Search by image - Alibaba Cloud Model Studio: Central de Documentação do Alibaba Cloud

A ferramenta de busca por imagem permite que o modelo pesquise na internet imagens visualmente semelhantes com base em uma imagem de entrada. O modelo analisa os resultados da pesquisa e faz inferências. Esse recurso é útil para cenários como encontrar produtos similares ou rastrear a origem de conteúdo visual.

Como usar

Chame o recurso de busca por imagem pela Responses API. Adicione a ferramenta image_search ao parâmetro tools e passe a imagem no parâmetro input em formato multimodal.

O parâmetro input deve conter conteúdo de imagem. Passe a URL da imagem usando o tipo input_image . Você também pode passar texto usando o tipo input_text para fornecer informações adicionais à pesquisa.

# Import dependencies and create a client...
input_content = [
    {"type": "input_text", "text": "Find landscape images with a style similar to this one"},
    {"type": "input_image", "image_url": "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png"}
]
response = client.responses.create(
    model="qwen3.8-max",
    input=[{"role": "user", "content": input_content}],
    tools=[{"type": "image_search"}]
)

print(response.output_text)

Modelos suportados

Modelos recomendados

Para obter o melhor desempenho na chamada de ferramentas, use os seguintes modelos:

Qwen-Plus: série Qwen3.7-Plus, série Qwen3.6-Plus, série Qwen3.5-Plus

Qwen-Max: série Qwen3.8-Max, qwen3.7-max-2026-06-08

qwen3.8-27b

Outros modelos

Os modelos abaixo também oferecem suporte a esta ferramenta, mas apresentam desempenho inferior aos modelos recomendados.

Qwen-Flash: série Qwen3.7-Flash, série Qwen3.6-Flash, série Qwen3.5-Flash

Este recurso só pode ser chamado pela Responses API.

Primeiros passos

Execute o código a seguir para chamar a ferramenta de busca por imagem pela Responses API e pesquisar imagens semelhantes ou relacionadas com base na imagem de entrada.

Antes de começar, obtenha uma chave de API e configure a chave de API como uma variável de ambiente .

Substitua o valor de image_url no código de exemplo por uma URL de imagem publicamente acessível.

python

import os
import json
from openai import OpenAI

client = OpenAI(
    # If you have not configured the environment variable, replace the following line with your Model Studio API key: api_key="sk-xxx",
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    # The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
    base_url="https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"
)

input_content = [
    {"type": "input_text", "text": "Find landscape images with a style similar to this one"},
    # Replace image_url with the actual public URL of the image
    {"type": "input_image", "image_url": "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png"}
]

response = client.responses.create(
    model="qwen3.8-max",
    input=[{"role": "user", "content": input_content}],
    tools=[
        {
            "type": "image_search"
        }
    ]
)

# Traverse the output to display each step
for item in response.output:
    if item.type == "image_search_call":
        print(f"[Tool Call] Search by image (status: {item.status})")
        # Parse and display the list of searched images
        if item.output:
            images = json.loads(item.output)
            print(f"  Found {len(images)} images:")
            for img in images[:5]:  # Display the first 5 images
                print(f"  [{img['index']}] {img['title']}")
                print(f"      {img['url']}")
            if len(images) > 5:
                print(f"  ... {len(images)} images in total")
    elif item.type == "message":
        print(f"\n[Model Response]")
        print(response.output_text)

# Display token usage and tool call statistics
print(f"\n[Token Usage] Input: {response.usage.input_tokens}, Output: {response.usage.output_tokens}, Total: {response.usage.total_tokens}")
if hasattr(response.usage, 'x_tools') and response.usage.x_tools:
    for tool_name, info in response.usage.x_tools.items():
        print(f"[Tool Statistics] {tool_name} call count: {info.get('count', 0)}")

javascript

import OpenAI from "openai";
import process from 'process';

const openai = new OpenAI({
    // If you have not configured the environment variable, replace the following line with your Model Studio API key: apiKey: "sk-xxx",
    apiKey: process.env.DASHSCOPE_API_KEY,
    // The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
    baseURL: "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"
});

async function main() {
    const response = await openai.responses.create({
        model: "qwen3.8-max",
        input: [
            {
                role: "user",
                content: [
                    { type: "input_text", text: "Find landscape images with a style similar to this one" },
                    // Replace image_url with the actual public URL of the image
                    { type: "input_image", image_url: "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png" }
                ]
            }
        ],
        tools: [
            { type: "image_search" }
        ]
    });

