Every indie level design guide skips the part that actually eats a solo dev's month: the level isn't short on gameplay, it's short on stuff. A corridor needs shelves, crates, pipes, rubble, signage, and forty other things nobody will consciously notice unless they're missing — and no two-person team is commissioning forty custom props for one corridor. This is exactly the gap indie game dev tools built around AI 3D generation are trying to close, and it's worth being honest about what that gap looks like before reaching for a tool to fix it.
Traditional options were already known quantities: hire a contractor per prop (too slow, too expensive at this volume), buy premade packs off a marketplace (licensing friction, and everything looks like it came from a marketplace), or model it yourself (fine for ten props, brutal for four hundred). AI prop generation is a real fourth option, but it only works cleanly if it slots into the production pipeline your level designers were already using — greybox first, art later — rather than replacing that pipeline outright.
TL;DR — the questions people actually ask
| Question | Direct answer |
|---|---|
| When do I generate props — before or after blocking out the level? | After. Lock the layout in greybox first; generate once the space is proven to work. |
| How do I keep dozens of separate AI generations from looking like a mismatched pile? | One written style brief plus 2-3 locked reference images pasted into every prompt. |
| What's the honest failure mode of AI props at scale? | Style drift across separate generations, plus texel-density mismatches against hand-made assets. |
| Do I need to retopologize everything AI makes? | No — only hero props and anything the camera lingers on. Background clutter ships as-is. |
| Can I use marketplace packs instead? | You can, but licensing terms vary pack to pack and mixed-source props rarely match stylistically. |
| Is this good enough for a full prop library on a tiny budget? | For background density, yes. For signature hero props, plan on a manual finishing pass. |
What's actually hard about filling a level with props?
Level design tutorials spend most of their time on layout, pacing, and encounter design — for good reason, that's the part that determines whether the level is fun. But a shipped level also needs enough visual density that it doesn't read as an empty box, and that density requirement scales with square footage in a way solo devs chronically underestimate.
The 2026 State of the Game Industry report, based on a survey of over 3,000 developers conducted by Omdia for GDC, found that 21% of respondents now work entirely solo — up from 18% the year before, and more developers report working alone than at studios of 500-plus people. That's the structural reality behind the prop problem: a growing share of the people shipping games have nobody to hand "model forty crates" to.
Marketplace asset packs exist precisely because this problem predates AI generation. But they carry two real costs the storefront page won't mention. Licenses vary pack to pack — some restrict resale-adjacent use, some require attribution, some forbid modification — so a "prop library game dev" strategy built on stacking packs from different sellers means auditing several separate EULAs before you ship. And visually, a shelf from one artist's pack next to a barrel from another's rarely reads as one coherent world; players may not name the problem, but "this looks like an asset flip" is a common, real criticism of cheaply assembled indie levels.
What is the greybox-to-final-art pipeline, and where do AI props actually fit?
Greyboxing — also called blockout or graybox — is standard practice across professional level design, not an indie shortcut. The Level Design Book describes it as building a level with primitive shapes that represent walls, cover, and pathways so you can playtest the layout before spending any art budget on it. Epic's own Unreal Engine documentation frames blockout the same way: rough geometry standing in for the eventual level, built fast enough that throwing it away costs nothing.
Robert Yang — a game designer and NYU professor — put the underlying discipline plainly in his writeup on the process: the greybox stage should stay in "very basic shapes and polygons" deliberately, because investing in finished art before the layout survives playtesting means you'll be reluctant to cut geometry that isn't working. Unreal's own tech blog on Kine's development shows this in practice — the studio's level layouts changed drastically in composition and theme between blockout and final art pass, which is only survivable because nothing expensive was committed yet.
Where AI generation actually slots in
This is the part most "AI for game dev" pitches gloss over: AI prop generation belongs at the art pass stage, not the blockout stage. Trying to generate finished props while you're still moving walls around defeats the entire point of greyboxing — you'll regenerate the same crate three times as the room dimensions around it keep changing.
The workflow that actually holds up:
- Block the level in primitive shapes — cubes standing in for crates, cylinders for barrels, flat planes for shelving.
- Playtest the layout. Cut, extend, and rearrange freely, because nothing has art invested in it yet.
- Once the layout is locked, walk the greybox and list every prop slot by rough silhouette and scale — this becomes your generation shot list.
- Generate final props against that list, matching each greybox stand-in's approximate dimensions.
- Swap greybox for final art incrementally, one room or one prop category at a time, so you can catch scale and style problems before they propagate across the whole level.
Our text-to-3D workflow guide covers the generation and import-checklist side of step 4 in more depth — poly count, UV seams, pivot placement — all of which matter more once you're committing dozens of props to a real scene rather than testing one in isolation.
How do you keep AI-generated props visually consistent across a whole level?
This is the practical crux of using AI at prop-library scale rather than for a single hero asset, and it's a genuinely solvable problem — it just takes more discipline than firing off forty unrelated prompts.
Write a style brief before you generate anything. Two or three sentences fixing material palette, edge treatment (hard-surface vs. worn/organic), and stylization level (realistic, painterly, low-poly) gives you a document to paste into every prompt rather than re-deciding tone each time. Inconsistent tone across props is the single most common tell that a level's dressing came from scattered, undirected generations.
