Gpt-image-1 Model Parameter Ignored by Image Variations API

API
1

Hey OpenAI dev community!

I’ve discovered what appears to be a discrepancy between the documentation and actual behavior of the Image Variations API when using gpt-image-1. I’m hoping someone can confirm if this is expected behavior or a bug.

The Problem

According to the documentation, gpt-image-1 supports non-square sizes:

  • 1024x1024 (square)
  • 1536x1024 (landscape)
  • 1024x1536 (portrait)
  • auto (default)

However, when I explicitly set model: "gpt-image-1" in a request to the variations endpoint (/v1/images/variations), the API rejects non-square sizes with an error message that lists only DALL·E sizes.

Reproduction Code

Here’s a simplified version of what I’m sending:

// Create FormData
const formData = new FormData();
formData.append("model", "gpt-image-1");  // Explicitly set model
formData.append("size", "1024x1536");     // Portrait size supported by gpt-image-1
formData.append("response_format", "b64_json");
formData.append("n", "1");
formData.append("image", imageBlob, "image.png");

// Send request
const response = await fetch("https://api.openai.com/v1/images/variations", {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${apiKey}`,
  },
  body: formData,
});

Error Response

{
  "error": {
    "code": null,
    "message": "'1024x1536' is not one of ['256x256', '512x512', '1024x1024'] - 'size'",
    "param": null,
    "type": "invalid_request_error"
  }
}

Note that the error message lists sizes for DALL·E models, not gpt-image-1.

What I’ve Verified

  1. My organization is verified and has access to gpt-image-1
  2. The same API key works correctly with gpt-image-1 when using the generations endpoint
  3. I’ve added logging to confirm the model parameter is being correctly included in the FormData:
FormData entries:
"0": "model: gpt-image-1",
"1": "size: 1024x1536",
"2": "response_format: b64_json",
"3": "n: 1",
"4": "image: [blob]"

Questions for the Community

  1. Has anyone successfully used gpt-image-1 with the variations endpoint?
  2. Is there a different way to specify the model for variations that I’m missing?
  3. Is this a known limitation that should be added to the docs?

Thanks! Joe

2

Read closely:

The image shows API documentation for creating variations of a given image using the endpoint "https://api.openai.com/v1/images/variations," specifying request body requirements for the image file and model. (Captioned by AI)
The image shows API documentation for creating variations of a given image using the endpoint "https://api.openai.com/v1/images/variations," specifying request body requirements for the image file and model. (Captioned by AI)586×324 7.76 KB

Variations does not take a prompt, and is a dalle-2 image-to-image that is quite poor in quality.

When gpt-image-1 can make very close copies of originals, to almost fool you that it does infill on an existing image instead of making a new one, you’d want the guidance of sending to edits with a prompt to instruct the transformation you desire.

1 Like