diff --git a/torchvision/tv_tensors/__init__.py b/torchvision/tv_tensors/__init__.py
index 1ba47f60a36..2fd9b9fbc7d 100644
--- a/torchvision/tv_tensors/__init__.py
+++ b/torchvision/tv_tensors/__init__.py
@@ -1,3 +1,5 @@
+from typing import TypeVar, cast
+
 import torch
 
 from ._bounding_boxes import BoundingBoxes, BoundingBoxFormat
@@ -7,12 +9,14 @@
 from ._tv_tensor import TVTensor
 from ._video import Video
 
+TVTensorLike = TypeVar("TVTensorLike", bound=TVTensor, covariant=True)
+
 
 # TODO: Fix this. We skip this method as it leads to
 # RecursionError: maximum recursion depth exceeded while calling a Python object
 # Until `disable` is removed, there will be graph breaks after all calls to functional transforms
 @torch.compiler.disable
-def wrap(wrappee, *, like, **kwargs):
+def wrap(wrappee: torch.Tensor, *, like: TVTensorLike, **kwargs) -> TVTensorLike:  # type: ignore
     """Convert a :class:`torch.Tensor` (``wrappee``) into the same :class:`~torchvision.tv_tensors.TVTensor` subclass as ``like``.
 
     If ``like`` is a :class:`~torchvision.tv_tensors.BoundingBoxes`, the ``format`` and ``canvas_size`` of
@@ -26,10 +30,10 @@ def wrap(wrappee, *, like, **kwargs):
             Ignored otherwise.
     """
     if isinstance(like, BoundingBoxes):
-        return BoundingBoxes._wrap(
+        return cast(TVTensorLike, BoundingBoxes._wrap(
             wrappee,
             format=kwargs.get("format", like.format),
             canvas_size=kwargs.get("canvas_size", like.canvas_size),
-        )
+        ))
     else:
         return wrappee.as_subclass(type(like))