diff --git a/src/_bentoml_sdk/io_models.py b/src/_bentoml_sdk/io_models.py
index 50f898fa9a9..3bacfc86b7d 100644
--- a/src/_bentoml_sdk/io_models.py
+++ b/src/_bentoml_sdk/io_models.py
@@ -432,7 +432,8 @@ def from_output(cls, func: t.Callable[..., t.Any]) -> type[IODescriptor]:
             )
         media_type: str | None = None
         if is_iterator_type(return_annotation):
-            return_annotation = get_args(return_annotation)[0]
+            iterator_args = get_args(return_annotation)
+            return_annotation = iterator_args[0] if iterator_args else t.Any
         elif is_annotated(return_annotation):
             content_type = next(
                 (a for a in get_args(return_annotation) if isinstance(a, ContentType)),
diff --git a/tests/unit/_bentoml_sdk/test_io_models.py b/tests/unit/_bentoml_sdk/test_io_models.py
new file mode 100644
index 00000000000..75ff3405ef2
--- /dev/null
+++ b/tests/unit/_bentoml_sdk/test_io_models.py
@@ -0,0 +1,32 @@
+from __future__ import annotations
+
+import collections.abc as cabc
+import typing as t
+
+import pytest
+
+from _bentoml_sdk.io_models import IODescriptor
+
+
+@pytest.mark.parametrize(
+    "return_annotation",
+    [
+        t.Iterator,
+        t.Generator,
+        t.AsyncIterator,
+        t.AsyncGenerator,
+        cabc.Iterator,
+        cabc.Generator,
+        cabc.AsyncIterator,
+        cabc.AsyncGenerator,
+    ],
+)
+def test_from_output_accepts_bare_iterators(return_annotation: t.Any) -> None:
+    def stream() -> t.Iterator[t.Any]:
+        yield None
+
+    stream.__annotations__["return"] = return_annotation
+
+    descriptor = IODescriptor.from_output(stream)
+
+    assert descriptor.model_fields["root"].annotation is t.Any