microsoft/semantic-kernel · error · ServiceInvalidExecutionSettingsError

The model does not support multi-modality

Error message

The model does not support multi-modality

What it means

Raised inside _get_images_from_history when chat_history contains an ImageContent item but the loaded model is not multi-modal (self.enable_multi_modality is False). The ONNX model was loaded without a 'vision' key in its config, so it cannot process images. Raised as ServiceInvalidExecutionSettingsError during the chat completion call.

Source

Thrown at python/semantic_kernel/connectors/ai/onnx/services/onnx_gen_ai_chat_completion.py:191

        return self.tokenizer.apply_chat_template(
            json.dumps(self._chat_messages_to_dicts(chat_history)),
            add_generation_prompt=True,
        )

    def _chat_messages_to_dicts(self, chat_history: "ChatHistory") -> list[dict[str, Any]]:
        return [
            message.to_dict(role_key="role", content_key="content")
            for message in chat_history.messages
            if isinstance(message, ChatMessageContent)
        ]

    def _get_images_from_history(self, chat_history: "ChatHistory") -> list[ImageContent] | None:
        images = []
        for message in chat_history.messages:
            for image in message.items:
                if isinstance(image, ImageContent):
                    if not self.enable_multi_modality:
                        raise ServiceInvalidExecutionSettingsError("The model does not support multi-modality")
                    if image.uri:
                        images.append(image)
                    else:
                        raise ServiceInvalidExecutionSettingsError(
                            "Image Content URI needs to be set, because onnx can only work with file paths"
                        )
        return images if len(images) else None

    def _get_audios_from_history(self, chat_history: "ChatHistory") -> list[AudioContent] | None:
        audios = []
        for message in chat_history.messages:
            for audio in message.items:
                if isinstance(audio, AudioContent):
                    if not self.enable_multi_modality:
                        raise ServiceInvalidExecutionSettingsError("The model does not support multi-modality")
                    if audio.uri:
                        audios.append(audio)
                    else:

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Use a multi-modal (vision) ONNX model if you need image inputs
2. Remove ImageContent items from the chat history before sending to a text-only model
3. Switch to OnnxGenAIChatCompletion with a vision model and a template

Example fix

// before (text-only model + image in history)
history.add_message(ChatMessageContent(role=AuthorRole.USER, items=[ImageContent(uri='...')]))
await chat.get_chat_message_contents(chat_history=history, settings=s)
// after (remove image or use vision model)
history.add_message(ChatMessageContent(role=AuthorRole.USER, items=[TextContent(text='describe the scene')]))

from semantic_kernel.contents import ImageContent

def check_images_supported(chat_service, chat_history):
    if not chat_service.enable_multi_modality:
        for msg in chat_history.messages:
            if any(isinstance(item, ImageContent) for item in msg.items):
                raise ValueError('Chat history contains images but model is not multi-modal')

check_images_supported(chat, history)

from semantic_kernel.contents import ImageContent

def history_has_images(chat_history) -> bool:
    return any(
        isinstance(item, ImageContent)
        for msg in chat_history.messages
        for item in msg.items
    )

from semantic_kernel.exceptions.service_exceptions import ServiceInvalidExecutionSettingsError

try:
    result = await chat.get_chat_message_contents(chat_history=history, settings=settings)
except ServiceInvalidExecutionSettingsError as e:
    if 'does not support multi-modality' in str(e):
        logger.error('Remove image content or switch to a vision model')
        raise

When it happens

Trigger: Calling get_chat_message_contents on a text-only ONNX model with chat_history that includes one or more ImageContent items. The check fires per-message during _get_images_from_history iteration.

Common situations: Switching from a vision model to a text-only model but keeping image inputs in the pipeline; adding image content to a chat that was designed for text-only; using the wrong model folder that happens to be text-only.

Related errors

• Image Content URI needs to be set, because onnx can only wor
• Audio Content URI needs to be set, because onnx can only wor
• ImageContent in function result must contain binary data.
• When using a multi-modal model, a template must be specified
• ImageContent cannot be converted to ResponseContentPart. Onl

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/088ca07bee3cb91d. Report an issue: GitHub.