Error when using the DALL·E 3 model

API
2

I’ve got a little utility that does not download URLs, it receives via response format.

dall-e-2:

Two cartoon carrots with faces and arms are having a conversation outdoors with a blank speech bubble above them. (Captioned by AI)
Two cartoon carrots with faces and arms are having a conversation outdoors with a blank speech bubble above them. (Captioned by AI)1024×1024 106 KB

dall-e-3:

Two cheerful, cartoon carrots with expressive faces and arms are happily conversing in a vibrant, flowery garden under a bright, blue sky. (Captioned by AI)
Two cheerful, cartoon carrots with expressive faces and arms are happily conversing in a vibrant, flowery garden under a bright, blue sky. (Captioned by AI)1024×1024 142 KB

gpt-image-1:

image
image406×211 4.47 KB

Log of API call parameters passed (showing as Python streamable):

{'model': 'dall-e-2', 'size': '1024x1024', 'prompt': 'Two cartoon carrots are having a conversation in a garden.', 'response_format': 'b64_json', 'timeout': 40}
{'model': 'dall-e-3', 'size': '1024x1024', 'prompt': 'Two cartoon carrots are having a conversation in a garden.', 'response_format': 'b64_json', 'timeout': 40}
{'model': 'gpt-image-1', 'size': '1024x1024', 'prompt': 'Two cartoon carrots are having a conversation in a garden.', 'timeout': 60, 'background': None, 'moderation': None, 'output_compression': None, 'output_format': 'png', 'quality': 'medium', 'user': 'dalle-edit-user'}

before this snippet with the Python SDK:

        if PRINT_API_CALLS:
            print(image_params)
        response = await self.client.images.edit(**image_params)

        b64_data = response.data[0].b64_json
        image_data = base64.b64decode(b64_data)
        image_file = BytesIO(image_data)

Note how the response_format is NOT sent for the image-1 model, per API Reference.

There’s someone here that also could not resolve an issue with dall-e-2 on their organization - while their script ran fine for me.

So check that you are making the same call as were successful before, and then discover if it is your org that has gone wrong, if a new project, new API key, using that API key without sending organization or project headers can resolve the issue, on an innocent-sounding prompt.

2 Likes