Create Chat Vision API (Streaming / Non-Streaming)

This API supports multimodal inputs, allowing users to send text and image URLs within a conversation. The model will generate appropriate responses based on the provided prompts and image content. It supports streaming output to provide a smoother, more interactive experience.

• Endpoint: https://api.codingplanx.ai/v1/chat/completions
• Method: POST
• Official Reference: OpenAI Chat Completion API

---

Request Parameters

Header Parameters

Body Parameters

messages Object Structure

Each item in the message list contains the following fields:

• role: (string) The role of the message's author. Options: system, user, assistant, tool.
• content: (string or array) The contents of the message.
  • In Vision Mode, content is an array of objects containing types text and image_url.

---

Request Example

Mixed Text and Vision Request (JSON)

{
    "model": "gpt-4o",
    "messages": [
      {
        "role": "system",
        "content": "You are a professional image analysis assistant."
      },
      {
            "role": "user",
            "content": [
                {
                    "type": "text", 
                    "text": "What is in this image? Please describe it in detail."
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example.com/image.png"
                    }
                }
            ]
        }
    ],
    "stream": true
}

---

Response Explanation

Non-Streaming Response (stream: false)

Response Example

{
    "id": "chatcmpl-123",
    "object": "chat.completion",
    "created": 1677652288,
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "This image shows a tranquil lake with a backdrop of rolling mountains."
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 120,
        "completion_tokens": 35,
        "total_tokens": 155
    }
}

Streaming Response (stream: true)

When stream is set to true, the API returns a text/event-stream. Each line begins with data: followed by a JSON string. The stream is terminated by a data: [DONE] message.

---

FAQs (Frequently Asked Questions)

Q1: How can I upload local images instead of using URLs? A: This API primarily supports image URLs. If you only have local images, it is recommended to upload them to your image hosting service or cloud storage (OSS) first. Alternatively, you can convert the image into a Base64 encoded string and format it as data:image/jpeg;base64,{base64_encode_data}, then pass it into the url field.

Q2: Why are there no usage statistics in my streaming response? A: In standard OpenAI-compatible streaming protocols, usage is typically only returned in the final data chunk or by setting a specific stream_options parameter. Please check if your model version supports returning token counts within a stream.

Q3: What are the size and format limits for image recognition? A: Generally, PNG, JPEG, WEBP, and GIF (non-animated) formats are supported. It is recommended that image sizes do not exceed 20MB. For optimal recognition results, a resolution of 512x512 or higher is advised.

Q4: What should I do if my request returns a "401 Unauthorized" error? A: Please ensure that you are correctly passing the Authorization field in the Header. The format should be Bearer followed by your API KEY. Also, verify that the KEY is valid and your account balance is sufficient.

Q5: Does the temperature parameter affect image recognition? A: Yes, it does. The temperature affects the linguistic creativity of the model when describing an image. If you need highly objective and rigorous image descriptions, a lower value (e.g., 0.2) is recommended. If you want more vivid and engaging descriptions, you can set a higher value (e.g., 0.8).

Q6: How do I process multiple images in a single request? A: You can place multiple objects with the image_url type into the messages.content array. The model will attempt to understand the context and content of all provided images simultaneously.