diff --git a/tests/integ/test_tools.py b/tests/integ/test_tools.py
index 9bd195eb..8c01f78d 100644
--- a/tests/integ/test_tools.py
+++ b/tests/integ/test_tools.py
@@ -124,7 +124,7 @@ def test_florence2_phrase_grounding_video():
         frames=frames,
     )
     assert len(result) == 10
-    assert 24 <= len([res["label"] for res in result[0]]) <= 26
+    assert 2 <= len([res["label"] for res in result[0]]) <= 26
 
 
 def test_florence2_phrase_grounding_video_fine_tune_id():
@@ -138,7 +138,7 @@ def test_florence2_phrase_grounding_video_fine_tune_id():
         fine_tune_id=FINE_TUNE_ID,
     )
     assert len(result) == 10
-    assert 24 <= len([res["label"] for res in result[0]]) <= 26
+    assert 16 <= len([res["label"] for res in result[0]]) <= 26
 
 
 def test_template_match():
diff --git a/vision_agent/tools/tool_utils.py b/vision_agent/tools/tool_utils.py
index 772d6bc3..924b96e6 100644
--- a/vision_agent/tools/tool_utils.py
+++ b/vision_agent/tools/tool_utils.py
@@ -37,6 +37,7 @@ def send_inference_request(
     files: Optional[List[Tuple[Any, ...]]] = None,
     v2: bool = False,
     metadata_payload: Optional[Dict[str, Any]] = None,
+    is_form: bool = False,
 ) -> Any:
     # TODO: runtime_tag and function_name should be metadata_payload and not included
     # in the service payload
@@ -64,7 +65,7 @@ def send_inference_request(
     elif metadata_payload is not None and "function_name" in metadata_payload:
         function_name = metadata_payload["function_name"]
 
-    response = _call_post(url, payload, session, files, function_name)
+    response = _call_post(url, payload, session, files, function_name, is_form)
 
     # TODO: consider making the response schema the same between below two sources
     return response if "TOOL_ENDPOINT_AUTH" in os.environ else response["data"]
@@ -75,6 +76,7 @@ def send_task_inference_request(
     task_name: str,
     files: Optional[List[Tuple[Any, ...]]] = None,
     metadata: Optional[Dict[str, Any]] = None,
+    is_form: bool = False,
 ) -> Any:
     url = f"{_LND_API_URL_v2}/{task_name}"
     headers = {"apikey": _LND_API_KEY}
@@ -87,7 +89,7 @@ def send_task_inference_request(
     function_name = "unknown"
     if metadata is not None and "function_name" in metadata:
         function_name = metadata["function_name"]
-    response = _call_post(url, payload, session, files, function_name)
+    response = _call_post(url, payload, session, files, function_name, is_form)
     return response["data"]
 
 
@@ -203,6 +205,7 @@ def _call_post(
     session: Session,
     files: Optional[List[Tuple[Any, ...]]] = None,
     function_name: str = "unknown",
+    is_form: bool = False,
 ) -> Any:
     files_in_b64 = None
     if files:
@@ -210,6 +213,8 @@ def _call_post(
     try:
         if files is not None:
             response = session.post(url, data=payload, files=files)
+        elif is_form:
+            response = session.post(url, data=payload)
         else:
             response = session.post(url, json=payload)
 
diff --git a/vision_agent/tools/tools.py b/vision_agent/tools/tools.py
index 7faa123a..d99a38fc 100644
--- a/vision_agent/tools/tools.py
+++ b/vision_agent/tools/tools.py
@@ -201,6 +201,7 @@ def owl_v2_image(
             data,
             "florence2-ft",
             v2=True,
+            is_form=True,
             metadata_payload={"function_name": "owl_v2_image"},
         )
         # get the first frame
@@ -432,6 +433,7 @@ def florence2_sam2_image(
             req_data,
             "florence2-ft",
             v2=True,
+            is_form=True,
             metadata_payload={"function_name": "florence2_sam2_image"},
         )
         # get the first frame
@@ -1193,6 +1195,7 @@ def florence2_phrase_grounding_image(
             data,
             "florence2-ft",
             v2=True,
+            is_form=True,
             metadata_payload={"function_name": "florence2_phrase_grounding_image"},
         )
         # get the first frame
@@ -1268,18 +1271,14 @@ def florence2_phrase_grounding_video(
             )
 
         data_obj = Florence2FtRequest(
-            video=buffer_bytes,
             task=PromptTask.PHRASE_GROUNDING,
             prompt=prompt,
             job_id=UUID(fine_tune_id),
         )
-        data = data_obj.model_dump(by_alias=True, exclude_none=True)
     else:
-        data_obj = Florence2FtRequest(
-            video=buffer_bytes, task=PromptTask.PHRASE_GROUNDING, prompt=prompt
-        )
-        data = data_obj.model_dump(by_alias=True, exclude_none=True)
+        data_obj = Florence2FtRequest(task=PromptTask.PHRASE_GROUNDING, prompt=prompt)
 
+    data = data_obj.model_dump(by_alias=True, exclude_none=True, mode="json")
     detections = send_inference_request(
         data,
         "florence2-ft",