[Bug] OpenAI Responses `model_type` fails with `Image` inputs

[dbreunig]

What happened?

When using the Responses API for OpenAI models, Image inputs aren't handled correctly.

Given the following code:

import python

class ImageDescriptionGeneratorSignature(dspy.Signature):
    image: dspy.Image = dspy.InputField(desc="")
    image_description: str = dspy.OutputField(desc="")

descriptor = dspy.Predict(ImageDescriptionGeneratorSignature)

lm = dspy.LM(
    model="openai/gpt-5-mini",
    api_key="MY-KEY",
    max_tokens=16000,
    temperature=1.0,
    model_type="responses"
)

dspy.configure(lm=lm)
img = dspy.Image("overload.jpg")

result = descriptor(image=img)
print(result.image_description)

This throws a LiteLLM error:

BadRequestError: litellm.BadRequestError: OpenAIException - {
  "error": {
    "message": "Invalid value: 'text'. Supported values are: 'input_text', 'input_image', 'output_text', 'refusal', 'input_file', 'computer_screenshot', and 'summary_text'.",
    "type": "invalid_request_error",
    "param": "input[0].content[1].type",
    "code": "invalid_value"
  }
}

To fix the issue, we need to make sure we're formatting images correctly for the Responses API.

Steps to reproduce

import python

class ImageDescriptionGeneratorSignature(dspy.Signature):
image: dspy.Image = dspy.InputField(desc="")
image_description: str = dspy.OutputField(desc="")

descriptor = dspy.Predict(ImageDescriptionGeneratorSignature)

lm = dspy.LM(
model="openai/gpt-5-mini",
api_key="MY-KEY",
max_tokens=16000,
temperature=1.0,
model_type="responses"
)

dspy.configure(lm=lm)
img = dspy.Image("overload.jpg")

result = descriptor(image=img)
print(result.image_description)

DSPy version

3.0.4b1

[TomeHirata]

Hi, @dbreunig. Can you help verify if this fix solves the issue?