Anchor every batch to the same 2-3 reference images. Whether you're working from concept art or photos of a real object, feeding the same visual anchors into each generation constrains color grading and proportions far more reliably than text alone. Our image-to-3D guide and concept-art-to-3D guide both cover how to prep reference material so the model reads it consistently instead of drifting toward its own defaults.
Generate in themed batches, not one prop at a time. A batch of "rusted industrial containers" generated back-to-back from the same brief tends to hang together better than the same containers generated across scattered sessions days apart, simply because you're actively holding the style brief in mind while reviewing output.
Do a pass-level color and scale check before final placement. Even with a style brief, individual generations will vary slightly in saturation and proportion — a five-minute review pass across the whole batch, discarding or regenerating outliers, catches drift before it's baked into a shipped level.
Where do AI 3D props for games still fall short?
Being straight about the limits matters more than another list of upsides, because these are the failure modes that show up after you've already committed art time to a level.
Style drift is real even with a brief. A written style brief and locked references narrow the variance considerably, but they don't eliminate it — expect to discard or regenerate a meaningful fraction of any batch, and budget review time for that rather than assuming every output ships as-is.
Texel density mismatches are easy to miss until props sit side by side. AI generators typically output textures at a fixed resolution regardless of the object's actual in-game scale, so a small prop can come out looking oversharp while a large one looks blurry once placed next to hand-authored assets tuned to your project's standard pixels-per-meter. This isn't a defect in the mesh — it's a mismatch you have to catch and correct with a resize/re-bake pass, the same way you'd standardize any asset from an outside source.
Hero props still need a human hand. Background clutter that reads at a glance from ten feet away tolerates rough topology and minor asymmetry just fine. A prop the camera lingers on, or one that deforms, or one central to a puzzle or interaction, benefits from manual retopology — auto-generated meshes optimize for matching the reconstructed surface, not for clean, animation-friendly edge flow.
Licensing still needs a read-through, same as marketplace assets. Generating instead of buying a pack doesn't automatically simplify the licensing question — terms differ by tool and by whether you're on a free or paid tier. We break down what "commercial use" actually means across current tools in our AI 3D model licensing guide.
Which indie game dev tools should actually be in your prop pipeline?
A workable stack for a solo or two-person team doesn't need to be complicated — it needs the pieces above to actually connect to each other.
- A greybox/blockout tool — your engine's own primitive brushes (BSP in Unreal, ProBuilder in Unity) are enough; you don't need anything specialized for this stage.
- A text-to-3D or image-to-3D generator for the art pass once layout is locked — this is where bunpav fits, taking a prompt or a reference photo to a textured, exportable mesh with GLB, FBX, OBJ, or USDZ output. bunpav is currently in private beta with an open waitlist; the Studio credit pack ($39 for 550 credits) is explicitly framed for "a jam weekend or a small prop library," which maps directly onto the level-dressing workload this guide is about. See current tiers on the pricing page.
- A format reference so exports actually import cleanly — our GLB vs. FBX vs. OBJ vs. USDZ breakdown covers which format to reach for depending on target engine and whether a prop needs rig data at all (most static props don't).
- An engine-specific import checklist — our Unity/Unreal import guide covers what commonly breaks (materials, scale, collision) when a generated mesh crosses from generator to engine.
None of this replaces a real environment artist on a funded team. It replaces the option a two-person team never actually had — hiring one.
What should you do this week if you're filling a level solo?
- Finish and playtest your greybox first. Resist the urge to generate a single final prop until the layout has survived at least one playthrough where you deliberately try to break it.
- Write a one-paragraph style brief covering material palette, edge treatment, and stylization level, and save it somewhere you'll actually reuse it.
- Pick 2-3 reference images — photos or concept art — that represent the tone you're going for, and reuse the same set across every prop in that area of the level.
- Walk your locked greybox and write a shot list of every prop slot with its rough scale and silhouette, so generation has a concrete target instead of a vague vibe.
- Generate in themed batches, review for drift and texel-density mismatches as a batch, and only commit the winners to the level.
- Reserve manual cleanup time for hero props only — anything the camera lingers on, anything that deforms, anything central to an interaction. Let background clutter ship as generated.
- Read the license terms for whatever tool you used before you assume a shipped, monetized build is in the clear.
Related reading
- Text-to-3D AI: How to Turn a Prompt Into a Game-Ready Model
- Image-to-3D: Turn a Single Photo Into a Textured 3D Model
- From Sketch to Mesh: Turning Concept Art Into 3D Game Assets With AI
- GLB vs FBX vs OBJ vs USDZ: Picking the Right Format for Game-Ready 3D Assets
- AI-Generated 3D Models in Unity and Unreal: What Actually Survives the Import
- Can You Sell AI-Generated 3D Models? Commercial Use and Licensing in 2026
- The Game Jam 3D Asset Pipeline: Generate, Rig, and Export in Under an Hour
- Top 10 Friendslop Games in 2026: The Co-op Chaos Your Group Chat Actually Plays — the same tiny-team, chaotic-scope constraints show up across the friendslop wave, just applied to gameplay systems instead of level dressing.
Workflow descriptions, cited surveys, and pricing are accurate as of publication (July 17, 2026). bunpav remains in private beta; check bunpav.com for current waitlist and tier details before making a production decision based on pricing here.