diff --git a/pyproject.toml b/pyproject.toml
index 850b06b..dc20fb5 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -26,7 +26,7 @@ dependencies = [
 TORCH = ["torch", "torchvision"]
 VIDEO = ["av>=10.0.0", "imageio[ffmpeg]>=2.23.0"]
 PROC = ["joblib>=1.2.0"]
-DEV = ["pytest", "pytest-cov", "lovely-tensors>=0.1.14", "lovely-numpy>=0.2.8"]
+DEV = ["pytest", "pytest-cov", "lovely-tensors>=0.1.14", "lovely-numpy>=0.2.8", "opencv-python"]
 ALL = ["image_utils[TORCH,VIDEO,PROC,DEV]"]
 
 [project.urls]
diff --git a/src/image_utils/im.py b/src/image_utils/im.py
index 8a94654..57ac707 100644
--- a/src/image_utils/im.py
+++ b/src/image_utils/im.py
@@ -11,21 +11,22 @@
 from io import BytesIO
 from math import ceil
 from pathlib import Path
-from typing import TYPE_CHECKING, Callable, Iterable, Optional, Tuple, Type, Union
+from typing import TYPE_CHECKING, Callable, Iterable, Optional, Tuple, Type, Union, Any
 
 import numpy as np
 from einops import rearrange, repeat
 from jaxtyping import Bool, Float, Integer
 from numpy import ndarray
 from PIL import Image
-from torch import rand
+
 
 if importlib.util.find_spec("torch") is not None:
     import torch
-    from torch import Tensor
+    from torch import Tensor, device # type: ignore
 else:
-    class Tensor:
-        pass
+    class device:
+        def __init__(self, type: str):
+            self.type = type
 
 if importlib.util.find_spec("image_utils") is not None:
     from image_utils.file_utils import get_date_time_str, load_cached_from_url
@@ -35,6 +36,7 @@ class Tensor:
     from imageio import v3 as iio
 
 if TYPE_CHECKING:
+    from torch import Tensor
     ImArr = Union[ndarray, Tensor]  # The actual array itself
     ImArrType = Type[Union[ndarray, Tensor]]  # The object itself is just a type
     ImDtype = Union[torch.dtype, np.dtype]
@@ -126,7 +128,7 @@ def __init__(self, arr: Union["Im", Tensor, Image.Image, ndarray, str, Path], ch
                     setattr(self, attr, getattr(arr, attr))
             return
 
-        self.device: torch.device
+        self.device: device
         self.arr_type: ImArrType
 
         # To handle things in a unified manner, we choose to always convert PIL Images -> NumPy internally
@@ -137,7 +139,7 @@ def __init__(self, arr: Union["Im", Tensor, Image.Image, ndarray, str, Path], ch
         self.arr: ImArr = arr
         if isinstance(self.arr, ndarray):
             self.arr_type = ndarray
-            self.device = "cpu"
+            self.device = device("cpu")
         elif isinstance(self.arr, Tensor):
             self.device = self.arr.device
             self.arr_type = Tensor
@@ -154,6 +156,7 @@ def __init__(self, arr: Union["Im", Tensor, Image.Image, ndarray, str, Path], ch
 
         # We normalize all arrays to (B, H, W, C) and record their original shape so
         # we can re-transform then when we need to output them
+        self.arr_transform: Callable[[ImArr], ImArr]
         if len(self.arr.shape) == 2:
             self.channel_order = ChannelOrder.HWC
             self.arr_transform = partial(rearrange, pattern="() h w () -> h w")
@@ -165,7 +168,7 @@ def __init__(self, arr: Union["Im", Tensor, Image.Image, ndarray, str, Path], ch
             extra_dims = self.arr.shape[:-3]
             mapping = {k: v for k, v in zip(string.ascii_uppercase, extra_dims)}
             transform_str = f'({" ".join(sorted(list(mapping.keys())))}) a b c -> {" ".join(sorted(list(mapping.keys())))} a b c'
-            self.arr_transform = partial(rearrange, pattern=transform_str, **mapping)  # lambda x: rearrange(x, transform_str, g)
+            self.arr_transform = partial(rearrange, pattern=transform_str, **mapping)  # type: ignore # lambda x: rearrange(x, transform_str, g)
         else:
             raise ValueError("Must be between 3-5 dims")
 
@@ -239,7 +242,7 @@ def wrapper(self: Im, *args, **kwargs):
 
         return custom_decorator
 
-    def _handle_order_transform(self, im, desired_order: ChannelOrder, desired_range: ChannelRange, select_batch=None):
+    def _handle_order_transform(self, im: ImArr, desired_order: ChannelOrder, desired_range: ChannelRange, select_batch=None) -> ImArr:
         if select_batch is not None:
             im = im[select_batch]
         else:
@@ -278,9 +281,9 @@ def _handle_order_transform(self, im, desired_order: ChannelOrder, desired_range
                 raise ValueError("Not supported")
 
             if desired_range == ChannelRange.UINT8:
-                im = im.astype(np.uint8) if is_ndarray(im) else im.to(torch.uint8)
+                im = im.astype(np.uint8) if isinstance(im, ndarray) else im.to(torch.uint8)
             elif desired_range == ChannelRange.FLOAT:
-                im = im.astype(np.float32) if is_ndarray(im) else im.to(torch.float32)
+                im = im.astype(np.float32) if isinstance(im, ndarray) else im.to(torch.float32)
 
         return im
 
@@ -290,6 +293,7 @@ def get_np(self, order=ChannelOrder.HWC, range=ChannelRange.UINT8) -> ndarray:
             arr = torch_to_numpy(arr)  # type: ignore
 
         arr = self._handle_order_transform(arr, order, range)
+        assert isinstance(arr, ndarray)
         return arr
 
     def get_torch(self, order=ChannelOrder.CHW, range=ChannelRange.FLOAT) -> Tensor:
@@ -298,6 +302,7 @@ def get_torch(self, order=ChannelOrder.CHW, range=ChannelRange.FLOAT) -> Tensor:
             arr = torch.from_numpy(arr)
 
         arr = self._handle_order_transform(arr, order, range)
+        assert isinstance(arr, Tensor)
         if self.device is not None:
             arr = arr.to(self.device)
         return arr
@@ -376,8 +381,8 @@ def random(h: int = 1080, w: int = 1920) -> Im:
 
