AI 3D props for indie games: what works and what doesn't

The prop problem

An indie level needs dozens of props, and every one of them takes time to model, unwrap and texture. That's the work AI generation is best at taking off your plate, but not every prop is equally suited. Knowing which ones generate well saves you from fighting the tool on the ones that don't.

Props that generate well

These share a few traits: one solid object, a readable silhouette, and surfaces that are opaque.

• Containers: crates, barrels, chests, sacks, baskets, jars.
• Furniture: chairs, tables, beds, shelves, wardrobes, thrones.
• Natural objects: rocks, boulders, logs, stumps, mushrooms.
• Decor: statues, busts, vases, candlesticks, banners on poles.
• Tools and simple weapons: hammers, axes, shovels, swords, shields.
• Food and small items: bread, fruit, potions, books, scrolls.
• Stylised anything: chunky, simplified shapes are easier than realism.

For these, a prompt or a quick sketch is often enough, and the result needs little more than a scale and pivot check.

Props that need more care

• Transparent or reflective parts: glass bottles, windows, chandeliers, mirrors. Generators make opaque meshes with textures, so glass usually needs a material change in the engine.
• Thin, repeated detail: chain-link fences, rope nets, wicker, wire. These thicken or blur. A texture with an alpha mask on a plane is usually the better game solution anyway.
• Foliage: trees, bushes and grass. Real-time foliage relies on cards and alpha textures that generators don't produce. See 3D tree models for games.
• Mechanical assemblies with moving parts: a door in a frame, a lever with a base, a cart with wheels. You'll get one mesh; split it in Blender if parts need to move.
• Modular kits: wall pieces that must snap to a grid. Generation won't hit exact dimensions; model these by hand or use a kit. See free modular assets for prototyping.
• Text and logos: signs with readable writing come out garbled. Add text as a decal or texture afterwards.

Budgeting a prop list

A rough way to plan an indie level's props:

1. List every prop from your blockout.
2. Tag each as "generate", "generate and edit" or "model or source".
3. Estimate cost. In Vi3W, a default sketch-to-3D model (image plus 3D) is about 30 XP; the pricing page lists the exact cost per model choice. Budget for a few retries on the trickier ones.
4. Estimate clean-up time for "generate and edit" props: splitting parts, fixing a texture, setting pivots.
5. Assign the "model or source" props to a modeller, a free pack, or the store.

Getting variety without chaos

Indie levels look repetitive when one barrel is copied 40 times. With generation you can afford variations: three barrels, two crates, four rocks. Keep them consistent by reusing style words and a reference image, and vary one thing at a time (size, damage, material). Consistent art style across AI assets has techniques.

Checks before a prop goes in the level

• Real-world scale.
• Pivot at the base (or the hinge, or the grip).
• Simple collision.
• Triangle count within budget for how often it appears. Poly count for game assets.
• Textures without baked shadows.
• Tested under the game's lighting.

Where to start

Start with the props that appear most often and are seen least closely. They give you the most time back and expose the fewest flaws. From greybox to dressed level walks through swapping placeholders for generated props without breaking the layout.

Have a list of props? Generate the first five in Vi3W and see how much of your list moves to "done".