add support for Amazon Nova Canvas model (#7838)

* add initial support for Amazon Nova Canvas model Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * adjust name to AmazonNovaCanvas and map function variables to config Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * tighten model name check Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * fix quality mapping Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * add premium quality in config Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * support all Amazon Nova Canvas tasks * remove unused import Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * add tests for image generation tasks and fix payload Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * add missing util file Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * update model prices backup file Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> * remove image tasks other than text->image Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> --------- Signed-off-by: omrishiv <327609+omrishiv@users.noreply.github.com> Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
2026-09-11 22:51:28 +00:00 · 2025-03-10 08:02:00 -07:00 · 2025-03-10 08:02:00 -07:00 · 0674491386
commit 0674491386
parent 0fa3ec98ab
8 changed files with 212 additions and 22 deletions
--- a/litellm/init.py
+++ b/litellm/init.py
@ -899,6 +899,7 @@ from .llms.bedrock.chat.invoke_transformations.base_invoke_transformation import

 from .llms.bedrock.image.amazon_stability1_transformation import AmazonStabilityConfig
 from .llms.bedrock.image.amazon_stability3_transformation import AmazonStability3Config
+from .llms.bedrock.image.amazon_nova_canvas_transformation import AmazonNovaCanvasConfig
 from .llms.bedrock.embed.amazon_titan_g1_transformation import AmazonTitanG1Config
 from .llms.bedrock.embed.amazon_titan_multimodal_transformation import (
    AmazonTitanMultimodalEmbeddingG1Config,
--- a/litellm/llms/bedrock/image/amazon_nova_canvas_transformation.py
+++ b/litellm/llms/bedrock/image/amazon_nova_canvas_transformation.py
@ -0,0 +1,106 @@
+import types
+from typing import List, Optional
+
+from openai.types.image import Image
+
+from litellm.types.llms.bedrock import (
+    AmazonNovaCanvasTextToImageRequest, AmazonNovaCanvasTextToImageResponse,
+    AmazonNovaCanvasTextToImageParams, AmazonNovaCanvasRequestBase,
+)
+from litellm.types.utils import ImageResponse
+
+
+class AmazonNovaCanvasConfig:
+    """
+    Reference: https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/model-catalog/serverless/amazon.nova-canvas-v1:0
+
+    """
+
+    @classmethod
+    def get_config(cls):
+        return {
+            k: v
+            for k, v in cls.__dict__.items()
+            if not k.startswith("__")
+               and not isinstance(
+                v,
+                (
+                    types.FunctionType,
+                    types.BuiltinFunctionType,
+                    classmethod,
+                    staticmethod,
+                ),
+            )
+               and v is not None
+        }
+
+    @classmethod
+    def get_supported_openai_params(cls, model: Optional[str] = None) -> List:
+        """
+        """
+        return ["n", "size", "quality"]
+
+    @classmethod
+    def _is_nova_model(cls, model: Optional[str] = None) -> bool:
+        """
+        Returns True if the model is a Nova Canvas model
+
+        Nova models follow this pattern:
+
+        """
+        if model:
+            if "amazon.nova-canvas" in model:
+                return True
+        return False
+
+    @classmethod
+    def transform_request_body(
+            cls, text: str, optional_params: dict
+    ) -> AmazonNovaCanvasRequestBase:
+        """
+        Transform the request body for Amazon Nova Canvas model
+        """
+        task_type = optional_params.pop("taskType", "TEXT_IMAGE")
+        image_generation_config = optional_params.pop("imageGenerationConfig", {})
+        image_generation_config = {**image_generation_config, **optional_params}
+        if task_type == "TEXT_IMAGE":
+            text_to_image_params = image_generation_config.pop("textToImageParams", {})
+            text_to_image_params = {"text" :text, **text_to_image_params}
+            text_to_image_params = AmazonNovaCanvasTextToImageParams(**text_to_image_params)
+            return AmazonNovaCanvasTextToImageRequest(textToImageParams=text_to_image_params, taskType=task_type,
+                                                      imageGenerationConfig=image_generation_config)
