External corners that should be crisp come out rounded, with a visible thickening on both sides of the turn. Measure across the part at the corner and it is wider than the model; measure across a flat face and it is correct.
That last observation is what separates this from over-extrusion, where everything is oversized. Here the error lives only where the toolhead changes direction.
Two systems, one of which slows down
Approaching a corner, the motion planner decelerates the toolhead so it can change direction without overshooting. It then accelerates away on the new heading.
The extruder is driven from the same planner, so in principle it slows too. The problem is that the plastic does not. Pressure in the melt takes time to fall, and while it is falling the nozzle is travelling slowly — so a length of wall that should have received a small amount of plastic receives what was queued up for full speed. The reverse happens on the way out: the head accelerates before pressure recovers, and the wall just past the corner is slightly thin.
The result is a bead that thickens into the turn and thins out of it, which reads to the eye as a swollen, rounded corner.
The fix is a calibration, and it takes half an hour
Pressure advance, or linear advance depending on your firmware, models that pressure lag and compensates for it by running the extruder slightly ahead of the motion during acceleration and slightly behind it during deceleration.
- Find your firmware's calibration print. Most produce a tall thin object printed with a different compensation value at each height, with speed changes built in so the effect is exaggerated.
- Print it in the material you actually use. The right value is a property of the filament as much as the machine — a stiff PLA and a soft TPU behave differently, and a change of material is a reason to re-check.
- Look at the corners, not the flats. The value you want is the lowest one at which the corners stop bulging, before the corners start looking hollow.
- Enter it in the filament profile rather than the machine profile if your slicer allows, so it travels with the material.
- Reprint a square tower and measure across the corners to confirm.
Expect the number itself to be small and unfamiliar: a direct drive machine typically lands in the low hundredths, while a bowden machine needs substantially more because there is far more compliance in its filament path.
When the firmware cannot do it
Not every controller supports pressure compensation, and on those machines the levers are indirect.
- Raise acceleration. Counter-intuitive but correct: a higher acceleration limit means the head spends less time in the slow region around the corner, so less excess is deposited there. This is the opposite of the advice for ringing, which is why the two have to be balanced rather than optimised separately.
- Reduce print speed overall. If the head is not going fast anywhere, the difference between corner speed and straight speed shrinks, and so does the error.
- Check flow calibration. Corner bulging on a machine that is also over-extruding will not resolve until the underlying flow is right, because you are then looking at two errors stacked.
Internal corners fail the other way
The same pressure lag that swells an external corner starves an internal one. On the inside of a turn the head is also slowing, but the geometry means the surplus has somewhere to spread, and the wall coming out of the corner is under-filled instead. So a part can show fat outside corners and small voids at inside corners simultaneously, from one cause — and if you tune for one while ignoring the other you will end up compensating in the wrong direction.
How to measure it honestly
Print a simple square tower, let it cool, and measure across the flats and across the corners at the same height. On a well-tuned machine those two numbers are close. Use the flats as your reference dimension — a caliper reading taken across a bulged corner has led plenty of people to apply a negative size compensation that then makes every flat face undersized.
What makes it more visible
PETG shows this more than PLA because it is more viscous and holds pressure longer. Fast profiles show it more than slow ones for the reason above. Small parts show it more than large ones, because a short wall never reaches steady state at all — on a part only a few millimetres across the toolhead is accelerating or decelerating for the whole path, which is also why miniature prints so often measure larger than they should.