Materials, bevels, and making it read as real
The agent's model looks fake for a reason you can name in one word — edges. Real objects have no perfectly sharp edges; the agent leaves every edge razor-sharp because nothing in the prompt told it that perfection is the tell.
Choose your track
Read this module through your lens
Why “realistic” means nothing to an agent
You look at the agent’s chair and immediately feel it is fake, but if asked why, you might say “it just looks CG.” Push past that. Almost always the tell is the edges: in the real world nothing is perfectly sharp, because manufacturing, wear, and physics all round edges by a fraction of a millimetre. That tiny radius catches a highlight, and your visual system reads “real.”
The agent leaves every edge mathematically perfect because “make a chair” never said otherwise. “Make it realistic” does not help either — it is an adjective, and the agent will happily report success without changing a single edge.
Realism in a static product decomposes into three concrete operations: bevel the hard edges, assign materials per part, and vary the roughness. Name those, and the fake look collapses.
The case study: a kitchen kettle
”Make this kettle look realistic.”
The agent reports it “applied a realistic metallic material.” Nothing about the geometry changed. Every edge is still razor-sharp, the whole body is one material slot, and the roughness is a single flat 0.4. It reads like a render of a CAD file.
”Bevel all hard edges at 1.5 mm, 2 segments. Assign three materials: brushed steel body, matte black handle, glossy chrome spout ring. Give the steel a roughness that varies slightly across the surface. Report the material slot count and confirm the bevel was applied.”
Three slots, edges now catch a highlight, handle reads as a different material. The chrome ring is too wide — one targeted fix — but it now reads as an object, not a diagram.
The inspection: look at an edge under a light
Do not judge realism from a flat orthographic thumbnail — it hides exactly the cues that matter. Put a light in the scene and look at a hard edge. A real edge shows a thin bright line where the bevel catches the light; a fake edge shows a hard black-to-light discontinuity.
Then count material slots and compare to the real object: a kettle has at least a body, a handle, and a spout, so one slot is automatically wrong. Finally, check that roughness is not a single flat number — uniform roughness is what makes things look like injection- moulded plastic even when they should be brushed metal.
Three checks, three seconds. They catch the difference between “the agent says realistic” and “it is.”
Hands-on exercise
Take a model from an earlier module. Prompt once with “make it realistic” and record what the agent actually changes (often nothing structural). Then prompt with the three explicit operations — bevel width, per-part materials, roughness variation — and capture an edge close-up under a light, before and after. Note which operation moved the needle most.
Second case · a wooden side table
The same lesson, a different object
Make this side table look realistic.
One oak material is assigned to everything, including the metal legs. The tabletop edge is razor-sharp and the roughness is a single flat value. It reads as a CAD render, not a table.
Bevel the tabletop edge 2 mm, 2 segments. Assign two materials: oak top, matte-black steel legs. Vary the oak roughness slightly. Report the material slot count and confirm the bevel.
Two slots, the top edge now catches a highlight, the legs read as metal. Same three operations — bevel, per-part materials, roughness variation — that fixed the kettle.
Common failures · spot them fast
The failure gallery
Every edge razor-sharp (the
One material assigned to the entire object
Flat uniform roughness — looks like plastic
Agent claims "realistic" with no concrete change
Bevel applied so wide it rounds the whole form
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
Add realism to [object]:
Bevel all hard edges (width ~1-2 mm, 2 segments) so no edge is perfectly sharp.
Assign separate materials per part: [list parts + finishes].
Vary roughness across surfaces; no single flat value.
Report material slot count and whether a bevel modifier/operation was applied.
Failure modes
- Every edge razor-sharp (the
- One material assigned to the entire object
- Flat uniform roughness — looks like plastic
- Agent claims "realistic" with no concrete change
- Bevel applied so wide it rounds the whole form
Key operations
- Bevel modifier or bevel on hard edges (small width, few segments)
- Multiple material slots, assigned per face group
- Roughness variation (map, or per-material values)
- Shade smooth + auto-smooth angle for curved parts
Quality gates
- Are hard edges beveled (catch a highlight), not razor-sharp?
- More than one material where the real object has more than one?
- Does roughness vary, or is it a single flat value?
- Did the agent report slot count + bevel applied?
Workflow steps
- Name the parts and their real finishes
- Prompt for bevel + per-part materials + roughness variation
- Inspect edges under a light; check material slots
- Refine the one finish that reads wrong
- Log the before/after edge close-up
Next module
- product_variations_and_parametrics — generating a controlled family of variants.
Reflection
Reflection card
Active retrieval — answer from memory before re-reading. Saved to this browser.
You've completed this module when…
- A before/after with edges beveled and at least two materials assigned.
- The specific bevel width and material slots named in the prompt.
- One sentence on why edge treatment dominates perceived realism.
Next: product_variations_and_parametrics — generating a controlled family of variants.
Finish — back to Product Design →