Forum Discussion

RobQ0's avatar
RobQ0
Brass Contributor
Jun 17, 2025

Azure OpenAI - GPT-4.1 + tools/image_generation doesn't work

Hi,

 

response = client.responses.create(model='gpt-4.1', input=prompt, tools=[{'type': 'image_generation'}])

 

When used with Azure/OpenAI it fails with:

openai.BadRequestError: Error code: 400 - {'error': {'message': 'There was an issue with your request. Please check your inputs and try again', 'type': 'invalid_request_error', 'param': None, 'code': None}}

 

It works fine with OpenAI direct.

 

 

1 Reply

  • hazem's avatar
    hazem
    MVP
    Jul 22, 2025

    Hi RobQ0​,

    Please keep in mind that when you call tools=[{"type":"image_generation"}] in Azure’s Responses API, you must tell Azure OpenAI which image model deployment to use, otherwise it doesn’t know where to send your DALL·E-style request and returns a generic 400.

    However, You can deploy the GPT-image-1 model In the Azure Portal

     

    OR, Instantiate your client with the image‐deployment header When you create your AzureOpenAI client.

    Add a default header set to your deployment name

    x-ms-oai-image-generation-deployment

     

    Then, use the preview API version:

    from openai import AzureOpenAI
from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default"
)
client = AzureOpenAI(
    base_url="https://<YOUR-RESOURCE>.openai.azure.com/openai/v1/",
    azure_ad_token_provider=token_provider,
    api_version="preview",
    default_headers={
      "x-ms-oai-image-generation-deployment": "gpt-image-1-deployment"
    }
)

    This header routes any image_generation tool calls to your DALL·E deployment

     

    Invoke the image tool

    response = client.responses.create(
  model="gpt-4.1",
  input="Generate an image of a golden retriever puppy playing in autumn leaves",
  tools=[{"type": "image_generation"}]
)

# extract the Base64-encoded image from the response
image_b64 = next(
  out.result for out in response.output
  if out.type == "image_generation_call"
)
with open("puppy.png", "wb") as f:
  f.write(base64.b64decode(image_b64))
``` :contentReference[oaicite:1]{index=1}

    Make sure you’re in a supported region The Responses API (and image tool) is in preview in select regions, check that your resource is in one of: eastus, westus, francecentral, etc.

    With those steps, deploying gpt-image-1, adding the header, and using api_version="preview"—your tools=[{"type":"image_generation"}] calls will work just like they do against OpenAI’s platform.

Resources