diff --git a/openviking/storage/queuefs/semantic_processor.py b/openviking/storage/queuefs/semantic_processor.py
index 7e300666..2dcd7c2d 100644
--- a/openviking/storage/queuefs/semantic_processor.py
+++ b/openviking/storage/queuefs/semantic_processor.py
@@ -187,9 +187,42 @@ async def _generate_single_file_summary(self, file_path: str) -> Dict[str, str]:
 
         try:
             # Read file content (limit length)
-            content = await viking_fs.read_file(file_path)
-            if isinstance(content, bytes):
-                content = content.decode("utf-8")
+            content_bytes = await viking_fs.read_file(file_path)
+            content = ""
+
+            # Check if it's binary data
+            if isinstance(content_bytes, bytes):
+                try:
+                    content = content_bytes.decode("utf-8")
+                except UnicodeDecodeError:
+                    # It's a binary file (likely image, video, etc.)
+                    # Use VLM to describe it if available
+                    if vlm.is_available():
+                        # Determine prompt based on file type
+                        prompt = "Please describe this file content briefly for summary."
+                        if file_name.lower().endswith((".png", ".jpg", ".jpeg", ".gif", ".webp")):
+                            prompt = "Describe this image in detail for a file summary."
+
+                        try:
+                            # Use the new vision capabilities we added!
+                            # Check if get_vision_completion_async exists (it should now)
+                            if hasattr(vlm.get_vlm_instance(), "get_vision_completion_async"):
+                                summary = await vlm.get_vlm_instance().get_vision_completion_async(
+                                    prompt, [content_bytes]
+                                )
+                                return {"name": file_name, "summary": summary.strip()}
+                            else:
+                                logger.warning("VLM instance does not support vision completion")
+                                return {"name": file_name, "summary": f"[Binary file: {file_name}]"}
+                        except Exception as vlm_err:
+                            logger.warning(
+                                f"VLM vision completion failed for {file_name}: {vlm_err}"
+                            )
+                            return {"name": file_name, "summary": f"[Binary file: {file_name}]"}
+                    else:
+                        return {"name": file_name, "summary": f"[Binary file: {file_name}]"}
+            else:
+                content = content_bytes
 
             # Limit content length (about 10000 tokens)
             max_chars = 30000
diff --git a/tests/storage/test_semantic_processor_vlm.py b/tests/storage/test_semantic_processor_vlm.py
new file mode 100644
index 00000000..a1869521
--- /dev/null
+++ b/tests/storage/test_semantic_processor_vlm.py
@@ -0,0 +1,70 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from openviking.storage.queuefs.semantic_processor import SemanticProcessor
+
+
+@pytest.mark.asyncio
+async def test_semantic_processor_image_description():
+    """Test that image files trigger VLM vision capabilities"""
+    processor = SemanticProcessor()
+
+    # Mock VikingFS to return bytes that are not valid utf-8 (simulating image binary)
+    mock_fs = MagicMock()
+    mock_fs.read_file = AsyncMock(return_value=b"\xff\xff\xff\xff")
+
+    # Mock VLM Config
+    mock_vlm_config = MagicMock()
+    mock_vlm_config.is_available.return_value = True
+
+    # Mock VLM Instance with vision capability
+    mock_vlm_instance = MagicMock()
+    mock_vlm_instance.get_vision_completion_async = AsyncMock(
+        return_value="A detailed description of the image."
+    )
+    mock_vlm_config.get_vlm_instance.return_value = mock_vlm_instance
+
+    with (
+        patch("openviking.storage.queuefs.semantic_processor.get_viking_fs", return_value=mock_fs),
+        patch(
+            "openviking.storage.queuefs.semantic_processor.get_openviking_config"
+        ) as mock_get_config,
+    ):
+        mock_get_config.return_value.vlm = mock_vlm_config
+
+        # Test with png extension to trigger vision path
+        result = await processor._generate_single_file_summary("test_image.png")
+
+        assert result["name"] == "test_image.png"
+        assert result["summary"] == "A detailed description of the image."
+
+        # Verify vision method was called
+        mock_vlm_instance.get_vision_completion_async.assert_called_once()
+        # Verify the prompt is vision-specific
+        args = mock_vlm_instance.get_vision_completion_async.call_args
+        assert "image" in args[0][0].lower()
+
+
+@pytest.mark.asyncio
+async def test_semantic_processor_binary_file_fallback():
+    """Test fallback when VLM is not available for binary files"""
+    processor = SemanticProcessor()
+
+    mock_fs = MagicMock()
+    mock_fs.read_file = AsyncMock(return_value=b"\xff\xff")
+
+    mock_vlm_config = MagicMock()
+    mock_vlm_config.is_available.return_value = False
+
+    with (
+        patch("openviking.storage.queuefs.semantic_processor.get_viking_fs", return_value=mock_fs),
+        patch(
+            "openviking.storage.queuefs.semantic_processor.get_openviking_config"
+        ) as mock_get_config,
+    ):
+        mock_get_config.return_value.vlm = mock_vlm_config
+
+        result = await processor._generate_single_file_summary("unknown_file.bin")
+
+        assert "[Binary file: unknown_file.bin]" in result["summary"]