Fixing Overlapping Labels and Clipped Titles in Matplotlib Subplots

The recognizable condition

You build a figure with several subplots, and the result looks wrong in a familiar way: tick labels from one Axes collide with the neighboring plot, the suptitle sits on top of the top-row titles, an axis label is sliced off at the figure edge, or a colorbar squeezes everything into a corner. It often appears only after you add something — a legend placed outside the Axes, long tick labels, a second row of plots — to a figure that previously looked fine.

The useful takeaway: this is almost never a drawing bug. It is a layout problem, and Matplotlib ships two layout engines — constrained_layout and tight_layout — that exist specifically to solve it. The job is to confirm which engine (if any) is active, enable the right one, and know the cases where neither works.

Why the default layout fails

By default, Matplotlib places each Axes at a fixed rectangle within the figure. Those rectangles are computed without knowing how large your text will be, because text extents are only measurable at draw time (they depend on the font, the renderer, and the DPI). So the default spacing parameters — left, right, hspace, wspace — are guesses that match average content. Long tick labels, rotated labels, colorbars, and outside legends all exceed those guesses, and the overflow either overlaps a neighbor or falls off the figure edge.

Cause and diagnostic table

Ordered checks

1. Confirm which layout engine is active

Run this in the same script or notebook session that creates the figure (Matplotlib 3.6 or newer; on older versions this method does not exist):

import matplotlib.pyplot as plt

fig, axs = plt.subplots(2, 2)
print(fig.get_layout_engine())

A None result means nothing is managing spacing, which fully explains the overlap. No special permissions are needed; this is a read-only inspection and changes no state.

2. Check whether a style sheet changed the default

Styles and config files can silently alter layout behavior:

print(plt.rcParams['figure.constrained_layout'])

If this prints True and you did not set it, a style sheet or your matplotlibrc did — useful to know before you start passing conflicting arguments.

3. Reproduce minimally

Build a 2x2 grid with deliberately long tick labels and a suptitle, once with the default layout and once with an engine enabled. Comparing the two side by side confirms the engine is the fix and not something else in your plotting code.

Fixes tied to the findings

Fix 1: enable constrained_layout at figure creation

This is the preferred remedy for standard subplot grids. The engine measures text extents at draw time and reserves room automatically, and it re-runs on every draw, so later additions are accommodated:

fig, axs = plt.subplots(2, 2, constrained_layout=True, figsize=(8, 6))
for ax in axs.flat:
    ax.plot([0, 1], [0, 1])
    ax.set_xticklabels(['a very long tick label', 'another long one'],
                       rotation=45)
fig.suptitle('Figure title that needs room')

Alternatively set it globally with plt.rcParams['figure.constrained_layout'] = True. Expected result: labels, titles, and colorbars get reserved space; nothing overlaps or clips. If you still see clipping after this, move to the escalation section rather than stacking more fixes.

Fix 2: use tight_layout when constrained_layout is not an option

If your figure relies on manual GridSpec positioning that constrained_layout handles poorly, call tight_layout() once, after every artist has been added:

fig.tight_layout()

Two differences matter. First, it runs once — add a colorbar or new title afterward and you must call it again. Second, it is mutually exclusive with constrained_layout: enabling both produces warnings and unpredictable spacing, so remove one before relying on the other.

Fix 3: check the saved output, not just the screen

A figure can look correct on screen and still clip in the file. savefig honors the figure's declared size unless you pass bbox_inches='tight', which crops to the drawn artists:

fig.savefig('out.png')                            # declared figure bounds
fig.savefig('out_tight.png', bbox_inches='tight') # cropped to artists

Verify by comparing the pixel dimensions of the two files. Be aware that bbox_inches='tight' produces output sizes that differ from the declared figure size, which can surprise automated pipelines that expect fixed dimensions — pick one convention and use it consistently.

Escalation criteria

If a layout engine is active and content still overlaps or clips, the problem is no longer standard layout. Check, in order:

1. Artists outside layout management. Text added with fig.text() in figure coordinates, or artists with clip_on=False, are not accounted for by either engine. Reposition them manually or convert to Axes-coordinates text.
2. Manually placed Axes. Axes created with fig.add_axes(), some inset Axes, and certain 3D Axes have historically interacted poorly with constrained_layout. For these, fall back to fig.subplots_adjust() or an explicit GridSpec with hand-set spacing.
3. Version-sensitive behavior. GridSpec support in constrained_layout improved across releases and differs meaningfully before Matplotlib 3.2. Check matplotlib.__version__ before assuming engine behavior matches current documentation; upgrade if you are on an old release.

If none of these apply, the remaining step is to reduce the figure to the smallest script that still shows the problem — that minimal case is what you want before filing a bug or asking for help, since layout issues are otherwise hard to describe.