Auto Z Offset

Auto Z-offset (the load sensor) — and why we don't rely on it

Your KE can set its Z-offset automatically: the nozzle taps the bed, a sensor feels the contact, and Klipper writes the offset for you — no paper, no eyeballing. It's built into the stock firmware and it works, so you're welcome to use it. But we tested it carefully and found it isn't repeatable enough to trust for a good first layer. This page explains what it is, shows our numbers, and tells you what we do instead.

The short version: auto Z-offset on the KE is convenient but noisy — it can land ~0.08 mm apart run to run, where a good first layer needs to be within ~0.01–0.02 mm. That's a hardware limit of the sensor Creality chose, not a screen or OpenKE problem. Use the manual on-screen baby-step instead (it's the reliable path, and it's why your KE also has a BLTouch).

---

What "auto Z-offset" actually is

The KE has a strain-gauge / ceramic pressure sensor in the toolhead. When the nozzle pushes down on the bed, that sensor detects the force and reports "contact." Creality's firmware turns that into a few gcode commands — Z_OFFSET_CALIBRATION, Z_OFFSET_AUTO — that home, clean the nozzle, probe the bed with the nozzle tip itself, compute the Z-offset, and save it automatically.

The idea is genuinely nice: the nozzle is the probe, so there's no probe-to-nozzle offset to worry about, and you never touch a feeler gauge.

Why we don't trust it — our test

We ran the no-frills calibration routine five times back to back on a real KE and recorded the offset it computed each time (one run was lost to a midnight log rotation):

Run	Computed Z-offset (mm)
1	2.676
2	2.746
3	2.714
4	2.759

• Spread (range): ~0.083 mm. Standard deviation ~0.032 mm.
• A first layer that looks the same every time needs the offset to be stable to about ±0.01–0.02 mm.
• So the auto routine is roughly 2–4× noisier than the tolerance that matters.

Interestingly, within a single run the two probe points agreed nicely (~0.006–0.015 mm). It's the run-to-run result that drifts ~0.08 mm — the classic signature of strain-gauge preload and thermal drift. In plain terms: the sensor's "zero" wanders a little each time it heats, cleans, and presses, so the number it lands on wanders with it.

Bottom line: if you auto-set your Z-offset today and again next week, you can easily get a first layer that's visibly different — gappy one time, smeared the next — through no fault of your own.

Why this is the sensor, not the screen

This matters, so it's worth being clear: the unreliability comes from Creality's choice of sensor, not from OpenKE or the touchscreen.

• The probing logic lives entirely in Creality's stock Klipper module (a compiled .so). OpenKE doesn't run, wrap, or modify it — the screen has no way to make a strain gauge more repeatable.
• Strain/load-cell nozzle probing is inherently drift-prone unless it's built with careful ADC wiring and compensation. The KE's implementation is the convenient, low-cost version.
• The strongest evidence is on the printer itself: the KE also ships a BLTouch as its real Z probe. If the nozzle-load sensor were accurate enough on its own, that second probe wouldn't need to be there.
• One more wrinkle: the auto routine probes at around 140 °C, not your printing temperature (~210–230 °C). The nozzle hasn't fully expanded yet, so even a perfectly repeatable reading would be a slightly different absolute offset than your true hot offset.

A gotcha if you do run it

The auto routine saves immediately (Creality's own save step), with no Klipper restart. Two things to know:

1. It overwrites your saved Z-offset without asking — there's no "test only" mode exposed.
2. If you've just done a KAMP (adaptive-mesh) print, that save can also persist the small temporary adaptive mesh over your saved full bed mesh, which throws off the first layer across the whole plate. If you ever run auto Z-offset, check your saved bed mesh afterwards (or restore a config backup).

What to do instead — the reliable path

Set Z-offset with the screen's live baby-stepping, which is repeatable because you judge the actual first layer:

1. Start a first-layer test (a single-layer square or patch).
2. On the Tune tab — or tap the Z-offset readout while it prints — nudge Z in 0.01–0.05 mm steps.
3. Aim for lines that are squished together and fused: not gappy/round (too high), not translucent or smeared (too low).
4. With Save Z-Offset installed (an installer option), it persists automatically.

Done when: the first layer is uniform with no gaps between the lines — and it stays that way next time, because the value didn't come from a drifting sensor.

Full first-layer walkthrough: Calibrate step by step → Z-offset. For uneven left↔right first layers, that's a different fix — Axis Twist Compensation.

"So should I ever use auto Z-offset?"

It's fine as a rough starting point — e.g. after a nozzle change, to get in the ballpark fast — as long as you follow it with a baby-step pass on a real first-layer print and save that. Just don't treat the number it prints as your final, trustworthy offset.