You can deploy this model and start earning money today!

ZiggyBakerloo/
clip-image-embeddings-endpoint

Name: clip-image-embeddings-endpoint
Availability: InStock

endpoints_compatible

Views

No views yet

CLIP Image Embedding Inference Endpoint

A Hugging Face Inference Endpoint that generates 512-dimensional embeddings from images using OpenAI's CLIP model. Supports multiple image formats including WebP, JPEG, PNG, and others.

Features

✅ WebP support - Handles WebP images natively
✅ Multiple input formats - Flexible JSON input handling
✅ 512-dimensional embeddings - Standard CLIP image embeddings
✅ Normalized vectors - Ready for similarity calculations
✅ GPU acceleration - Automatic GPU detection and usage

Supported Image Formats

WebP
JPEG/JPG
PNG
GIF
BMP
TIFF

Input Formats

The endpoint accepts several input formats:

json

// Option 1: Direct image_url
{
  "image_url": "https://example.com/image.webp"
}

// Option 2: String inputs
{
  "inputs": "https://example.com/image.webp"
}

// Option 3: Nested object
{
  "inputs": {
    "image_url": "https://example.com/image.webp"
  }
}

Response Format

json

{
  "embeddings": [0.123, -0.456, ...], // 512-dimensional array
  "shape": [512],
  "model": "openai/clip-vit-base-patch32"
}

Deployment to Hugging Face

Create an Inference Endpoint on Hugging Face Hub
Upload these files to your endpoint repository:
- handler.py - Main inference handler
- requirements.txt - Python dependencies
Configure your endpoint with:
- Python 3.9+
- GPU instance (recommended for faster inference)

Local Testing

Run the example locally:

bash

pip install -r requirements.txt
python example_usage.py

Usage Example

python

import requests

# Your HF Inference Endpoint URL
endpoint_url = "https://your-endpoint.huggingface.co"

# Request with WebP image
response = requests.post(
    endpoint_url,
    headers={"Authorization": f"Bearer {your_token}"},
    json={"image_url": "https://example.com/image.webp"}
)

embeddings = response.json()["embeddings"]
print(f"Generated {len(embeddings)}-dimensional embedding")

Error Handling

The endpoint returns error details in case of failures:

json

{
  "error": "Failed to download image from URL: HTTP 404",
  "message": "Failed to process image and generate embeddings"
}

ZiggyBakerloo/clip-image-embeddings-endpoint

CLIP Image Embedding Inference Endpoint

Features

Supported Image Formats

Input Formats

Response Format

Deployment to Hugging Face

Local Testing

Usage Example

Error Handling

ZiggyBakerloo/
clip-image-embeddings-endpoint