    // Traverse the output to display each step
    for (const item of response.output) {
        if (item.type === "image_search_call") {
            console.log(`[Tool Call] Search by image (status: ${item.status})`);
            // Parse and display the list of searched images
            if (item.output) {
                const images = JSON.parse(item.output);
                console.log(`  Found ${images.length} images:`);
                images.slice(0, 5).forEach(img => {
                    console.log(`  [${img.index}] ${img.title}`);
                    console.log(`      ${img.url}`);
                });
                if (images.length > 5) {
                    console.log(`  ... ${images.length} images in total`);
                }
            }
        } else if (item.type === "message") {
            console.log(`\n[Model Response]`);
            console.log(response.output_text);
        }
    }

    // Display token usage and tool call statistics
    console.log(`\n[Token Usage] Input: ${response.usage.input_tokens}, Output: ${response.usage.output_tokens}, Total: ${response.usage.total_tokens}`);
    if (response.usage && response.usage.x_tools) {
        for (const [toolName, info] of Object.entries(response.usage.x_tools)) {
            console.log(`[Tool Statistics] ${toolName} call count: ${info.count || 0}`);
        }
    }
}

main();

bash

# The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
curl -X POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/responses \
-H "Authorization: Bearer $DASHSCOPE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
    "model": "qwen3.8-max",
    "input": [
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "Find landscape images with a style similar to this one"},
                {"type": "input_image", "image_url": "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png"}
            ]
        }
    ],
    "tools": [
        {"type": "image_search"}
    ]
}'

Após executar o código, você receberá uma resposta semelhante à seguinte:

[Tool Call] Search by image (status: completed)
  Found 2 images:
  [1] QingMing Festival Holiday Notice 2024
      https://www.healthcabin.net/blog/wp-content/uploads/2024/04/QingMing-Festival-Holiday-Notice-2024.jpg
  [2] Serene Asian Landscape Stone Bridge Reflecting in Misty Water
      https://thumbs.dreamstime.com/b/serene-asian-landscape-stone-bridge-reflecting-misty-water-tranquil-illustration-traditional-arch-spanning-lake-style-376972039.jpg

[Model Response]
OK. I have found several landscape images with a similar style.

These images all display the artistic conception of typical Chinese ink wash paintings or traditional landscape paintings, and they share the following common points:
*   **Traditional architecture**: such as pavilions, towers, and arch bridges.
*   **Natural elements**: such as distant mountains, lakes, weeping willows, and lotus flowers.
*   **Artistic style**: uses elegant colors and soft lines to create a quiet and serene atmosphere.

...

[Token Usage] Input: 2753, Output: 181, Total: 2934
[Tool Statistics] image_search call count: 1

Saída em streaming

A ferramenta de busca por imagem tem um tempo de processamento longo. Ative a saída em streaming para recuperar resultados intermediários em tempo real.

python

import os
import json
from openai import OpenAI

client = OpenAI(
    # If you have not configured the environment variable, replace the following line with your Model Studio API key: api_key="sk-xxx",
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    # The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
    base_url="https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"
)

input_content = [
    {"type": "input_text", "text": "Find landscape images with a style similar to this one"},
    # Replace image_url with the actual public URL of the image
    {"type": "input_image", "image_url": "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png"}
]

stream = client.responses.create(
    model="qwen3.8-max",
    input=[{"role": "user", "content": input_content}],
    tools=[{"type": "image_search"}],
    stream=True
)

for event in stream:
    # Tool call starts
    if event.type == "response.output_item.added":
        if event.item.type == "image_search_call":
            print("[Tool Call] Searching by image...")
    # Tool call is complete. Parse and display the list of searched images.
    elif event.type == "response.output_item.done":
        if event.item.type == "image_search_call":
            print(f"[Tool Call] Search by image complete (status: {event.item.status})")
            if event.item.output:
                images = json.loads(event.item.output)
                print(f"  Found {len(images)} images:")
                for img in images[:5]:  # Display the first 5 images
                    print(f"  [{img['index']}] {img['title']}")
                    print(f"      {img['url']}")
                if len(images) > 5:
                    print(f"  ... {len(images)} images in total")
    # Model response starts
    elif event.type == "response.content_part.added":
        print(f"\n[Model Response]")
    # Streamed text output
    elif event.type == "response.output_text.delta":
        print(event.delta, end="", flush=True)
    # Response is complete. Output the usage.
    elif event.type == "response.completed":
        usage = event.response.usage
        print(f"\n\n[Token Usage] Input: {usage.input_tokens}, Output: {usage.output_tokens}, Total: {usage.total_tokens}")
        if hasattr(usage, 'x_tools') and usage.x_tools:
            for tool_name, info in usage.x_tools.items():
                print(f"[Tool Statistics] {tool_name} call count: {info.get('count', 0)}")

javascript

import OpenAI from "openai";
import process from 'process';

const openai = new OpenAI({
    // If you have not configured the environment variable, replace the following line with your Model Studio API key: apiKey: "sk-xxx",
    apiKey: process.env.DASHSCOPE_API_KEY,
    // The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
    baseURL: "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"
});

