The underside of an overhang lifts away from where it should be. On the next layer the nozzle rides over a surface that is higher than the machine thinks, and by twenty layers up there is a rhythmic knock every time the toolhead crosses that feature.
This is not the same complaint as a rough sloped underside. If the surface sags downward and looks fuzzy, you want overhang droop. Curling goes the other way: the edge rises.
It is a feedback loop, which is why it escalates
An overhang layer is partly cantilevered over air. Molten plastic laid with nothing beneath it contracts freely as it cools, and free contraction on a beam anchored at one end pulls the free end upward.
Now the important part. The next layer is printed at the height the slicer planned, but the surface it lands on has risen. So the nozzle presses into it, compacting the extrusion and adding heat exactly where the part is already deformed — which curls it further. Each layer amplifies the last.
That is why this defect is mild for a few layers and then suddenly severe, why it looks fine in a preview and appalling on the plate, and why catching the first two layers of it matters far more than any single setting does.
The three things that decide it
Cooling. The curl happens in the window between the plastic leaving the nozzle and solidifying. Shorten that window and there is less time for it to lift. Cooling is the strongest lever here and it is available on nearly every machine.
How much of each bead is unsupported. Set by the overhang angle and by layer height together. A shallower layer height means each new bead hangs over the previous one by less, so a 0.2 mm layer curls where a 0.3 mm layer at the same angle fails outright.
Speed at the overhang. Slower gives each bead more time to set before its neighbour arrives — the opposite of the advice for bridges, where speed keeps the strand in tension. The two look similar and behave oppositely, which is a common source of confusion.
In order of what to try
- Give overhang regions everything the part-cooling fan has. Most slicers can raise cooling for overhangs specifically without changing it elsewhere.
- Slow overhang perimeters to somewhere near half of normal perimeter speed.
- Drop the nozzle temperature by 5 to 10 °C. Cooler plastic sets sooner. Watch for weaker layer bonding if you go far.
- Halve the layer height for the region, or for the whole part if adaptive layers are not available.
- Reorient or support it. If the feature is at a steep angle on a functional part, the settings above are managing a problem the model does not have to have.
If the curl is already causing knocks, enable Z hop as a stopgap so the toolhead clears it — but treat that as buying time, not as a fix. See nozzle hitting the print for what happens when it is not addressed.
Where the enclosure works against you
Anyone printing ABS or ASA is deliberately running with cooling off and a warm chamber, because that is what stops those materials cracking. Both of those choices make curling worse, and there is no setting that resolves the contradiction.
The workable compromise is a small amount of cooling for overhang regions only — enough to set the free edge, not enough to chill the whole part — plus accepting that a machine tuned for delamination resistance will not produce crisp unsupported edges. On those materials, supports and reorientation do more than cooling ever will.
The materials that curl
PETG curls readily and is the material most people first meet this on. ABS and ASA curl for the contraction reason and are then made worse by the cooling compromise above. Polyamides curl and are stringy while doing it. PLA is the best-behaved material here by a distance: it sets quickly, contracts little, and with good cooling will hold overhangs that no other common filament manages.
Do not confuse the cure with the symptom
Increasing supports everywhere is the usual response and it is a blunt one. Supports leave marked surfaces of their own, cost material and time, and are unnecessary on a machine with adequate cooling for anything up to a fairly steep angle. Test the cooling change first on a small overhang test print; it is a ten-minute experiment that decides whether supports are needed at all.