STL Export Pipeline¶

The STL pipeline converts GLB models from AI providers into print-ready STL files with automated quality assurance.

Architecture¶

GLB Model (Tripo/Meshy)
    |
    v
Download & Convert (trimesh)
    |
    v
Auto-Repair (admesh)
    |
    v
Quality Assessor (GPT-4o) → REJECT / FAST / FULL
    |
    v
Scale & Validate
    |
    v
Remix Decision (GPT-4o) → PASS / REMIX / ESCALATE
    |
    v
Upload STL to MinIO

The pipeline runs as a FastAPI microservice (mdu-stl-pipeline) on port 8090, proxied through mdu-api.

Endpoints¶

Direct (STL Pipeline)¶

Method	Path	Auth	Description
GET	`/api/stl/health`	-	Health check
GET	`/api/stl/profiles`	-	List printer profiles
POST	`/api/stl/export/stl`	-	Start STL export job
GET	`/api/stl/export/status/:jobId`	-	Poll job status

Proxied via mdu-api¶

Method	Path	Auth	Description
GET	`/api/stl-profiles`	-	List printer profiles
POST	`/api/stl-export`	Required	Start STL export job
GET	`/api/stl-status/:jobId`	Required	Poll job status

Printer Profiles¶

5 built-in profiles optimized for different printer types:

Profile	Type	Layer Height	Nozzle
`fdm_standard`	FDM	0.2mm	0.4mm
`fdm_fine`	FDM	0.1mm	0.4mm
`resin_standard`	Resin	0.05mm	-
`resin_fine`	Resin	0.025mm	-
`resin_ultra`	Resin	0.01mm	-

Export STL¶

POST /api/stl-export
Authorization: Bearer <token>
Content-Type: application/json

{
  "source": "tripo",
  "tripo_task_id": "abc123",
  "printer_profile": "fdm_standard",
  "target_scale": "32mm_standard",
  "auto_repair": true,
  "original_prompt": "dwarf warrior"
}

Response (immediate):

{
  "job_id": "abc12345",
  "status": "processing",
  "message": "STL export started"
}

Source Options¶

Source	Required Field	Description
`tripo`	`tripo_task_id`	Tripo3D task
`meshy`	`meshy_task_id` + `meshy_task_type`	Meshy task

meshy_task_type: text-to-3d or image-to-3d

Poll Status¶

GET /api/stl-status/abc12345
Authorization: Bearer <token>

Response (success):

{
  "job_id": "abc12345",
  "status": "success",
  "progress": 100,
  "stage": "done",
  "stl_url": "https://minidreamuniverse.com/api/assets/mdu-assets/...",
  "metadata": {
    "face_count": 42800,
    "is_watertight": true,
    "volume_mm3": 12450.5,
    "bounding_box_mm": [24.5, 31.2, 38.7],
    "quality_status": "PASS"
  },
  "qa_path": "FAST",
  "qa_score": 92,
  "remix_decision": { "action": "PASS", "confidence": 0.95 }
}

Stages: downloading → converting → repairing → validating → scaling → uploading → done

AI Quality Assurance¶

Flow 2 — Quality Assessor¶

After mesh repair, GPT-4o evaluates the mesh and routes it:

Decision	Action
FAST	Skip detailed validation, proceed directly
FULL	Run complete validation suite
REJECT	Fail the job immediately

Flow 3 — Remix Decision¶

After validation, GPT-4o evaluates the final result:

Decision	Action
PASS	Model is print-ready
REMIX	Badge stored in metadata (suggest re-generation)
ESCALATE	Flag for manual review

Base Generation¶

POST /api/generate-base
Authorization: Bearer <token>

{
  "miniature_bbox": [24.5, 31.2, 38.7],
  "style": "gaming",
  "base_type": "round",
  "base_size": "32mm",
  "magnet_slots": true,
  "printer_profile": "fdm_standard"
}

Uses Flow 4 (Base Customizer) to generate build123d Python code, executes it in the mdu-base-builder container, and returns a MinIO-hosted STL URL.

Usage Limits¶

STL export counts are enforced per subscription plan:

Plan	STL Exports/month
Free	10
Starter	30
Pro	50

Performance¶

Tripo3D pipeline: ~12 seconds end-to-end
Meshy pipeline: ~154 seconds (uses Remesh), GLB fallback available
Validation timeout: 240 seconds