In short: describe the asset and the job it has to do - silhouette, real-world size, poly budget, art style - and the assistant generates it in the scene. From there it is an ordinary Blender object: retopologise it down to budget, texture it by describing surfaces, check pivot and scale, export FBX or GLB. Kits stay consistent when the style sentence stays the same, and nothing about the workflow is locked - every asset remains editable, re-exportable and yours.
What prompt-driven asset work covers
- Props and weapons - crates, barrels, swords, tools, pickups, at the poly count you name.
- Modular environment pieces - wall sections, floors, stairs, door frames generated to a grid so they snap together.
- Characters - stylised humans, creatures and robots with clean silhouettes, ready for the rigging workflow.
- Style consistency - a family of assets requested with one style sentence reads as one kit.
- In-scene iteration - "make the blade wider" edits the object you keep, not a copy stuck on a web tab.
Write an asset brief that lands
A game brief has four ingredients: what it is, how big it is, what it is allowed to cost, and how it should look. Compare:
| Vague prompt | Prompt that lands |
|---|---|
| Make a sword | A low-fantasy sword, 1.2 m long, chunky readable silhouette, about 1,000 triangles, a single 1024 texture, hand-painted look |
| A crate | A sci-fi supply crate, 0.8 m cube, hard edges with slight chamfers, flat-coloured panels, game-ready at about 600 triangles |
| An adventurer | A stylised adventurer character, 1.7 m tall, readable silhouette from 10 m away, A-pose, flat colours, ready for a rig |
| Some houses | Three modular medieval house wall pieces, 4 m wide, snapping to a 1 m grid, wood and plaster, low-poly |
The asset pipeline, in-scene
01 - Generate against the brief. Name the budget in the request; treat the result as the high-quality starting point, not the shipping mesh.
02 - Judge the silhouette first. Step back to game distance - the camera the player actually plays from - before judging detail. A prop that reads at 10 m needs no detail that only reads at 50 cm.
03 - Bring it to budget. Run retopology or a decimate pass to the triangle count the platform allows, keeping UVs intact.
04 - Texture for the renderer. Describe the surfaces ("scratched steel, wooden grip, leather wrap") and finalise channels: packed ORM where the engine expects it, resolution to match the budget.
05 - Export and verify. FBX or GLB, real-world scale, applied transforms, pivot where the engine wants it - then a round-trip check in a test scene before it joins the project.
Poly budgets worth aiming for
- Hand props and pickups - 300 to 1,500 triangles.
- Weapons - 1,000 to 3,000 triangles for first-person readability.
- Environment modules - 500 to 3,000 triangles per piece, designed to repeat.
- Stylised characters - 5,000 to 20,000 triangles; more only if the camera earns it.
These are starting points, not rules - the game's platform and camera decide. State the target in the brief, then verify it in the mesh statistics panel the way you would for any asset.
The export checklist
- Scale in metres - 1 unit equals 1 metre keeps engines and physics happy.
- Applied transforms - rotation and scale applied before export; nothing screams "imported asset" like a 0.0137 scale factor.
- Sensible pivots - at the base for props, at the grip for weapons, at the origin for modules.
- Names that travel - clear object and material names survive the round trip and keep the outliner readable at 2 a.m.
Where hand work still wins
- Art direction - the palette, the proportion language and the one silhouette that makes the game look like itself.
- Hero assets - the flagship character or weapon the store page zooms into deserves a sculptor, not a prompt.
- Platform optimisation - the final LOD chain, occlusion data and platform-specific trims are profiling work, not generation work.
- Animation authoring - generation gets you an asset; making it feel good in the hands is still animation craft - start with the walk cycle.
From asset to game
A prop is a pipeline in miniature: generate, budget, texture, export. Once that loop is comfortable, the same session extends naturally - animate a character with a walk cycle, keep a family of assets stylistically tight with the low-poly workflow, or take the Roblox export path for a platform that ships avatars and UGC.