+        raise NotImplementedError(f"Task type {task_type} is not supported")
+
+    @classmethod
+    def map_openai_params(cls, non_default_params: dict, optional_params: dict) -> dict:
+        """
+        Map the OpenAI params to the Bedrock params
+        """
+        _size = non_default_params.get("size")
+        if _size is not None:
+            width, height = _size.split("x")
+            optional_params["width"], optional_params["height"] = int(width), int(height)
+        if non_default_params.get("n") is not None:
+            optional_params["numberOfImages"] = non_default_params.get("n")
+        if non_default_params.get("quality") is not None:
+            if non_default_params.get("quality") in ("hd", "premium"):
+                optional_params["quality"] = "premium"
+            if non_default_params.get("quality") == "standard":
+                optional_params["quality"] = "standard"
+        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
+        """
+
+        nova_response = AmazonNovaCanvasTextToImageResponse(**response_dict)
+        openai_images: List[Image] = []
+        for _img in nova_response.get("images", []):
+            openai_images.append(Image(b64_json=_img))
+
+        model_response.data = openai_images
+        return model_response
--- a/litellm/llms/bedrock/image/image_handler.py
+++ b/litellm/llms/bedrock/image/image_handler.py
@ -266,6 +266,8 @@ class BedrockImageGeneration(BaseAWSLLM):
                    "text_prompts": [{"text": prompt, "weight": 1}],
                    **inference_params,
                }
+        elif provider == "amazon":
+            return dict(litellm.AmazonNovaCanvasConfig.transform_request_body(text=prompt, optional_params=optional_params))
        else:
            raise BedrockError(
                status_code=422, message=f"Unsupported model={model}, passed in"
@ -301,6 +303,7 @@ class BedrockImageGeneration(BaseAWSLLM):
        config_class = (
            litellm.AmazonStability3Config
            if litellm.AmazonStability3Config._is_stability_3_model(model=model)
+            else litellm.AmazonNovaCanvasConfig if litellm.AmazonNovaCanvasConfig._is_nova_model(model=model)
            else litellm.AmazonStabilityConfig
        )
        config_class.transform_response_dict_to_openai_response(
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@ -6434,7 +6434,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-micro-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 300000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.000000035,
@ -6472,7 +6472,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-lite-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 128000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.00000006,
@ -6514,7 +6514,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-pro-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 300000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.0000008,
@ -6527,6 +6527,12 @@
        "supports_prompt_caching": true,
        "supports_response_schema": true
    },
+    "1024-x-1024/50-steps/bedrock/amazon.nova-canvas-v1:0": {
+      "max_input_tokens": 2600,
+      "output_cost_per_image": 0.06,
+      "litellm_provider": "bedrock",
+      "mode": "image_generation"
+    },
    "eu.amazon.nova-pro-v1:0": {
        "max_tokens": 4096, 
        "max_input_tokens": 300000,
@ -7871,22 +7877,22 @@
        "mode": "image_generation"
    },
    "stability.sd3-5-large-v1:0": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.08,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
    },
    "stability.stable-image-core-v1:0": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.04,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
    },
    "stability.stable-image-core-v1:1": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.04,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
@ -7899,8 +7905,8 @@
        "mode": "image_generation"
    },
    "stability.stable-image-ultra-v1:1": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.14,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
--- a/litellm/types/llms/bedrock.py
+++ b/litellm/types/llms/bedrock.py
@ -365,6 +365,63 @@ class AmazonStability3TextToImageResponse(TypedDict, total=False):
    finish_reasons: List[str]


+class AmazonNovaCanvasRequestBase(TypedDict, total=False):
+    """
+    Base class for Amazon Nova Canvas API requests
+    """
+
+    pass
+
+
+class AmazonNovaCanvasImageGenerationConfig(TypedDict, total=False):
+    """
+    Config for Amazon Nova Canvas Text to Image API
+
