# GFPGAN - Core.Today AI API

> Tencent ARC's face restoration model — 115M+ runs, the most-run model in the restoration collection. Restores old/blurry photos and fixes faces in AI-generated images.

- **Provider**: Tencent ARC
- **Model ID**: tencentarc/gfpgan
- **Category**: Image Generation
- **Credits**: 6 per image
- **Speed**: Fast
- **Quality**: High

## Features
- 115M+ runs — the standard for face restoration
- Restores old, blurry, or damaged photos
- Fixes distorted faces in AI-generated images
- Four versions incl. RestoreFormer
- Adjustable rescaling factor

## Use Cases
- Old family photo restoration
- Fixing AI-generated faces
- Profile photo enhancement
- Archive digitization cleanup
- Low-quality selfie improvement

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Required
- **img**: string - Input image with faces to restore

### Optional
- **version**: string (default: v1.4) - v1.3: better quality. v1.4: more details and better identity. RestoreFormer: alternative architecture
  Options: v1.2, v1.3, v1.4, RestoreFormer
- **scale**: number (default: 2) - Rescaling factor for the output

## Examples

### Old Photo Restoration
Restore a blurry old family photo at 2x
```json
{
  "model": "tencentarc/gfpgan",
  "input": {
    "img": "https://example.com/old-photo.jpg",
    "version": "v1.4",
    "scale": 2
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- v1.4 keeps identity best; try v1.3 if v1.4 looks over-sharpened
- Works on all detected faces in the image at once
- Pair with Real-ESRGAN (face_enhance) when you also need background upscaling
- For adjustable quality/fidelity balance, compare with CodeFormer

## Documentation
https://replicate.com/tencentarc/gfpgan