microsoft/semantic-kernel · error · ServiceResponseException

Failed to generate image.

Error message

Failed to generate image.

What it means

Raised by the deprecated generate_image method after the OpenAI images API returns. It checks that response.data exists and that its first element carries either a url or b64_json; if data is empty/None or the first image has neither field, it throws ServiceResponseException. This signals the upstream API returned an unusable payload.

Source

Thrown at python/semantic_kernel/connectors/ai/open_ai/services/open_ai_text_to_image_base.py:72

            if settings.size and not settings.size.height:
                settings.size.height = height
        if not settings.size and width and height:
            settings.size = ImageSize(width=width, height=height)

        if not settings.prompt:
            settings.prompt = description

        if not settings.prompt:
            raise ServiceInvalidRequestError("Prompt is required.")

        if not settings.ai_model_id:
            settings.ai_model_id = self.ai_model_id

        response = await self._send_request(settings)

        assert isinstance(response, ImagesResponse)  # nosec
        if not response.data or not (response.data[0].url or response.data[0].b64_json):
            raise ServiceResponseException("Failed to generate image.")

        return response.data[0].url or response.data[0].b64_json  # type: ignore[return-value]

    async def generate_images(
        self,
        prompt: str,
        settings: PromptExecutionSettings | None = None,
        **kwargs: Any,
    ) -> list[str]:
        """Generate one or more images from text. Returns URLs or base64-encoded images.

        Args:
            prompt: Description of the image(s) to generate.
            settings: Execution settings for the prompt.
            kwargs: Additional arguments, check the openai images.generate documentation for the supported arguments.

        Returns:
            list[str]: Image URLs or base64-encoded images.

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Retry the request with a less policy-sensitive prompt
2. Set response_format explicitly in settings (e.g. url or b64_json) to match what you expect to read back
3. Verify you are hitting the real OpenAI endpoint and not a proxy that altered the response
4. Migrate to generate_images, which has a more defensive data check and returns a list

from semantic_kernel.exceptions.service_exceptions import ServiceResponseException

try:
    img = await service.generate_image(description=desc)
except ServiceResponseException as e:
    if "Failed to generate image" in str(e):
        # upstream returned empty/ unusable payload; retry or fall back
        ...
    raise

When it happens

Trigger: The OpenAI image endpoint returned a 200 but with an empty data array, or response.data[0] has neither url nor b64_json (e.g. a content-filtered or partially-flagged response, or an unexpected response shape from a custom/proxy endpoint).

Common situations: Content policy filtered the generation so no image payload was returned; a gateway/proxy stripped fields; an unsupported response_format was requested; intermittent upstream issue.

Related errors

• Prompt is required.
• No valid image data found in response.
• Failed to edit image.
• The OpenAI text to image model ID is required.
• Provide either 'image_paths' or 'image_files', and only one.

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/79a4c40d172eb644. Report an issue: GitHub.