Complete Guide to Running z-image-turbo on RunPod Serverless | Building an API Environment

For advanced users who want to use z-image-turbo for API-based bulk generation and automation, this guide explains how to set up a RunPod Serverless environment.

If you’re a beginner, we recommend starting with ConoHa AI Canvas first.

What Is RunPod Serverless?

RunPod is a cloud GPU platform. Using its Serverless function, you can run AI inference as an API endpoint.

Advantages:

• Pay-as-you-go billing (zero cost when idle)
• Auto-scaling support
• Callable via API for automation and batch processing
• Parallel execution with multiple workers possible

Disadvantages:

• Technical knowledge required (Docker, API)
• Latency of a few seconds to tens of seconds on cold start

For cost comparisons with other services, see the Cloud GPU Comparison article.

Setup Steps

Step 1: Create a RunPod Account

1. Visit RunPod and create an account
2. Register a credit card (for pay-as-you-go billing)
3. Add initial credits ($10+ recommended)

Step 2: Create a Serverless Endpoint

1. In the RunPod dashboard, navigate to the Serverless section
2. Click New Endpoint
3. Enter the following settings:

Step 3: Get an API Key

1. In the RunPod dashboard, go to Settings → API Keys
2. Generate a new API key
3. Also note down the Endpoint ID

Step 4: Set Environment Variables

# Save to ~/.config/z-image/.env
RUNPOD_API_KEY=your_api_key_here
RUNPOD_ENDPOINT_ID=your_endpoint_id_here

Step 5: Test Request

Use curl to test the API:

curl -X POST "https://api.runpod.ai/v2/${RUNPOD_ENDPOINT_ID}/runsync" \
  -H "Authorization: Bearer ${RUNPOD_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "prompt": "a beautiful Japanese woman, portrait, natural lighting, 85mm lens",
      "width": 1024,
      "height": 1024,
      "steps": 8,
      "sampler": "euler",
      "scheduler": "ddim_uniform"
    }
  }'

Cost Optimization

GPU Selection

z-image-turbo can generate in 8 steps, so expensive GPUs aren’t necessary. RTX A4000 is the best cost-performance option.

Worker Configuration Tips

• Min Workers = 0: Zero cost when idle. However, cold starts will occur.
• Min Workers = 1: Keep one worker always running. For cases where latency is unacceptable.
• Max Workers: Upper limit for concurrent requests. Set according to your batch processing parallelism.

Batch Processing

Using RunPod API’s asynchronous endpoint (/run) lets you send multiple requests in parallel for batch processing.

Prompt Configuration

For prompt writing, see these articles:

• Prompt Basics — Word order and emphasis syntax fundamentals
• Prompt Best Practices — Verified prompt knowledge
• Prompt Examples — Actual examples with prompts

The same parameter settings used in the ComfyUI workflow can also be used with the API.

Troubleshooting

Cold Start Is Slow

• Set Min Workers to 1 to keep a worker always running (increased cost)
• Enable the FlashBoot option (if supported for the GPU)

Images Not Generating

• Check that the API key is correct
• Check that the Endpoint ID is correct
• Check the endpoint status in the RunPod dashboard

Costs Higher Than Expected

• Shorten the Idle Timeout (5 seconds recommended)
• Verify Min Workers is set to 0
• Delete unused endpoints

Summary

z-image-turbo environment setup on RunPod Serverless:

1. Create account → Configure endpoint → Get API key
2. RTX A4000 is the best cost-performance option
3. Min Workers = 0 for zero idle cost
4. curl or script for API requests

Related Articles

• What Is z-image-turbo — Model features and selection guide
• ConoHa AI Canvas Getting Started — Beginners start here
• ComfyUI Workflow — Pre-configured workflow
• Cloud GPU Comparison — Cost comparison between services