BerriAI/litellm · error · BedrockError

"; ".join(finish_reasons)

Error message

"; ".join(finish_reasons)

What it means

After a Stability 3 image request on Bedrock returns 200, the response is parsed into AmazonStability3TextToImageResponse. If finish_reasons[] contains any non-empty entries, Bedrock reported generation failures and litellm raises BedrockError(status_code=400) with the reasons joined by '; '. This is a server-side content- or input-quality failure delivered in a successful HTTP response.

Source

Thrown at litellm/llms/bedrock/image_generation/amazon_stability3_transformation.py:98

        No OpenAI params are mapped for Stability 3, so directly return the optional_params
        """
        return optional_params

    @classmethod
    def transform_response_dict_to_openai_response(
        cls, model_response: ImageResponse, response_dict: dict
    ) -> ImageResponse:
        """
        Transform the response dict to the OpenAI response
        """

        stability_3_response: Final = AmazonStability3TextToImageResponse(**response_dict)

        finish_reasons = stability_3_response.get("finish_reasons", [])
        finish_reasons = [reason for reason in finish_reasons if reason]
        if len(finish_reasons) > 0:
            raise BedrockError(status_code=400, message="; ".join(finish_reasons))

        openai_images: Final[list[Image]] = []
        for _img in stability_3_response.get("images", []):
            openai_images.append(Image(b64_json=_img))

        model_response.data = openai_images
        return model_response

    @classmethod
    def cost_calculator(
        cls,
        model: str,
        image_response: ImageResponse,
        size: str | None = None,
        optional_params: dict | None = None,
    ) -> float:
        get_model_info: Final = get_cached_model_info()
        model_info: Final = get_model_info(

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Read the joined finish_reasons in the BedrockError message — it states the exact cause (usually content filter or invalid input).
2. Rephrase the prompt to remove flagged terms; try a simpler prompt to confirm the filter is the cause.
3. Check/adjust Stability-specific params (aspect_ratio, seed, negative_prompt) against the Stability 3 API spec.
4. If the filter is a false positive, switch to a different image model (e.g. Nova Canvas / Titan Image).

from litellm.exceptions import BedrockError

try:
    resp = litellm.image_generation(model="bedrock/stability.stable-image-core-1:0", prompt=p)
except BedrockError as e:
    if e.status_code == 400 and "finish_reason" not in str(e):
        raise  # unrelated 400
    log.warning("Stability rejected prompt: %s", e.message)
    return fallback_prompt_or_model(p)

When it happens

Trigger: Calling litellm.image_generation() with model='bedrock/stability.stable-image-core-1:0' (or ultra/core variants) where Stability rejects the prompt (content filter) or the request has invalid parameters; finish_reasons like 'Invalid prompt: Input text prompt contains unsafe content' come back non-empty.

Common situations: Prompts with NSFW/violent content triggering Stability's filter, negative-prompt misuse, aspect_ratio/seed-strength values Stability rejects, or intermittent safety-filter false positives.

Related errors

• Model needs to be set for bedrock
• completion_response must be of type ImageResponse for bedroc
• Parameter {k} is not supported for model {model}. Supported
• Error transforming text to image params: {e}. Got params: {t
• Error transforming image generation config: {e}. Got params:

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/b12243596d32a17c. Report an issue: GitHub.