diff --git a/litellm/llms/vertex_ai/gemini/transformation.py b/litellm/llms/vertex_ai/gemini/transformation.py
index d49a16fa90..5eaac5e48c 100644
--- a/litellm/llms/vertex_ai/gemini/transformation.py
+++ b/litellm/llms/vertex_ai/gemini/transformation.py
@@ -139,14 +139,20 @@ def _apply_gemini_metadata(
 ) -> PartType:
     """
     Apply media_resolution and video_metadata parameters to a Gemini part.
-    Both are supported across all Gemini models (1.x, 2.x, 3+).
+
+    - Per-part media_resolution: Gemini 3+ only (2.x uses generation_config global).
+    - video_metadata (fps, startOffset, endOffset): all Gemini models (1.x, 2.x, 3+).
     """
     if model is None:
         return part
 
+    from .vertex_and_google_ai_studio_gemini import VertexGeminiConfig
+
     part_dict = dict(part)
 
-    if media_resolution_enum is not None:
+    if media_resolution_enum is not None and VertexGeminiConfig._is_gemini_3_or_newer(
+        model
+    ):
         part_dict["media_resolution"] = media_resolution_enum
 
     if video_metadata is not None:
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
index f77bba56cb..1ea7486a51 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
@@ -3511,8 +3511,18 @@ def test_video_metadata_supported_for_all_gemini_models():
         assert file_part is not None, f"{model}: file part should exist"
         assert "video_metadata" in file_part, f"{model}: video_metadata should be present"
         assert file_part["video_metadata"]["fps"] == 5, f"{model}: fps should be 5"
+
+    # Per-part media_resolution is Gemini 3+ only; 2.x uses generation_config global
+    for model in ["gemini-3-pro-preview"]:
+        contents = _gemini_convert_messages_with_history(messages=messages, model=model)
+        file_part = next(p for p in contents[0]["parts"] if "file_data" in p)
         assert "media_resolution" in file_part, f"{model}: media_resolution should be present"
 
+    for model in ["gemini-1.5-pro", "gemini-2.5-flash", "gemini-2.5-pro"]:
+        contents = _gemini_convert_messages_with_history(messages=messages, model=model)
+        file_part = next(p for p in contents[0]["parts"] if "file_data" in p)
+        assert "media_resolution" not in file_part, f"{model}: per-part media_resolution should not be set"
+
 
 def test_chunk_parser_handles_prompt_feedback_block():
     """Test chunk_parser correctly handles promptFeedback.blockReason"""