README

A simple PHP client for interacting with the Replicate API, enabling the ability to run machine learning models from your PHP applications. This library provides a clean and reusable way to interact with the API in a Laravel environment but can also be used in other PHP applications.

Features

• Simple API client for Replicate
• Supports Laravel out of the box
• Uses Guzzle HTTP client for HTTP requests
• Allows easy integration with machine learning models using Replicate

Installation

To install the package, simply use Composer:

composer require arikislam/replicate-php-client

Laravel Configuration

For Laravel applications, the package will automatically register the ReplicatePhpClientServiceProvider and the Replicate alias.

You can publish the config file with:

php artisan vendor:publish --provider="Arikislam\ReplicatePhpClient\Providers\ReplicatePhpClientServiceProvider"

Usage

Configuration

To use this package, you must have an API key from Replicate.

Set up the API key in your environment file:

REPLICATE_API_KEY=your_api_key_here

Example Request

To create an inference using the Replicate API, use the Replicate facade in your application as shown below:

use Arikislam\ReplicatePhpClient\Replicate;

$response = Replicate::createPrediction([
    'input' => [
        'prompt' => 'Test',
        'go_fast' => true,
        'num_outputs' => 1,
        'aspect_ratio' => '16:9',
        'output_format' => 'png',
        'output_quality' => 80,
        'megapixels' => '.25',
    ],
]);

// Get the prediction results
$predictionResult = $response->getResult();

This example demonstrates how to send a request to the Replicate API with the required parameters.

Detailed Breakdown

1. Input

The input parameter specifies the model input options. In this case:

• prompt: The text prompt for generating the image.
• go_fast: A boolean indicating whether the model should prioritize speed over quality.
• num_outputs: The number of outputs required from the model.
• aspect_ratio: The aspect ratio of the output (e.g., 16:9).
• output_format: The format of the generated output (e.g., png).
• output_quality: The quality of the output (percentage-based, 80 means 80% quality).
• megapixels: Controls the megapixel size for the image (e.g., .25).

2. Webhooks

Although webhooks can be used to track the progress of asynchronous tasks, this package allows you to specify a webhook URL and webhook_events_filter to listen for certain events. Currently, the webhook parameter is set to null in the example for simplicity.

3. API Version

The version is an identifier for the version of the Replicate API being used. You can specify which model version you are targeting by using the correct version hash.

Response Handling

The response from the API will contain details of the prediction, including the generated content or the status of the request. You can handle the response as follows:

if ($response->isSuccessful()) {
    $output = $response->getResult();
    // Do something with the output, such as saving the image or displaying it
} else {
    // Handle errors
    $errorMessage = $response->getErrorMessage();
    // Log or display the error message
}

Testing

This package includes support for unit testing. You can run tests using PHPUnit:

vendor/bin/phpunit

Advanced Usage

Custom HTTP Client

You can provide your own custom HTTP client if needed by extending the Guzzle client. For example:

use GuzzleHttp\Client;
use Arikislam\ReplicatePhpClient\Replicate;

$client = new Client([
    'base_uri' => 'https://api.replicate.com',
    'timeout'  => 10.0,
]);

$replicate = new Replicate($client);
$response = $replicate->createPrediction([
    // input options
]);

Facades in Laravel

For Laravel users, the Replicate facade provides a convenient way to interact with the API. However, if you are not using Laravel, you can directly instantiate the Replicate class and make API requests.

Contributing

Contributions are welcome! Please submit a pull request or open an issue to contribute to the development of this package.

License

This package is licensed under the MIT License. See the LICENSE file for more information.

Author

This package is maintained by S M Ariq Islam.