diff --git a/test/test_video.py b/test/test_video.py
index 63434fa9c1f..7dc6b8aeb93 100644
--- a/test/test_video.py
+++ b/test/test_video.py
@@ -10,7 +10,7 @@
 
 import torch
 import torchvision
-from torchvision.io import _HAS_VIDEO_OPT
+from torchvision.io import _HAS_VIDEO_OPT, Video
 
 try:
     import av
@@ -289,7 +289,7 @@ def test_read_video_tensor(self):
             tv_result, _, _ = torchvision.io.read_video(full_path, pts_unit="sec")
             tv_result = tv_result.permute(0, 3, 1, 2)
             # pass 2: decode all frames using new api
-            reader = torch.classes.torchvision.Video(full_path, "video")
+            reader = Video(full_path, "video")
             frames = []
             t, _ = reader.next()
             while t.numel() > 0:
@@ -310,7 +310,7 @@ def test_read_video_tensor(self):
     #         s = min(r)
     #         e = max(r)
 
-    #         reader = torch.classes.torchvision.Video(full_path, "video")
+    #         reader = Video(full_path, "video")
     #         results = _template_read_video(reader, s, e)
     #         tv_video, tv_audio, info = torchvision.io.read_video(
     #             full_path, start_pts=s, end_pts=e, pts_unit="sec"
@@ -329,7 +329,7 @@ def test_read_video_tensor(self):
     #             full_path, pts_unit="sec"
     #         )
     #         # pass 2: decode all frames using new api
-    #         reader = torch.classes.torchvision.Video(full_path, "video")
+    #         reader = Video(full_path, "video")
     #         pts = []
     #         t, p = reader.next()
     #         while t.numel() > 0:
@@ -353,7 +353,7 @@ def test_metadata(self):
         torchvision.set_video_backend("pyav")
         for test_video, config in test_videos.items():
             full_path = os.path.join(VIDEO_DIR, test_video)
-            reader = torch.classes.torchvision.Video(full_path, "video")
+            reader = Video(full_path, "video")
             reader_md = reader.get_metadata()
             self.assertAlmostEqual(
                 config.video_fps, reader_md["video"]["fps"][0], delta=0.0001
@@ -372,7 +372,7 @@ def test_video_reading_fn(self):
 
             ref_result = _decode_frames_by_av_module(full_path)
 
-            reader = torch.classes.torchvision.Video(full_path, "video")
+            reader = Video(full_path, "video")
             newapi_result = _template_read_video(reader)
 
             # First we check if the frames are approximately the same
diff --git a/torchvision/io/__init__.py b/torchvision/io/__init__.py
index 029dd311592..7a1026541c9 100644
--- a/torchvision/io/__init__.py
+++ b/torchvision/io/__init__.py
@@ -1,3 +1,6 @@
+import torch
+import warnings
+
 from ._video_opt import (
     Timebase,
     VideoMetaData,
@@ -20,9 +23,14 @@
     encode_jpeg,
     write_jpeg,
     encode_png,
-    write_png
+    write_png,
 )
 
+if _HAS_VIDEO_OPT:
+    Video = torch.classes.torchvision.Video
+else:
+    Video = None
+
 
 __all__ = [
     "write_video",
@@ -39,10 +47,11 @@
     "_read_video_meta_data",
     "VideoMetaData",
     "Timebase",
-    'read_image',
-    'decode_image',
-    'encode_jpeg',
-    'write_jpeg',
-    'encode_png',
-    'write_png',
+    "read_image",
+    "decode_image",
+    "encode_jpeg",
+    "write_jpeg",
+    "encode_png",
+    "write_png",
+    "Video",
 ]
diff --git a/torchvision/io/_video_opt.py b/torchvision/io/_video_opt.py
index ae4b0f7c869..e0a5fbd4499 100644
--- a/torchvision/io/_video_opt.py
+++ b/torchvision/io/_video_opt.py
@@ -26,7 +26,9 @@
         torch.ops.load_library(ext_specs.origin)
         _HAS_VIDEO_OPT = True
 except (ImportError, OSError):
-    pass
+    warnings.warn("Could not import video_reader bakcend."
+                  + " Make sure ffmpeg is installed properly"
+                  + " and rebuild torchvision.")
 
 
 default_timebase = Fraction(0, 1)