fal-3d

Generate 3D models (GLB/OBJ/PLY) from text descriptions or images using fal.ai.

Scripts

| Script | Purpose | |--------|---------| | generate-3d.sh | Generate a 3D model from text or image |

Usage

Image to 3D

./scripts/generate-3d.sh --image-url "https://example.com/object.jpg" --model fal-ai/hunyuan3d-v3/image-to-3d

Text to 3D

./scripts/generate-3d.sh --prompt "A medieval sword with ornate handle" --model fal-ai/meshy/v6/text-to-3d

Arguments

| Argument | Description | Required | |----------|-------------|----------| | --image-url | URL of image to convert to 3D | Yes (or --prompt) | | --prompt / -p | Text description for text-to-3D | Yes (or --image-url) | | --model / -m | Model endpoint | No (default: fal-ai/hunyuan3d-v3/image-to-3d) | | --param | Extra param as key=value (repeatable) | No |

Finding Models

To discover the best and latest 3D generation models, use the search API:

# Search for image-to-3D models
bash /mnt/skills/user/fal-generate/scripts/search-models.sh --category "image-to-3d"

# Search for text-to-3D models
bash /mnt/skills/user/fal-generate/scripts/search-models.sh --query "text to 3d"

Or use the search_models MCP tool with relevant keywords like "3d", "mesh", "image-to-3d".

Tips

• For best results from images: use a clear photo with a single object on a plain background
• Remove background first if needed (use fal-image-edit or fal-generate with birefnet)
• Simple, well-defined objects work best — complex scenes don't reconstruct well yet
• 3D generation takes 1-5 minutes — jobs use the queue API

Output Format

{
  "mesh": {
    "url": "https://fal.media/files/.../model.glb",
    "content_type": "model/gltf-binary",
    "file_name": "model.glb"
  }
}