How to Integrate Z Image Turbo API: A Production Guide to Low-Cost Image Generation ($0.0036/img)

This guide is intentionally objective and implementation-first. You'll learn:

• What ZImage Turbo is and its key capabilities
• How EvoLink's ZImage endpoint works (async tasks + webhook rules)
• How to integrate ZImage Turbo with cost control and operational guardrails
• When ZImage is the right default—and when it isn't

1. What Is ZImage Turbo

• Photorealistic image generation with high visual fidelity
• Bilingual text rendering (English & Chinese) directly in generated images
• Fast inference speed suitable for high-volume production
• Instruction adherence for consistent, predictable outputs

ZImage Turbo visual quality example A	ZImage Turbo visual quality example B

2. ZImage Pricing Context: Compare Against Published Tables

• EvoLink's model catalog lists ZImage Turbo "starting from" $0.0036 per image and shows a "market" reference of $0.0050.
• EvoLink also calls out ~$0.0036/image pricing in its changelog entry for ZImage Turbo.
• fal.ai's pricing page shows several widely used hosted image models around $0.03–$0.04 per image (model-specific).
• fal.ai also lists Z-Image Turbo pricing at $0.005/megapixel (useful as a market reference for ZImage itself).

Cost comparison table (transparent assumptions)

Assumption: "10k images" means 10,000 generated outputs at ~1MP each (e.g., 1024×1024 ≈ 1MP). Always confirm billing rules for your exact plan and output sizes.

3. How EvoLink's ZImage Turbo API Works (Async by Design)

1. You submit a ZImage generation request to POST /v1/images/generations
2. You receive a task ID immediately
3. You query the task status via GET /v1/tasks/{task_id} (or use a webhook callback)

4. ZImage Turbo Quickstart (Copy/Paste)

4.1 Create a ZImage generation task

curl --request POST \
  --url https://api.evolink.ai/v1/images/generations \
  --header 'Authorization: Bearer <YOUR_TOKEN>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "model": "z-image-turbo",
  "prompt": "A clean product hero shot, studio lighting, soft shadow, minimal background, no watermark",
  "size": "1:1"
}
'

{
  "id": "task_abc123",
  "status": "pending",
  "model": "z-image-turbo",
  "progress": 0,
  "created": 1735100000,
  "task_info": {
    "can_cancel": true,
    "estimated_time": 10
  },
  "usage": {
    "credits_reserved": 1
  }
}

4.2 Query your ZImage task status

curl --request GET \
  --url https://api.evolink.ai/v1/tasks/{task_id} \
  --header 'Authorization: Bearer <YOUR_TOKEN>'

5. ZImage Turbo Parameters

• prompt: Required, max 2000 characters.
• size: Supports ratios (1:1, 16:9, etc.) or custom dimensions (376x376 to 1536x1536).
• seed: Integer (1 - 2147483647) for reproducibility.
• callback_url: HTTPS URL for async notification.

6. ZImage Webhooks: Callback Rules

• HTTPS only: No internal IPs allowed.
• Retries: Max 3 retries (1s, 2s, 4s).
• Format: Callback body matches the Task Query API response.

7. Production Patterns for ZImage

• Pattern A: Variant Selection: Generate 4–8 candidates cheaply and pick the winner.
• Pattern B: Early Persistence: Store outputs in your S3/OSS immediately (URLs expire in 24h).
• Pattern C: Worker Queue: Use a decoupled worker to handle status polling or webhooks.

8. ZImage Prompt Templates

Template: Bilingual Poster Draft

Prompt: A clean retail poster layout with clear bilingual typography. Big title: "Winter Sale" and "冬季大促". Subtitle: "Up to 50% off" and "最高立减50%". Modern grid layout, high contrast, sans-serif.

9. Decision Matrix

10. FAQ

• Architecture: 6B model, S3-DiT architecture, 8 NFEs for Turbo.
• Retention: Links valid for 24 hours.
• Pricing: Starts from $0.0036 on EvoLink.ai.

Conclusion