Laravel电商集成Deep Infra CLIP API遇'A contents key is required'错误求助

在Laravel电商应用中集成Deep Infra的openai/clip-vit-large-patch14-336图像识别API时，遇到错误提示"A 'contents' key is required"，原代码如下：

public function imageSearch(Request $request){
    try {
        $api = 'UMIDKJCkmZBRmCmmbzTWB8JPq3wMbd1b';
        // Instantiate a Guzzle HTTP client
        $client = new Client();

        $image = time().rand(1,1000).'.'.$request->image->extension();
        $request->image->move(public_path('search_images'),$image);

        $candidateLabels = ['cat', 'dog','phone','laptop'];
        $imagePath = public_path('search_images') . '/' . $image; // Assuming $image contains the filename

        $imageFile = fopen($imagePath, 'r');
        if (!$imageFile) {
            throw new Exception("Failed to open image file: $imagePath");
        }

        // Prepare the request body
        $body = [
            'multipart' => [
                [
                    'name' => 'image',
                    'contents' => fopen($imagePath, 'r'), // Read the image file
                    
                ],
                [
                    'name' => 'candidate_labels',
                    'contents' => json_encode($candidateLabels),
                ],
            ],
        ];

        // Prepare the request headers
        $headers = [
            'Authorization' => 'Bearer ' . $api,
        ];
        $response = $client->post('https://api.deepinfra.com/v1/inference/openai/clip-vit-base-patch32', [
            'headers' => $headers,
            'multipart' => $body,
        ]);

        // Handle the response
        $statusCode = $response->getStatusCode();
        $responseData = json_decode($response->getBody(), true);
        dd($responseData);
    } catch (\Throwable $th) {
        return $th->getMessage();
    }
}

问题原因 #

错误提示表明请求结构不符合Deep Infra CLIP API的要求：

• 该API不接受multipart/form-data格式的请求，需要用application/json格式
• 核心参数应为inputs，而非自定义的image和candidate_labels字段
• 图像需要以base64编码的形式传入，而非直接上传文件

修正后的代码 #

public function imageSearch(Request $request){
    try {
        $apiKey = 'UMIDKJCkmZBRmCmmbzTWB8JPq3wMbd1b';
        $client = new Client();

        // 读取上传的图像文件并转为base64编码
        $imageBase64 = base64_encode(file_get_contents($request->image->getPathname()));
        $imageData = "data:image/{$request->image->extension()};base64,$imageBase64";

        $candidateLabels = ['cat', 'dog','phone','laptop'];

        // 构造符合API要求的JSON请求体
        $requestBody = [
            'inputs' => [
                'image' => $imageData,
                'text' => $candidateLabels
            ]
        ];

        $headers = [
            'Authorization' => "Bearer $apiKey",
            'Content-Type' => 'application/json'
        ];

        // 注意修正API端点为目标模型
        $response = $client->post('https://api.deepinfra.com/v1/inference/openai/clip-vit-large-patch14-336', [
            'headers' => $headers,
            'json' => $requestBody
        ]);

        $responseData = json_decode($response->getBody(), true);
        dd($responseData);
    } catch (\Throwable $th) {
        return $th->getMessage();
    }
}

关键修改点 #

• 请求格式：从multipart改为json格式，添加Content-Type: application/json请求头
• 图像处理：直接读取上传文件的内容转为base64，无需保存到本地，减少IO操作
• 参数结构：使用API要求的inputs字段，包含image（base64格式）和text（候选标签数组）
• API端点：修正为目标模型openai/clip-vit-large-patch14-336，原代码使用的是基础版模型

内容的提问来源于stack exchange，提问作者nadeesha