Then prompt with your budget plus silhouette preservation, read the final count, and put the high and low silhouettes side by side. Note which faces the agent should have deleted first.
Second case · a barrel prop
The same lesson, a different object
Make this barrel low-poly.
Decimated to 900 triangles, but the rim is now an octagon and the staves are visibly faceted — the round silhouette is gone. Meanwhile the hidden inner cylinder kept its full density.
Retopologise to no more than 1,200 triangles. Keep the rim circle and barrel silhouette smooth; delete the unseen interior cylinder. Report the final triangle count.
1,150 triangles, round rim, clean outline. The report confirms it deleted the interior first — the polygons that should always go first.
Common failures · spot them fast
The failure gallery
Decimate destroys the silhouette (lumpy, melted outline)
Barely reduces (no real budget hit)
Strips visible detail, keeps hidden interior geometry
Reports "optimised" with no triangle number
Non-uniform, broken topology (stray tris, holes)
Each of these is caught by a quality gate — keep the cheatsheet open while you work.
Worked example
See the journey
The pattern
Cheatsheet
Cheatsheet
Prompt skeleton
Retopologise [object] to a game budget.
Target: <= [N] triangles. Preserve the silhouette and any read-critical edges.
Spend polygons on curved/visible areas; strip interior and unseen faces.
Report the final triangle count and the % change from the source.
Failure modes
- Decimate destroys the silhouette (lumpy, melted outline)
- Barely reduces (no real budget hit)
- Strips visible detail, keeps hidden interior geometry
- Reports "optimised" with no triangle number
- Non-uniform, broken topology (stray tris, holes)
Key operations
- Set and verify a triangle budget (target tri count)
- Decimate vs. remesh vs. manual retopo (choose per case)
- Preserve boundary/silhouette edges
- Delete interior + never-seen faces
- Read final tri count, compare to budget
Quality gates
- Final tri count at or under budget?
- Silhouette still reads at the intended distance?
- Read-critical edges intact?
- No holes / stray geometry introduced?
Workflow steps
- Set a triangle budget from the use case
- Prompt with budget + silhouette-preservation
- Read final tri count vs. budget
- Compare silhouette to the high-poly source
- Refine where the outline broke; log counts
Next module
- game_uv_unwrap — UVs the agent quietly gets wrong.
Reflection
Reflection card
Active retrieval — answer from memory before re-reading. Saved to this browser.
You've completed this module when…
- A retopo run with a stated triangle budget and a reported final count.
- A silhouette comparison (high vs. low) showing what survived.
- One case where the agent broke the silhouette, and the fix prompt.
Next: game_uv_unwrap — UVs the agent quietly gets wrong.
Finish — back to Game Asset →