async function main() {
    const stream = await openai.responses.create({
        model: "qwen3.8-max",
        input: [
            {
                role: "user",
                content: [
                    { type: "input_text", text: "Find landscape images with a style similar to this one" },
                    // Replace image_url with the actual public URL of the image
                    { type: "input_image", image_url: "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png" }
                ]
            }
        ],
        tools: [{ type: "image_search" }],
        stream: true
    });

    for await (const event of stream) {
        // Tool call starts
        if (event.type === "response.output_item.added") {
            if (event.item.type === "image_search_call") {
                console.log("[Tool Call] Searching by image...");
            }
        }
        // Tool call is complete. Parse and display the list of searched images.
        else if (event.type === "response.output_item.done") {
            if (event.item && event.item.type === "image_search_call") {
                console.log(`[Tool Call] Search by image complete (status: ${event.item.status})`);
                if (event.item.output) {
                    const images = JSON.parse(event.item.output);
                    console.log(`  Found ${images.length} images:`);
                    images.slice(0, 5).forEach(img => {
                        console.log(`  [${img.index}] ${img.title}`);
                        console.log(`      ${img.url}`);
                    });
                    if (images.length > 5) {
                        console.log(`  ... ${images.length} images in total`);
                    }
                }
            }
        }
        // Model response starts
        else if (event.type === "response.content_part.added") {
            console.log(`\n[Model Response]`);
        }
        // Streamed text output
        else if (event.type === "response.output_text.delta") {
            process.stdout.write(event.delta);
        }
        // Response is complete. Output the usage.
        else if (event.type === "response.completed") {
            const usage = event.response.usage;
            console.log(`\n\n[Token Usage] Input: ${usage.input_tokens}, Output: ${usage.output_tokens}, Total: ${usage.total_tokens}`);
            if (usage && usage.x_tools) {
                for (const [toolName, info] of Object.entries(usage.x_tools)) {
                    console.log(`[Tool Statistics] ${toolName} call count: ${info.count || 0}`);
                }
            }
        }
    }
}

main();

bash

# The following URL is for the Singapore region. Replace {WorkspaceId} with your actual workspace ID. URLs vary by region.
curl -X POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/responses \
-H "Authorization: Bearer $DASHSCOPE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
    "model": "qwen3.8-max",
    "input": [
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "Find landscape images with a style similar to this one"},
                {"type": "input_image", "image_url": "https://img.alicdn.com/imgextra/i4/O1CN01YbrnSS1qtmsAkw0Ud_!!6000000005554-2-tps-788-450.png"}
            ]
        }
    ],
    "tools": [
        {"type": "image_search"}
    ],
    "stream": true
}'

Após executar o código, você receberá uma resposta semelhante à seguinte:

[Tool Call] Searching by image...
[Tool Call] Search by image complete (status: completed)
  Found 3 images:
  [1] QingMing Festival Holiday Notice 2024
      https://www.healthcabin.net/blog/wp-content/uploads/2024/04/QingMing-Festival-Holiday-Notice-2024.jpg
  [2] Serene Asian Landscape Stone Bridge Reflecting in Misty Water
      https://thumbs.dreamstime.com/b/serene-asian-landscape-stone-bridge-reflecting-misty-water-...
  [3] ...

[Model Response]
OK. I have found several landscape images with a similar style. These images all display the style of typical Chinese ink wash or fine-brush paintings...

[Token Usage] Input: 5339, Output: 164, Total: 5503
[Tool Statistics] image_search call count: 1

Faturamento

O faturamento envolve os seguintes aspectos:

Taxas de chamada do modelo: Os resultados da busca por imagem são anexados ao prompt, o que aumenta o número de tokens de entrada do modelo. A cobrança segue a tarifa padrão do modelo. Para detalhes de preços, consulte o console do Model Studio.
Taxas de chamada de ferramenta: A cobrança é feita a cada 1.000 chamadas. A taxa é de $8,00 para implantações na região de Singapore, enquanto a taxa para implantações em North China 2 (Beijing) é de $6.881.

Perguntas frequentes

P: Quais formatos de imagem e métodos de entrada são suportados?

R: Para mais informações, consulte Image limits e File input methods.

O SDK da OpenAI não oferece suporte à passagem de caminhos de arquivos locais.

P: Quantas imagens podem ser enviadas?

R: A quantidade de imagens permitida depende do comprimento máximo de entrada do modelo. O total de tokens das imagens e do texto não deve exceder o valor máximo suportado pelo modelo. O modelo pesquisa apenas uma imagem por chamada, mas você pode fazer múltiplas chamadas para processar várias imagens.

O modelo decide quantas imagens pesquisar.

P: Quantas imagens pesquisadas são retornadas?

R: O modelo determina a quantidade de imagens a serem retornadas. Esse número não é fixo, mas o limite máximo é de 100 imagens.