fix(proxy): parse form-encoded video edit/extension bodies after auth

Fixes #36487

video_edit, video_extension, and video_remix called request.body() after
user_api_key_auth had already parsed multipart/form bodies via
_read_request_body(), causing RuntimeError Stream consumed and 500s for
OpenAI SDK clients.

Use _read_request_body consistently and normalize bare-string or JSON-string
video references from form fields into video_id.

litellm/proxy/video_endpoints/endpoints.py

@@ -2,7 +2,6 @@
 
 from typing import Any, Final
 
-import orjson
 from fastapi import APIRouter, Depends, File, Form, Request, Response, UploadFile
 from fastapi.responses import ORJSONResponse
 
@@ -20,6 +19,7 @@ from litellm.proxy.video_endpoints.utils import (
     encode_character_id_in_response,
     extract_model_from_target_model_names,
     get_custom_provider_from_data,
+    pop_video_reference_to_video_id,
 )
 from litellm.types.videos.utils import (
     decode_character_id_with_provider,
@@ -451,9 +451,7 @@ async def video_remix(
         version,
     )
 
-    # Read request body
-    body: Final = await request.body()
-    data: Final = orjson.loads(body)
+    data: Final = await _read_request_body(request=request)
     data["video_id"] = video_id
 
     decoded: Final = decode_video_id_with_provider(video_id)
@@ -760,15 +758,10 @@ async def video_edit(
         version,
     )
 
-    body: Final = await request.body()
-    data: Final = orjson.loads(body)
+    data: Final = await _read_request_body(request=request)
+    pop_video_reference_to_video_id(data)
 
-    # Extract video_id from nested video object
-    video_ref: Final = data.pop("video", {})
-    video_id: Final = video_ref.get("id", "") if isinstance(video_ref, dict) else ""
-    data["video_id"] = video_id
-
-    decoded: Final = decode_video_id_with_provider(video_id)
+    decoded: Final = decode_video_id_with_provider(data["video_id"])
     provider_from_id: Final = decoded.get("custom_llm_provider")
     model_id_from_decoded: Final = decoded.get("model_id")
 
@@ -860,15 +853,10 @@ async def video_extension(
         version,
     )
 
-    body: Final = await request.body()
-    data: Final = orjson.loads(body)
+    data: Final = await _read_request_body(request=request)
+    pop_video_reference_to_video_id(data)
 
-    # Extract video_id from nested video object
-    video_ref: Final = data.pop("video", {})
-    video_id: Final = video_ref.get("id", "") if isinstance(video_ref, dict) else ""
-    data["video_id"] = video_id
-
-    decoded: Final = decode_video_id_with_provider(video_id)
+    decoded: Final = decode_video_id_with_provider(data["video_id"])
     provider_from_id: Final = decoded.get("custom_llm_provider")
     model_id_from_decoded: Final = decoded.get("model_id")
 

litellm/proxy/video_endpoints/utils.py

@@ -13,6 +13,27 @@ def extract_model_from_target_model_names(target_model_names: Any) -> str | None
     return target_model_names[0] if target_model_names else None
 
 
+def pop_video_reference_to_video_id(data: dict[str, Any]) -> None:
+    """
+    Normalize OpenAI video edit/extension payloads into ``video_id``.
+
+    JSON bodies use ``video: {"id": ...}``. Multipart and form-urlencoded bodies
+    may send a bare id string or a JSON-encoded reference object as a string field.
+    """
+    video_ref: Final = data.pop("video", {})
+    if isinstance(video_ref, dict):
+        video_id: Final = video_ref.get("id", "")
+    elif isinstance(video_ref, str):
+        try:
+            parsed_ref: Final = orjson.loads(video_ref)
+        except orjson.JSONDecodeError:
+            parsed_ref = None
+        video_id = parsed_ref.get("id", "") if isinstance(parsed_ref, dict) else video_ref
+    else:
+        video_id = ""
+    data["video_id"] = video_id
+
+
 def get_custom_provider_from_data(data: dict[str, Any]) -> str | None:
     custom_llm_provider: Final = data.get("custom_llm_provider")
     if custom_llm_provider:

tests/test_litellm/proxy/video_endpoints/test_endpoints.py

@@ -375,6 +375,7 @@ async def test_content__model_encoded_id(harness):
 async def call_edit(
     harness: Harness, *, body: Dict[str, Any], headers=None, query=None
 ):
+    harness.read_body.return_value = dict(body)
     return await endpoints.video_edit(
         request=FakeRequest(headers=headers, query=query, raw_body=orjson.dumps(body)),
         fastapi_response=Response(),
@@ -431,6 +432,27 @@ async def test_edit__missing_video_object_defaults_to_openai(harness):
     assert "video" not in data
 
