[Frontend][Responses] Accept chat-completions image format on /v1/responses (#46631)

tests/entrypoints/multimodal/openai/responses/test_image.py

@@ -113,6 +113,44 @@ async def test_single_chat_session_image_base64encoded(
     assert len(response.output_text) > 0
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+@pytest.mark.parametrize("part_type", ["input_image", "image_url"])
+@pytest.mark.parametrize("raw_image_url", TEST_IMAGE_ASSETS)
+async def test_single_chat_session_image_chat_completions_format(
+    client: openai.AsyncOpenAI,
+    model_name: str,
+    part_type: str,
+    raw_image_url: str,
+    url_encoded_image: dict[str, str],
+):
+    # #46631: accept chat-completions image shapes (image_url type or
+    # input_image with nested image_url, no detail)
+    content_text = "What's in this image?"
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": part_type,
+                    "image_url": {"url": url_encoded_image[raw_image_url]},
+                },
+                {"type": "input_text", "text": content_text},
+            ],
+        }
+    ]
+    response = await client.responses.create(
+        model=model_name,
+        input=messages,
+    )
+    assert response.status == "completed"
+    assert len(response.output) >= 1
+    message = next((out for out in response.output if out.type == "message"), None)
+    assert message is not None
+    assert message.role == "assistant"
+    assert len(response.output_text) > 0
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("model_name", [MODEL_NAME])
 @pytest.mark.parametrize(

tests/tool_use/test_responses_request_validations.py

@@ -208,6 +208,28 @@ def test_input_image_detail_defaults_or_preserves(detail: str | None) -> None:
     )
 
 
+@pytest.mark.parametrize("part_type", ["input_image", "image_url"])
+def test_input_image_accepts_chat_completions_format(part_type: str) -> None:
+    # #46631: chat-completions image parts must validate on /v1/responses.
+    item = {
+        "role": "user",
+        "content": [
+            {
+                "type": part_type,
+                "image_url": {"url": "https://example.com/image.png"},
+            }
+        ],
+    }
+
+    request = ResponsesRequest.model_validate({"input": [item]})
+
+    assert request.input[0]["content"][0] == {
+        "type": "input_image",
+        "image_url": "https://example.com/image.png",
+        "detail": "auto",
+    }
+
+
 # Regression tests for parallel_tool_calls=null crash in Responses API
 # (from_request() passed None to ResponsesResponse.parallel_tool_calls,
 #  a non-optional bool field, causing a Pydantic ValidationError during

vllm/entrypoints/openai/responses/protocol.py

@@ -136,8 +136,12 @@ class ResponseRawMessageAndToken(OpenAIBaseModel):
 ResponseInputOutputItem: TypeAlias = ResponseInputItemParam | ResponseOutputItem
 
 
-def _default_input_image_details(value: Any) -> Any:
-    """Set the API default for input images before SDK type validation."""
+def _normalize_input_images(value: Any) -> Any:
+    """Normalize input images before SDK type validation.
+
+    Sets the API default ``detail`` and accepts the chat-completions shapes
+    (``image_url`` part type, nested ``image_url: {"url": ...}``).
+    """
     if not isinstance(value, dict):
         return value
 
@@ -149,13 +153,13 @@ def _default_input_image_details(value: Any) -> Any:
     changed = False
     for part in content:
         new_part = part
-        if (
-            isinstance(part, dict)
-            and part.get("type") == "input_image"
-            and "detail" not in part
-        ):
-            new_part = {**part, "detail": "auto"}
-            changed = True
+        if isinstance(part, dict) and part.get("type") in ("input_image", "image_url"):
+            new_part = {**part, "type": "input_image"}
+            image_url = new_part.get("image_url")
+            if isinstance(image_url, dict) and "url" in image_url:
+                new_part["image_url"] = image_url["url"]
+            new_part.setdefault("detail", "auto")
+            changed = changed or new_part != part
         new_content.append(new_part)
 
     return {**value, "content": new_content} if changed else value
@@ -183,7 +187,7 @@ class ResponsesRequest(OpenAIBaseModel):
         | list[
             Annotated[
                 ResponseInputOutputItem,
-                BeforeValidator(_default_input_image_details),
+                BeforeValidator(_normalize_input_images),
             ]
         ]
     )