# Grounding DINO - Core.Today AI API

> Text-prompted object detection (39M+ runs) — describe what to find in natural language ('red car, person wearing a hat') and get bounding boxes with confidence scores plus an annotated image.

- **Provider**: Grounding DINO
- **Model ID**: adirik/grounding-dino
- **Category**: Image Generation
- **Credits**: 2 per image
- **Speed**: Fast
- **Quality**: High

## Features
- Zero-shot detection from natural language
- Bounding boxes + confidence scores (JSON)
- Annotated visualization image
- Adjustable box/text thresholds

## Use Cases
- Product detection in commerce photos
- Shelf/inventory auditing
- Dataset auto-labeling
- Region targeting for downstream edits
- Content compliance checks

## 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

### Optional
- **image**: string - Image to run detection on
- **query**: string - Comma-separated things to detect, e.g. 'red car, person wearing a hat'
- **box_threshold**: number (default: 0.25) - Confidence threshold for boxes (0-1)
- **text_threshold**: number (default: 0.25) - Confidence threshold for text matching (0-1)
- **show_visualisation**: boolean (default: true) - Also return an annotated result image

## Examples

### Detect Products
Find specific items in a shelf photo
```json
{
  "model": "adirik/grounding-dino",
  "input": {
    "image": "https://example.com/shelf.jpg",
    "query": "soda can, water bottle, snack bag",
    "box_threshold": 0.3
  }
}
```

## 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
- Phrase queries as short noun phrases separated by commas
- Raise thresholds (0.3-0.4) to cut false positives; lower to catch more
- detections JSON gives you boxes for cropping or downstream editing
- For detection + caption + OCR in one call, see Florence-2 Large

## Documentation
https://replicate.com/adirik/grounding-dino