 
+@pytest.mark.asyncio
+async def test_edit__bare_string_video_id_from_form_field(harness):
+    await call_edit(harness, body={"prompt": "brighter", "video": "video_plain"})
+
+    assert harness.processor_data() == {
+        "prompt": "brighter",
+        "video_id": "video_plain",
+        "custom_llm_provider": "openai",
+    }
+
+
+@pytest.mark.asyncio
+async def test_edit__json_string_video_reference_from_form_field(harness):
+    await call_edit(
+        harness,
+        body={"prompt": "brighter", "video": orjson.dumps({"id": "video_plain"}).decode()},
+    )
+
+    assert harness.processor_data()["video_id"] == "video_plain"
+
+
 # =========================================================================== #
 #   GET /v1/videos  -  video_list                                              #
 # =========================================================================== #
@@ -474,6 +496,7 @@ async def test_list__provider_from_header(harness):
 async def call_remix(
     harness: Harness, video_id: str, *, body, headers=None, query=None
 ):
+    harness.read_body.return_value = dict(body)
     return await endpoints.video_remix(
         video_id=video_id,
         request=FakeRequest(headers=headers, query=query, raw_body=orjson.dumps(body)),
@@ -632,6 +655,7 @@ async def test_get_character__plain_id_defaults_openai_no_encode(harness):
 
 
 async def call_extension(harness: Harness, *, body, headers=None, query=None):
+    harness.read_body.return_value = dict(body)
     return await endpoints.video_extension(
         request=FakeRequest(headers=headers, query=query, raw_body=orjson.dumps(body)),
         fastapi_response=Response(),

tests/test_litellm/test_video_generation.py

@@ -2321,6 +2321,53 @@ def test_edit_and_extension_support_custom_provider_from_extra_body(
     assert captured_data["custom_llm_provider"] == "vertex_ai"
 
 
+@pytest.mark.parametrize("endpoint", ["/v1/videos/edits", "/v1/videos/extensions"])
+def test_edit_and_extension_accept_form_encoded_after_auth_reads_body(
+    video_proxy_test_client, endpoint
+):
+    from fastapi import Request
+    from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing
+    from litellm.proxy.common_utils.http_parsing_utils import _read_request_body
+    from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
+
+    captured_data = {}
+
+    async def _mock_base_process(self, **kwargs):
+        captured_data.update(self.data)
+        return {
+            "id": "video_resp_123",
+            "object": "video",
+            "status": "queued",
+            "created_at": 1712697600,
+        }
+
+    async def auth_that_reads_body_first(request: Request):
+        await _read_request_body(request=request)
+        return MagicMock()
+
+    app = video_proxy_test_client.app
+    app.dependency_overrides[user_api_key_auth] = auth_that_reads_body_first
+
+    with patch.object(
+        ProxyBaseLLMRequestProcessing,
+        "base_process_llm_request",
+        new=_mock_base_process,
+    ):
+        response = video_proxy_test_client.post(
+            endpoint,
+            headers={"Authorization": "Bearer sk-1234"},
+            data={
+                "model": "my-video-model",
+                "prompt": "brighter",
+                "video": "video_123",
+            },
+        )
+
+    assert response.status_code == 200, response.text
+    assert captured_data["video_id"] == "video_123"
+    assert captured_data["prompt"] == "brighter"
+
+
 @pytest.mark.parametrize("endpoint", ["/v1/videos/edits", "/v1/videos/extensions"])
 def test_edit_and_extension_route_with_encoded_video_ids(
     video_proxy_test_client, endpoint