diff --git a/autolens/potential_correction/visualize.py b/autolens/potential_correction/visualize.py
index 895a12e29..43481c674 100644
--- a/autolens/potential_correction/visualize.py
+++ b/autolens/potential_correction/visualize.py
@@ -14,15 +14,15 @@
 import os
 
 import numpy as np
-import matplotlib as mpl
-import matplotlib.cm as cm
-import matplotlib.pyplot as plt
-import matplotlib.ticker as ticker
-from matplotlib.ticker import MaxNLocator
-from mpl_toolkits.axes_grid1 import make_axes_locatable
 from scipy.interpolate import griddata
 from scipy.spatial import Voronoi, voronoi_plot_2d
 
+# ``matplotlib`` is imported inside the drawing functions below, not here:
+# importing ``matplotlib.pyplot`` at module level cost every ``import autolens``
+# ~0.32s, since ``autolens/__init__.py`` imports this sub-package (census O3).
+# This matches ``potential_correction/mesh.py`` and ``iterative.py``, which
+# already defer their pyplot import the same way.
+
 
 def _plot_anchor_points(ax, anchor_points):
     if anchor_points is not None and np.shape(anchor_points) == (3, 2):
@@ -45,6 +45,11 @@ def imshow_masked_data(
     Shows a 1D (slim) data vector on its 2D mask, with masked pixels blanked,
     a colorbar and optional contours.
     """
+    from matplotlib import pyplot as plt
+    from matplotlib import ticker
+    from matplotlib.ticker import MaxNLocator
+    from mpl_toolkits.axes_grid1 import make_axes_locatable
+
     data_1d = np.asarray(data_1d)
     mask_2d = np.asarray(mask_2d)
     if centralize:
@@ -111,6 +116,8 @@ def show_image_irregular_interpolate(
     Shows values on irregular (y, x) points by linear interpolation onto a
     regular grid.
     """
+    from matplotlib import pyplot as plt
+
     points = np.asarray(points)
     points = points[:, ::-1]  # to scipy (x, y) order
 
@@ -141,6 +148,11 @@ def show_image_irregular(
     """
     Shows values on irregular (y, x) points as coloured Voronoi cells.
     """
+    import matplotlib as mpl
+    from matplotlib import cm
+    from matplotlib import pyplot as plt
+    from mpl_toolkits.axes_grid1 import make_axes_locatable
+
     points = np.asarray(points)
     points = points[:, ::-1]  # to scipy (x, y) order
 
@@ -180,6 +192,8 @@ def show_fit_dpsi(fit, output="result.png"):
     The six-panel summary of a ``FitDpsiImaging``: residual data, model,
     residual-of-residual, normalized residual, and the dpsi and dkappa maps.
     """
+    from matplotlib import pyplot as plt
+
     plt.figure(figsize=(15, 10))
     cmap = copy.copy(plt.get_cmap("jet"))
     cmap.set_bad(color="white")
@@ -256,6 +270,8 @@ def show_fit_dpsi_src(fit, output="result.png", show_src_grid=True, interpolate=
     model, residual, normalized residual, dpsi map, dkappa map and the
     source reconstruction.
     """
+    from matplotlib import pyplot as plt
+
     fig = plt.figure(figsize=(15, 10))
     cmap = copy.copy(plt.get_cmap("jet"))
     cmap.set_bad(color="white")
diff --git a/test_autolens/potential_correction/test_visualize_import.py b/test_autolens/potential_correction/test_visualize_import.py
new file mode 100644
index 000000000..561129157
--- /dev/null
+++ b/test_autolens/potential_correction/test_visualize_import.py
@@ -0,0 +1,49 @@
+"""
+Census O3: ``import autolens`` must not pay for ``matplotlib.pyplot``.
+
+``autolens/__init__.py`` imports the ``potential_correction`` sub-package,
+whose ``__init__`` imports ``visualize``. That module used to import
+``matplotlib.pyplot`` (and ``cm``/``ticker``/``mpl_toolkits``) at module level,
+so every process that touched ``autolens`` at all paid ~0.32s to build a
+pyplot state machine that only its six drawing functions use. Those imports now
+live inside the functions, matching ``potential_correction/mesh.py`` and
+``iterative.py``.
+"""
+import subprocess
+import sys
+
+from autolens.potential_correction import visualize
+
+
+def test__importing_autolens_does_not_import_pyplot():
+    result = subprocess.run(
+        [
+            sys.executable,
+            "-c",
+            "import autolens, sys; print('matplotlib.pyplot' in sys.modules)",
+        ],
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+
+    assert result.stdout.strip() == "False"
+
+
+def test__visualize_module_binds_no_matplotlib_names_at_module_level():
+    # The deferral is only real while these stay off the module: a name put back
+    # at the top of ``visualize.py`` would re-import pyplot on ``import autolens``
+    # without failing any drawing test.
+    for name in ("plt", "mpl", "cm", "ticker", "MaxNLocator", "make_axes_locatable"):
+        assert not hasattr(visualize, name)
+
+
+def test__drawing_functions_are_still_exported():
+    for name in (
+        "imshow_masked_data",
+        "show_image_irregular_interpolate",
+        "show_image_irregular",
+        "show_fit_dpsi",
+        "show_fit_dpsi_src",
+    ):
+        assert callable(getattr(visualize, name))