Comfy-Org/ComfyUI · error · ValueError

Gemini API blocked the request. Reason: {feedback.blockReaso

Error message

Gemini API blocked the request. Reason: {feedback.blockReason} ({feedback.blockReasonMessage})

What it means

Raised while extracting parts from a Gemini generateContent response when response.candidates is empty AND response.promptFeedback.blockReason is set. This is Google's server-side prompt safety filter blocking the request before generation; both the machine-readable blockReason and the human-readable blockReasonMessage are surfaced.

Source

Thrown at comfy_api_nodes/nodes_gemini.py:170

        return False
    return fnmatch(mime.value, pattern)


def get_parts_by_type(response: GeminiGenerateContentResponse, part_type: Literal["text"] | str) -> list[GeminiPart]:
    """
    Filter response parts by their type.

    Args:
        response: The API response from Gemini.
        part_type: Type of parts to extract ("text" or a MIME type).

    Returns:
        List of response parts matching the requested type.
    """
    if not response.candidates:
        if response.promptFeedback and response.promptFeedback.blockReason:
            feedback = response.promptFeedback
            raise ValueError(
                f"Gemini API blocked the request. Reason: {feedback.blockReason} ({feedback.blockReasonMessage})"
            )
        raise ValueError(
            "Gemini API returned no response candidates. If you are using the `IMAGE` modality, "
            "try changing it to `IMAGE+TEXT` to view the model's reasoning and understand why image generation failed."
        )
    parts = []
    blocked_reasons = []
    for candidate in response.candidates:
        if candidate.finishReason and candidate.finishReason.upper() == "IMAGE_PROHIBITED_CONTENT":
            blocked_reasons.append(candidate.finishReason)
            continue
        if candidate.content is None or candidate.content.parts is None:
            continue
        for part in candidate.content.parts:
            if part_type == "text" and part.text:
                parts.append(part)
            elif part.inlineData and _mime_matches(part.inlineData.mimeType, part_type):

View on GitHub (pinned to 1c6d8d45b3)

Solutions

Match the blockReason name to the cause: SAFETY -> rephrase prompt; COPYRIGHT -> remove celebrity/brand-style references; OTHER -> simplify and retry.
Remove or replace reference images that may be triggering the filter.
If the content is legitimate (medical, news, security research), add clinical/professional framing to the prompt.
Adjust the node's safety settings if exposed, then retry.

Defensive patterns

Strategy: try-catch

Try / catch

try:
    parts = extract_parts(response, "text")
except ValueError as e:
    if "blocked the request" in str(e):
        # prompt-level block: rephrase or swap images; retrying unchanged will fail again
        raise PolicyBlocked(str(e)) from e
    raise

Prevention

Keep prompts and attached images inside Gemini's content policy.
Map blockReason names (SAFETY, COPYRIGHT, ...) to concrete prompt changes instead of blind retries.
In automated pipelines, catch block errors and route to a sanitized-prompt fallback.

When it happens

Trigger: Gemini API returns no candidates with promptFeedback.blockReason such as SAFETY, LANGUAGE, COPYRIGHT, or PROHIBITED_CONTENT — triggered by the prompt text, inline image content, or system instruction.

Common situations: Prompts with violence/sexual/political content; requests about real people that trip COPYRIGHT depictions filters; images with disallowed content attached as reference; aggressive safetySettings configured upstream.

Related errors

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/eb5a6c757caac2d6. Report an issue: GitHub.