[Bugfix] Accept RFC 2397 parameters in base64 data URLs (#48973)

Signed-off-by: Thomas Fahrner <thomas.fahrner@parasail.io>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

tests/multimodal/media/test_connector.py

@@ -225,6 +225,44 @@ async def test_fetch_image_local_files_with_space_in_name(image_url: str):
         assert not ImageChops.difference(image_sync, image_async).getbbox()
 
 
+@pytest.mark.asyncio
+async def test_fetch_image_data_url_with_params():
+    """RFC 2397 allows parameters between the mediatype and the base64
+    marker; they must not be rejected or leak into the media type."""
+    connector = MediaConnector()
+
+    image = Image.new("RGB", (4, 4), color=(255, 0, 0))
+    with NamedTemporaryFile(suffix=".png") as f:
+        image.save(f.name)
+        base64_image = base64.b64encode(f.read()).decode("utf-8")
+
+    data_url = f"data:image/png;charset=utf-8;base64,{base64_image}"
+    image_sync = connector.fetch_image(data_url)
+    image_async = await connector.fetch_image_async(data_url)
+    assert _image_equals(image_sync, image_async)
+
+
+def test_fetch_image_data_url_malformed():
+    connector = MediaConnector()
+
+    with pytest.raises(ValueError, match="missing ','"):
+        connector.fetch_image("data:image/png;base64")
+
+    with pytest.raises(NotImplementedError, match="base64"):
+        connector.fetch_image("data:text/plain,hello")
+
+    # ";base64" requires the ";"; here "base64" is a (bogus) media type.
+    with pytest.raises(NotImplementedError, match="base64"):
+        connector.fetch_image("data:base64,aGVsbG8=")
+
+    # Strict RFC 2397 grammar: lowercase "base64", no whitespace.
+    with pytest.raises(NotImplementedError, match="base64"):
+        connector.fetch_image("data:image/png;BASE64,aGVsbG8=")
+
+    with pytest.raises(NotImplementedError, match="base64"):
+        connector.fetch_image("data:image/png; base64,aGVsbG8=")
+
+
 @pytest.mark.asyncio
 async def test_fetch_image_error_conversion():
     connector = MediaConnector()

vllm/multimodal/media/connector.py

@@ -302,14 +302,18 @@ class MediaConnector:
         media_io: MediaIO[_M],
     ) -> _M:  # type: ignore[type-var]
         # Format per RFC 2397:
-        # data:[<mediatype>][;base64],<data>
-        data_spec, data = url[5:].split(",", 1)
-        media_type, data_type = data_spec.split(";", 1)
+        # data:[<mediatype>][;<param>=<value>]*[;base64],<data>
+        data_spec, sep, data = url[5:].partition(",")
+        if not sep:
+            msg = f"Invalid data URL {url[:32]!r}: missing ',' separator."
+            raise ValueError(msg)
 
-        if data_type != "base64":
+        media_type, sep, encoding = data_spec.rpartition(";")
+        if not sep or encoding != "base64":
             msg = "Only base64 data URLs are supported for now."
             raise NotImplementedError(msg)
 
+        media_type = media_type.partition(";")[0]
         return media_io.load_base64(media_type, data)
 
     def _load_file_url(