fix(images): forward n/response_format/size/user in image_variation

`image_variation()` declares `n`, `response_format`, `size` and `user`, but
built the request with a hard-coded `optional_params={}` for both the OpenAI
and Topaz branches, so none of them ever reached a provider.
`litellm.image_variation(image=f, n=4, size="512x512")` silently returned one
image at the provider default size.

The mapping layer already existed and was simply never called:
`ProviderConfigManager.get_provider_image_variation_config` returns a config
whose `map_openai_params` handles these (OpenAI passes them through; Topaz
renames `size` to `output_width`/`output_height` and `response_format` to
`output_format`). Build `optional_params` through it, as `image_generation`
does.

`n` and `response_format` become `| None = None`, matching
`image_generation`, so an omitted argument stays omitted instead of pinning
`n=1`/`response_format="url"` onto every provider.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

litellm/images/main.py

@@ -616,8 +616,8 @@ async def aimage_variation(*args, **kwargs) -> ImageResponse:
 def image_variation(
     image: FileTypes,
     model: str = "dall-e-2",  # set to dall-e-2 by default - like OpenAI.
-    n: int = 1,
-    response_format: Literal["url", "b64_json"] = "url",
+    n: int | None = None,
+    response_format: Literal["url", "b64_json"] | None = None,
     size: str | None = None,
     user: str | None = None,
     **kwargs,
@@ -662,6 +662,33 @@ def image_variation(
     api_key: Final = provider_config.get_api_key(litellm_params.get("api_key", None))
     api_base = provider_config.get_api_base(litellm_params.get("api_base", None))
 
+    # Map the OpenAI-style arguments onto the provider's request. Without this
+    # every provider was called with an empty dict, so `n`, `response_format`,
+    # `size` and `user` never left this function.
+    image_variation_config: Final = ProviderConfigManager.get_provider_image_variation_config(
+        model=model,
+        provider=llm_provider,
+    )
+    non_default_params: Final = {
+        key: value
+        for key, value in {
+            "n": n,
+            "response_format": response_format,
+            "size": size,
+            "user": user,
+        }.items()
+        if value is not None
+    }
+    optional_params: dict = {}
+    if image_variation_config is not None and non_default_params:
+        drop_params: Final = kwargs.get("drop_params")
+        optional_params = image_variation_config.map_openai_params(
+            non_default_params=non_default_params,
+            optional_params=optional_params,
+            model=model,
+            drop_params=bool(litellm.drop_params if drop_params is None else drop_params),
+        )
+
     if image_variation_provider == LITELLM_IMAGE_VARIATION_PROVIDERS.OPENAI:
         if api_key is None:
             raise ValueError("API key is required for OpenAI image variations")
@@ -677,7 +704,7 @@ def image_variation(
             timeout=litellm_params.get("timeout", None),
             custom_llm_provider=custom_llm_provider,
             logging_obj=litellm_logging_obj,
-            optional_params={},
+            optional_params=optional_params,
             litellm_params=litellm_params,
         )
     elif image_variation_provider == LITELLM_IMAGE_VARIATION_PROVIDERS.TOPAZ:
@@ -695,7 +722,7 @@ def image_variation(
             timeout=litellm_params.get("timeout", None) or DEFAULT_REQUEST_TIMEOUT,
             custom_llm_provider=custom_llm_provider,
             logging_obj=litellm_logging_obj,
-            optional_params={},
+            optional_params=optional_params,
             litellm_params=litellm_params,
             client=client,
         )

tests/unit/images/test_image_variation_optional_params.py

@@ -0,0 +1,77 @@
+"""`image_variation()` must forward its OpenAI-style arguments to the provider.
+
+`n`, `response_format`, `size` and `user` are declared on the public function
+but were dropped before the request was built.
+"""
+
+from unittest.mock import patch
+
+import litellm
+from litellm.images.main import image_variation
+
+
+def _image_response() -> "litellm.utils.ImageResponse":
+    return litellm.utils.ImageResponse(
+        created=1234567890,
+        data=[{"url": "https://example.com/image.png"}],
+    )
+
+
+def _optional_params(mock_call) -> dict:
+    return mock_call.call_args.kwargs["optional_params"]
+
+
+class TestImageVariationOptionalParams:
+    @patch("litellm.images.main.openai_image_variations")
+    def test_openai_receives_the_requested_params(self, mock_openai) -> None:
+        mock_openai.image_variations.return_value = _image_response()
+
+        image_variation(
+            image=b"fake-image-bytes",
+            model="openai/dall-e-2",
+            n=4,
+            size="512x512",
+            response_format="b64_json",
+            user="user-123",
+            api_key="sk-test",
+            api_base="https://api.openai.com/v1",
+        )
+
+        optional_params = _optional_params(mock_openai.image_variations)
+        assert optional_params["n"] == 4
+        assert optional_params["size"] == "512x512"
+        assert optional_params["response_format"] == "b64_json"
+        assert optional_params["user"] == "user-123"
+
+    @patch("litellm.images.main.openai_image_variations")
+    def test_omitted_params_are_not_invented(self, mock_openai) -> None:
+        """Nothing the caller left out may be sent, so provider defaults stand."""
+        mock_openai.image_variations.return_value = _image_response()
+
+        image_variation(
+            image=b"fake-image-bytes",
+            model="openai/dall-e-2",
+            api_key="sk-test",
+            api_base="https://api.openai.com/v1",
+        )
+
+        assert _optional_params(mock_openai.image_variations) == {}
+
+    @patch("litellm.images.main.base_llm_aiohttp_handler")
+    def test_topaz_maps_size_and_response_format(self, mock_handler) -> None:
+        """Topaz renames these, which only happens if map_openai_params runs."""
+        mock_handler.image_variations.return_value = _image_response()
+
+        image_variation(
+            image=b"fake-image-bytes",
+            model="topaz/Standard V2",
+            size="1024x768",
+            response_format="b64_json",
+            api_key="topaz-key",
+            api_base="https://api.topazlabs.com",
+        )
+
+        optional_params = _optional_params(mock_handler.image_variations)
+        assert optional_params["output_width"] == "1024"
+        assert optional_params["output_height"] == "768"
+        assert optional_params["output_format"] == "b64_json"