     @_convert_to_datatype(desired_datatype=Tensor, desired_order=ChannelOrder.CHW, desired_range=ChannelRange.FLOAT)
     def resize(self, height: int, width: int, resampling_mode: str = "bilinear"):
-        assert isinstance(self.arr, Tensor)
         from torchvision.transforms.functional import resize, InterpolationMode
+        assert isinstance(self.arr, torch.Tensor)
         arr = resize(self.arr, [height, width], InterpolationMode(resampling_mode), antialias=True)
         arr = self.arr_transform(arr)
         return Im(arr)
@@ -429,7 +434,7 @@ def save(
         if filepath is None:
             filepath = Path(get_date_time_str())
 
-        filepath = Im._save_data(filepath, filetype)
+        filepath: Path = Im._save_data(filepath, filetype)
 
         if self.batch_size > 1:
             img = self.get_torch()
@@ -538,7 +543,7 @@ def normalize(self, normalize_min_max: bool = False, **kwargs) -> Im:
             self.arr = (self.arr - mean) / std
         return self
 
-    def denormalize(self, clamp: Union[bool, tuple[float, float]] = (0, 1.0), **kwargs) -> Im:
+    def denormalize(self, clamp: tuple[float, float] = (0, 1.0), **kwargs) -> Im:
         self, mean, std = self.normalize_setup(**kwargs)
         self.arr = (self.arr * std) + mean
         if isinstance(self.arr, ndarray):
@@ -595,10 +600,10 @@ def encode_video(self, fps: int, format="mp4") -> BytesIO:
             import imageio
             if format == "webm":
                 writer = imageio.get_writer(
-                    ntp.name, format="webm", codec="libvpx-vp9", pixelformat="yuv420p", output_params=["-lossless", "1"], fps=fps
+                    ntp.name, format="webm", codec="libvpx-vp9", pixelformat="yuv420p", output_params=["-lossless", "1"], fps=fps # type: ignore
                 )
             elif format == "gif":
-                writer = imageio.get_writer(ntp.name, format="GIF", mode="I", duration=(1000 * 1 / fps))
+                writer = imageio.get_writer(ntp.name, format="GIF", mode="I", duration=(1000 * 1 / fps)) # type: ignore
             elif format == "mp4":
                 writer = imageio.get_writer(ntp.name, quality=10, pixelformat="yuv420p", codec="libx264", fps=fps)
             else:
@@ -728,7 +733,7 @@ def get_arr_hwc(im: Im):
     return im._handle_order_transform(im.arr, desired_order=ChannelOrder.HWC, desired_range=im.channel_range)
 
 
-def new_like(arr, shape, fill: Optional[tuple[int]] = None):
+def new_like(arr, shape, fill: Optional[tuple[int]] = None) -> ImArr:
     if is_ndarray(arr):
         new_arr = np.zeros_like(arr, shape=shape)
     elif is_tensor(arr):
@@ -748,6 +753,8 @@ def new_like(arr, shape, fill: Optional[tuple[int]] = None):
 
 def concat_along_dim(arr_1: ImArr, arr_2: ImArr, dim: int):
     if is_ndarray(arr_1) and is_ndarray(arr_2):
+        assert isinstance(arr_1, np.ndarray)
+        assert isinstance(arr_2, np.ndarray)
         return np.concatenate((arr_1, arr_2), axis=dim)
     elif is_tensor(arr_1) and is_tensor(arr_2):
         return torch.cat([arr_1, arr_2], dim=dim)  # type: ignore
@@ -755,7 +762,7 @@ def concat_along_dim(arr_1: ImArr, arr_2: ImArr, dim: int):
         raise ValueError("Must be numpy array or torch tensor")
 
 
-def broadcast_arrays(im1_arr, im2_arr):
+def broadcast_arrays(im1_arr, im2_arr) -> Tuple[ImArr, ImArr]:
     """
     Takes [..., H, W, C] and [..., H, W, C] and broadcasts them to the same shape.
 
diff --git a/src/image_utils/library_ops.py b/src/image_utils/library_ops.py
index 720564c..0691711 100644
--- a/src/image_utils/library_ops.py
+++ b/src/image_utils/library_ops.py
@@ -4,8 +4,11 @@ def enable():
         if importlib.util.find_spec("numpy") is not None:
             import numpy as np
             np.set_printoptions(suppress=True, precision=3, threshold=10, edgeitems=2, linewidth=120)
-            from lovely_numpy import lovely, set_config
-            set_config(repr=lovely)
+            try:
+                from lovely_numpy import lovely, set_config
+                set_config(repr=lovely)
+            except:
+                print(f"Failed to enable lovely_numpy.")
         
         if importlib.util.find_spec("torch") is not None:
             import torch