Z-Image-Turbo on RunPod Serverless

This repository deploys the Tongyi-MAI/Z-Image-Turbo model (6B parameters, S3-DiT architecture) on RunPod Serverless.

Features

• Fast Cold Starts: Model weights baked into Docker image
• API Key Security: Environment-variable based authentication
• Optimized for Speed: Uses bfloat16 and ZImagePipeline with 8-step inference
• PyTorch 2.5.1: Latest CUDA 12.4 support

Requirements

• GPU: RTX 4090 (24GB) or A100 recommended
• Container Disk: 15GB+
• VRAM: 14-16GB minimum

Deployment Instructions

1. Create RunPod Serverless Endpoint:

   • Use GitHub integration: https://github.com/Thetimii/z-images
   • Branch: master
   • Dockerfile path: Dockerfile

2. Configure Endpoint:

   • GPU: RTX 4090 (24GB)
   • Container Disk: 15GB+
   • Environment Variables:
     • API_KEY = your_secret_password

3. Wait for Build: First build takes ~10-15 minutes (downloads PyTorch 2.5.1 and model weights)

Testing

curl -X POST "https://api.runpod.ai/v2/YOUR_ENDPOINT_ID/runsync" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_RUNPOD_API_KEY" \
  -d '{
    "input": {
      "prompt": "A cinematic shot of a futuristic city with neon lights",
      "api_key": "your_secret_password",
      "num_inference_steps": 8,
      "guidance_scale": 0.0
    }
  }'

Z-Image-Turbo Parameters

• num_inference_steps: 8-10 (optimal, higher is worse)
• guidance_scale: Must be 0.0 (turbo model)
• torch_dtype: bfloat16 (for memory efficiency)

Technical Details

• Base Image: NVIDIA CUDA 12.4.1 with cuDNN
• PyTorch: 2.5.1 (required for ZImagePipeline)
• Diffusers: Installed from source (latest ZImagePipeline support)