Need help on how to use a fileID uploaded for an image

[rodriciru]

Hi.
Using the Responses API, currently I can upload a file to the files API and use that in a vector store if this file is an .docx for example.

VectorStoreClient vectorStoreClient = new VectorStoreClient(apiKey);
VectorStore createStoreOp = await vectorStoreClient.CreateVectorStoreAsync(
    new VectorStoreCreationOptions()
    {
        FileIds = { uploadedFileId },
    });
creationOptions.Tools.Add(ResponseTool.CreateFileSearchTool([createStoreOp.Id], null));

But if I want to upload an image, the vector store won't work, as it's only for text docs. So I want to use as an ImageContent()
The problem is that ImageContent() wants a URI, so "file-sadkhjasdjhas8sadjh" won't work.
I don't know how to pass a file ID to make use of that file.
The API allows that:

 curl https://api.openai.com/v1/responses   -H "Content-Type: application/json"   -H "Authorization: Bearer TOKEN"   -d '{
    "model": "gpt-4.1",
    "input": [
      {
        "role": "user",
        "content": [
          {"type": "input_text", "text": "what is in this image?"},
          {
            "type": "input_image",
            "file_id": "file-G1qgzqZ8JSuyYhnB1FFJ2F"
          }
        ]
      }
    ]
  }'

Also, FileReferenceContent() put a type of input_file so the API assumes it's not an image and fails.
A combination of both will be so good.
Allow in Responses API vision of images by FileID.

[Samielakkad]

Your reading is correct: the current Semantic Kernel conversion cannot represent an image that is referenced by an uploaded file ID. FileReferenceContent is converted with ResponseContentPart.CreateInputFilePart(...), which produces input_file, while the OpenAI .NET Responses API has a separate CreateInputImagePart(string imageFileId, ...) factory for input_image.

So file-... should not be put into ImageContent.Uri; it is not a URI and that would only hide the type mismatch.

The reliable options today are:

1. For this message, use the OpenAI .NET Responses client directly and construct the user content with ResponseContentPart.CreateInputImagePart(uploadedFileId) plus the text part.
2. If acceptable for your application, read/download the image bytes and pass an ImageContent with the bytes and MIME type instead of a file ID.

Conceptually the direct OpenAI content is:

ResponseItem.CreateUserMessageItem([
    ResponseContentPart.CreateInputTextPart("Describe this image"),
    ResponseContentPart.CreateInputImagePart(uploadedFileId)
]);

This looks like a genuine abstraction gap in Semantic Kernel rather than incorrect usage. A distinct image-file-reference content type (or an explicit image purpose on the reference) would let the SK adapter choose CreateInputImagePart without guessing. Until such support is added, using the lower-level Responses client for this one request is safer than trying to coerce FileReferenceContent or a fake URI.