Generating 3D models
Describe an object and get a textured 3D model placed in your scene. Choose a detail tier to match how you will use it.
Open in the interactive manual →The interactive version has full-text search, a table of contents, and light/dark themes.
With generation enabled, ask the Assistant for an object that isn't in your library ("generate a low-poly potted fern," "make a stylized treasure chest"), and it creates a textured 3D model and drops it into your scene. Cost is about $0.30 per model.
Detail tiers
You (or the Assistant) pick how dense the model should be, which matters a lot for scene performance:
| Tier | Budget | Use for |
|---|---|---|
| Low (default) | ~5k faces | Props you'll scatter many times, and the sensible default for everything else. |
| Mid | ~20k faces | General use when you want a little more detail. |
| High | ~50k faces + HD texture | A single hero centerpiece. |
Generations lean low-poly by default: if neither you nor the Assistant names a tier, you get Low. Leaner meshes load, render, and generate faster. When you do want more, just say so: a prompt that asks for "high detail," "photorealistic," "intricate," and the like automatically bumps the tier up to High.
When the confirm guard is on, the model's confirm card also shows a Low / Mid / High control (pre-set to the Assistant's pick) so you can nudge a hero piece up to High right before you approve the spend. Your choice there always wins.
Building from a reference image
The same confirm card carries + Reference image (optional). Attach a picture and the model is built from that image rather than from your words alone, the right move when you're recreating a specific object and describing it accurately would take a paragraph. On a material generation the image conditions the texture instead. The card shows a thumbnail of what you attached, with an ✕ to drop it again.
Image-conditioned generation runs on a different endpoint and costs 10 credits more; the quote on the button updates the moment you attach or remove the picture, so you always approve the real figure. The image is uploaded to your own Asset Library first and passed by id. The server checks you own it.
How it flows
Generation runs in the background: a model takes roughly 1–3 minutes. When it's ready, the Assistant places it in the scene and saves it to your Asset Library, so you can reuse it without paying again.