You can deploy this model and start earning money today!

smorand/
hf-sdxl-endpoint

Name: hf-sdxl-endpoint
Availability: InStock

endpoints_compatible

Views

No views yet

Stable Diffusion XL - Hugging Face Inference Endpoint

Custom handler for deploying Stable Diffusion XL as a text-to-image API on Hugging Face Inference Endpoints.

Features

Text-to-image generation with Stable Diffusion XL
Configurable parameters (steps, guidance, dimensions, seed)
Optional refiner for higher quality images
Base64 encoded output (JPEG/PNG)
Memory-optimized for GPU inference

Project Structure

hf-inference-endpoint/
├── handler.py          # Custom handler implementation
├── requirements.txt    # Python dependencies
├── README.md          # This file
└── CLAUDE.md          # AI documentation

Deployment

Option 1: Deploy from Hugging Face Hub

Create a new model repository on Hugging Face Hub
Upload handler.py and requirements.txt to the repo
Go to Inference Endpoints
Create new endpoint:
- Select your model repository
- Choose GPU instance (recommended: A10G or A100)
- Set "Custom Handler" as the task
Deploy

Option 2: Deploy with Model Files

If you have custom model weights:

Upload model files to your HF repo (safetensors format recommended)
Place handler.py and requirements.txt in the repo root
Deploy as above

API Usage

Request Format

bash

curl -X POST https://your-endpoint.endpoints.huggingface.cloud \
  -H "Authorization: Bearer $HF_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": "A majestic lion wearing a crown, digital art, highly detailed",
    "parameters": {
      "negative_prompt": "blurry, low quality, distorted",
      "num_inference_steps": 50,
      "guidance_scale": 7.5,
      "width": 1024,
      "height": 1024,
      "seed": 42,
      "output_format": "jpeg"
    }
  }'

Response Format

json

{
  "image": "<base64-encoded-image>",
  "content_type": "image/jpeg",
  "parameters": {
    "prompt": "A majestic lion...",
    "negative_prompt": "blurry...",
    "num_inference_steps": 50,
    "guidance_scale": 7.5,
    "width": 1024,
    "height": 1024,
    "seed": 42
  }
}

Parameters

Parameter	Type	Default	Description
`inputs`	string	required	The text prompt for image generation
`negative_prompt`	string	""	What to avoid in the image
`num_inference_steps`	int	50	Number of denoising steps (20-100)
`guidance_scale`	float	7.5	How closely to follow the prompt (1-20)
`width`	int	1024	Image width (512-2048, multiple of 8)
`height`	int	1024	Image height (512-2048, multiple of 8)
`seed`	int	null	Random seed for reproducibility
`output_format`	string	"jpeg"	Output format: "jpeg" or "png"
`use_refiner`	bool	false	Use SDXL refiner for higher quality
`refiner_strength`	float	0.3	Refiner denoising strength

Decoding the Image

Python:

python

import base64
from PIL import Image
import io

response = requests.post(endpoint_url, json=payload, headers=headers)
result = response.json()

image_data = base64.b64decode(result["image"])
image = Image.open(io.BytesIO(image_data))
image.save("output.jpg")

JavaScript:

javascript

const response = await fetch(endpointUrl, { method: 'POST', headers, body: JSON.stringify(payload) });
const result = await response.json();

const imageBuffer = Buffer.from(result.image, 'base64');
fs.writeFileSync('output.jpg', imageBuffer);

Hardware Requirements

Minimum: A10G (24GB VRAM) - base model only
Recommended: A100 (40GB+ VRAM) - base + refiner
Memory: ~15GB VRAM for base model in fp16

Customization

Enable Refiner

Uncomment the refiner loading in handler.py: