BerriAI/litellm · error · ValueError

Error transforming image generation config: {e}. Got params:

Error message

Error transforming image generation config: {e}. Got params: {image_generation_config}, Expected params: {AmazonNovaCanvasImageGenerationConfig.__annotations__}

What it means

Independently of the task params, LiteLLM validates the merged imageGenerationConfig against AmazonNovaCanvasImageGenerationConfig (fields like numberOfImages, quality, cfgScale, seed, width, height). Constructor failures (unknown keys, out-of-range or wrongly-typed values) raise this ValueError showing both the received config and the expected annotations.

Source

Thrown at litellm/llms/bedrock/image_generation/amazon_nova_canvas_transformation.py:88

        # Extract model_id parameter to prevent "extraneous key" error from Bedrock API
        # Following the same pattern as chat completions and embeddings
        unencoded_model_id: Final = optional_params.pop("model_id", None)  # noqa: F841

        image_generation_config = {**image_generation_config, **optional_params}
        if task_type == "TEXT_IMAGE":
            text_to_image_params: dict[str, Any] = image_generation_config.pop("textToImageParams", {})
            text_to_image_params = {"text": text, **text_to_image_params}
            try:
                text_to_image_params_typed: Final = AmazonNovaCanvasTextToImageParams(**text_to_image_params)
            except Exception as e:
                raise ValueError(
                    f"Error transforming text to image params: {e}. Got params: {text_to_image_params}, Expected params: {AmazonNovaCanvasTextToImageParams.__annotations__}"
                )

            try:
                image_generation_config_typed = AmazonNovaCanvasImageGenerationConfig(**image_generation_config)
            except Exception as e:
                raise ValueError(
                    f"Error transforming image generation config: {e}. Got params: {image_generation_config}, Expected params: {AmazonNovaCanvasImageGenerationConfig.__annotations__}"
                )

            return AmazonNovaCanvasTextToImageRequest(
                textToImageParams=text_to_image_params_typed,
                taskType=task_type,
                imageGenerationConfig=image_generation_config_typed,
            )
        if task_type == "COLOR_GUIDED_GENERATION":
            color_guided_generation_params: dict[str, Any] = image_generation_config.pop(
                "colorGuidedGenerationParams", {}
            )
            color_guided_generation_params = {
                "text": text,
                **color_guided_generation_params,
            }
            try:
                color_guided_generation_params_typed: Final = AmazonNovaCanvasColorGuidedGenerationParams(

View on GitHub (pinned to 6c2dcb801b)

Solutions

  1. Match Got params against Expected params in the message; delete keys not in the expected set.
  2. Use native config keys with valid ranges: numberOfImages 1-4ish per model, width/height among supported pixel sizes, quality 'standard'|'premium', seed int, cfgScale float.
  3. Set drop_params=True to have unsupported extras dropped before validation where litellm supports it on this path.

Example fix

# before
litellm.image(model="bedrock/amazon.nova-canvas-v1:0",
             prompt="a cat", numberOfImages="2")  # str type -> ValueError
# after
litellm.image(model="bedrock/amazon.nova-canvas-v1:0",
             prompt="a cat", n=2)  # OpenAI param, mapped to numberOfImages int
Defensive patterns

Strategy: validation

Validate before calling

ALLOWED_CFG = {"numberOfImages", "quality", "cfgScale", "seed", "width", "height"}

def valid_generation_config(c: dict) -> bool:
    return (
        set(c) <= ALLOWED_CFG
        and all(not isinstance(v, str) for k, v in c.items() if k in {"numberOfImages", "seed", "width", "height"})
    )

Type guard

def is_valid_image_generation_config(c: object) -> bool:
    if not isinstance(c, dict):
        return False
    ints = {"numberOfImages", "seed", "width", "height"}
    return all(
        (k not in ints or (isinstance(v, int) and not isinstance(v, bool)))
        for k, v in c.items()
    )

Prevention

When it happens

Trigger: Passing keys Bedrock doesn't accept in imageGenerationConfig (e.g. 'style', 'response_format'), wrong types (seed as string, numberOfImages as float), or Bedrock-invalid values (width not in {512,768,1024...} depending on model, quality misspelled) for nova-canvas generations of any task type (TEXT_IMAGE, COLOR_GUIDED_GENERATION, etc.).

Common situations: Forwarding OpenAI gpt-image-1 params (background, output_compression, moderation) that land in the config dict; passing size strings not convertible to width/height ints; user-supplied option dicts passed verbatim.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/b3fe9acc200638ff. Report an issue: GitHub.