HKUDS/DeepTutor · error · GenerationProviderError

Chat model returned no image. Check the model supports image

Error message

Chat model returned no image. Check the model supports image output (its output modalities must include `image`).

What it means

The chat-completions response parsed successfully but _extract_sources found no usable image in the assistant message, so the images list is empty. The message text tells you the likely root cause: the model does not declare `image` in its output modalities.

Source

Thrown at deeptutor/services/imagegen/adapters/chat_completions.py:66

        }
        payload: dict[str, Any] = {
            "model": config.model,
            "messages": [{"role": "user", "content": prompt}],
            "modalities": ["image", "text"],
        }

        logger.debug("imagegen(chat) url=%s model=%s", url, config.model)
        try:
            async with httpx.AsyncClient(timeout=config.request_timeout) as client:
                resp = await client.post(url, headers=headers, json=payload)
                raise_for_provider(resp, "Image generation")
                images = [
                    await self._materialize(client, src) for src in self._extract_sources(resp)
                ]
        except httpx.HTTPError as exc:
            raise GenerationProviderError(f"Image generation request error: {exc}") from exc
        if not images:
            raise GenerationProviderError(
                "Chat model returned no image. Check the model supports image output "
                "(its output modalities must include `image`)."
            )
        return images

    @staticmethod
    def _extract_sources(resp: httpx.Response) -> list[str]:
        """Pull image URLs / data URIs out of the assistant message."""
        data = resp.json()
        sources: list[str] = []
        choices = data.get("choices") if isinstance(data, dict) else None
        for choice in choices or []:
            message = (choice or {}).get("message") or {}
            for image in message.get("images") or []:
                if not isinstance(image, dict):
                    continue
                src = (image.get("image_url") or {}).get("url") or image.get("url")
                if isinstance(src, str) and src:

View on GitHub (pinned to 3e82f13042)

Solutions

1. Switch to a model that supports image output modalities (check the provider's model listing for `image` in output modalities)
2. Ensure the request payload requests image output (modality settings in extra body/config) as required by the provider
3. If image output isn't available on your chat endpoint, use the /images/generations adapter instead
4. Inspect the raw response (log resp.json()) to confirm what the model actually returned

Example fix

// before
adapter = ChatCompletionsImagegenAdapter()  # model: gpt-4o-mini (text-only)
// after
config.model = "gpt-4o"  # or a model whose output modalities include `image`

try:
    images = await chat_adapter.generate(prompt, config)
except GenerationProviderError as exc:
    if "no image" in str(exc):
        images = await openai_compat_adapter.generate(prompt, config)  # fallback endpoint

When it happens

Trigger: Calling a chat model (e.g. gpt-4o style) that only outputs text because the request didn't request image modality or the model/deployment doesn't support image output; the model returned images in a response shape the extractor doesn't recognize; the model refused or returned only text.

Common situations: Using a text-only model name with the chat-completions imagegen adapter; forgetting the modalities/image config in the payload; provider deployed without image output support; model returned a refusal instead of an image.

Related errors

• Image response had no image in the assistant message.
• Cohere model '{model_name}' does not support multimodal `con
• No endpoint URL configured for image generation.
• Image generation request error: {exc}
• Malformed image data URI.

AI-assisted analysis of HKUDS/DeepTutor@3e82f13042 (2026-08-27). Data as JSON: /api/errors/413b01551165ea78. Report an issue: GitHub.