La Plateforme

¶Pixtral: Vision Model Integration

Pixtral is Mistral's multimodal model capable of processing both text and images. Interacting with Pixtral is similar to a standard chat, but requires image inputs alongside text prompts.

You simply need to:

1. Use a vision-capable model like pixtral-12b-2409.
2. Construct a Message object that combines:
   • Text (MESSAGE_TYPE_TEXT)
   • Image URLs (MESSAGE_TYPE_IMAGE_URL)
   • Local images in base64 (MESSAGE_TYPE_BASE64)

Tip

This allows flexible input handling:

• Use URLs for hosted images.
• Use base64 for local files (e.g., charts, diagrams).
• Combine multiple images and texts for complex prompts.

¶Example: Describe Multiple Images

use Partitech\PhpMistral\Clients\Mistral\MistralClient;
use Partitech\PhpMistral\Message;

$apiKey = getenv('MISTRAL_API_KEY');
$client = new MistralClient($apiKey);

$message = $client->newMessage();
$message->setRole('user');

// Add initial text prompt
$message->addContent(
    type: Message::MESSAGE_TYPE_TEXT,
    content: 'Describe this image in detail please.'
);

// Add first image via URL
$message->addContent(
    type: Message::MESSAGE_TYPE_IMAGE_URL,
    content: 'https://s3.amazonaws.com/cms.ipressroom.com/338/files/201808/5b894ee1a138352221103195_A680%7Ejogging-edit/A680%7Ejogging-edit_hero.jpg'
);

// Add follow-up text
$message->addContent(
    type: Message::MESSAGE_TYPE_TEXT,
    content: 'And these two images as well. Answer in French.'
);

// Add second image via URL
$message->addContent(
    type: Message::MESSAGE_TYPE_IMAGE_URL,
    content: 'https://www.wolframcloud.com/obj/resourcesystem/images/a0e/a0ee3983-46c6-4c92-b85d-059044639928/6af8cfb971db031b.png'
);

// Add third image via local file (base64)
$message->addContent(
    type: Message::MESSAGE_TYPE_BASE64,
    content: realpath('./../../medias/pixtral_image_example_charts.jpeg')
);

$messages = $client->getMessages()->addMessage($message);

¶Send to the Pixtral Model

try {
    $result = $client->chat(
        $messages,
        [
            'model' => 'pixtral-12b-2409',
            'max_tokens' => 1024,
        ]
    );
} catch (\Partitech\PhpMistral\MistralClientException $e) {
    echo $e->getMessage();
    exit(1);
}

print($result->getMessage());

¶Sample Output (Translated in French)

### Description des Images

#### Première Image
Trois personnes courent dans un cadre naturel, souriantes et concentrées, entourées de verdure.

#### Deuxième Image
Une famille de cinq personnes, portant des chemises rouges, posant devant un fond blanc.

#### Troisième Image
Un graphique comparant les indicateurs socio-économiques entre les zones défavorisées et la France entière :
- **Classe ouvrière** : 33,5 % vs 14,5 %
- **Chômage** : 18,1 % vs 7,3 %
- **Jeunes sans emploi** : 25,2 % vs 12,9 %
- **Revenu médian** : 1 168 € vs 1 822 €
- **Pauvreté** : 43,3 % vs 15,5 %

Important

Pixtral excels at image captioning, chart analysis, and complex multimodal tasks. Make sure your image URLs are publicly accessible or use base64 for local files.