+    Ref: https://docs.aws.amazon.com/nova/latest/userguide/image-gen-req-resp-structure.html
+    """
+
+    cfgScale: int
+    seed: int
+    quality: Literal["standard", "premium"]
+    width: int
+    height: int
+    numberOfImages: int
+
+
+class AmazonNovaCanvasTextToImageParams(TypedDict, total=False):
+    """
+    Params for Amazon Nova Canvas Text to Image API
+    """
+
+    text: str
+    negativeText: str
+    controlStrength: float
+    controlMode: Literal["CANNY_EDIT", "SEGMENTATION"]
+    conditionImage: str
+
+
+class AmazonNovaCanvasTextToImageRequest(AmazonNovaCanvasRequestBase, TypedDict, total=False):
+    """
+    Request for Amazon Nova Canvas Text to Image API
+
+    Ref: https://docs.aws.amazon.com/nova/latest/userguide/image-gen-req-resp-structure.html
+    """
+
+    textToImageParams: AmazonNovaCanvasTextToImageParams
+    taskType: Literal["TEXT_IMAGE"]
+    imageGenerationConfig: AmazonNovaCanvasImageGenerationConfig
+
+
+class AmazonNovaCanvasTextToImageResponse(TypedDict, total=False):
+    """
+    Response for Amazon Nova Canvas Text to Image API
+
+    Ref: https://docs.aws.amazon.com/nova/latest/userguide/image-gen-req-resp-structure.html
+    """
+
+    images: List[str]
+
+
 if TYPE_CHECKING:
    from botocore.awsrequest import AWSPreparedRequest
 else:
--- a/litellm/utils.py
+++ b/litellm/utils.py
@ -2427,6 +2427,7 @@ def get_optional_params_image_gen(
        config_class = (
            litellm.AmazonStability3Config
            if litellm.AmazonStability3Config._is_stability_3_model(model=model)
+            else litellm.AmazonNovaCanvasConfig if litellm.AmazonNovaCanvasConfig._is_nova_model(model=model)
            else litellm.AmazonStabilityConfig
        )
        supported_params = config_class.get_supported_openai_params(model=model)
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@ -6434,7 +6434,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-micro-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 300000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.000000035,
@ -6472,7 +6472,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-lite-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 128000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.00000006,
@ -6514,7 +6514,7 @@
        "supports_response_schema": true
    },
    "us.amazon.nova-pro-v1:0": {
-        "max_tokens": 4096, 
+        "max_tokens": 4096,
        "max_input_tokens": 300000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.0000008,
@ -6527,6 +6527,12 @@
        "supports_prompt_caching": true,
        "supports_response_schema": true
    },
+    "1024-x-1024/50-steps/bedrock/amazon.nova-canvas-v1:0": {
+      "max_input_tokens": 2600,
+      "output_cost_per_image": 0.06,
+      "litellm_provider": "bedrock",
+      "mode": "image_generation"
+    },
    "eu.amazon.nova-pro-v1:0": {
        "max_tokens": 4096, 
        "max_input_tokens": 300000,
@ -7871,22 +7877,22 @@
        "mode": "image_generation"
    },
    "stability.sd3-5-large-v1:0": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.08,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
    },
    "stability.stable-image-core-v1:0": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.04,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
    },
    "stability.stable-image-core-v1:1": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.04,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
@ -7899,8 +7905,8 @@
        "mode": "image_generation"
    },
    "stability.stable-image-ultra-v1:1": {
-        "max_tokens": 77, 
-        "max_input_tokens": 77, 
+        "max_tokens": 77,
+        "max_input_tokens": 77,
        "output_cost_per_image": 0.14,
        "litellm_provider": "bedrock",
        "mode": "image_generation"
--- a/tests/image_gen_tests/test_image_generation.py
+++ b/tests/image_gen_tests/test_image_generation.py
@ -130,6 +130,16 @@ class TestBedrockSd1(BaseImageGenTest):
        return {"model": "bedrock/stability.sd3-large-v1:0"}


+class TestBedrockNovaCanvasTextToImage(BaseImageGenTest):
+    def get_base_image_generation_call_args(self) -> dict:
+        litellm.in_memory_llm_clients_cache = InMemoryCache()
+        return {"model": "bedrock/amazon.nova-canvas-v1:0",
+                "n": 1,
+                "size": "320x320",
+                "imageGenerationConfig": {"cfgScale":6.5,"seed":12},
+                "taskType": "TEXT_IMAGE"}
+
+
 class TestOpenAIDalle3(BaseImageGenTest):
    def get_base_image_generation_call_args(self) -> dict:
        return {"model": "dall-e-3"}