
I own an A1 and an A2L. I tried PancakeSlicer’s browser demo on 16 November and sliced one of its sample models. I have not printed a non-planar part, on my own printers or anyone else’s. The rest of this post is built from the project’s GitHub page, coverage by Tom’s Hardware and Hackaday, Daniel Ahlers’ 2018 thesis, CNC Kitchen’s experiments and published research on strength.
A clip has been doing the rounds of a Bambu Lab A1 mini printing a cube at an angle. The cube is split diagonally into two colours, and the nozzle is laying plastic along the sloped face, rising and falling as it goes, on a surface tilted nearly 40 degrees from the bed. No flat layers. It looks wrong, and that is why people share it.
The clip is not from Bambu Lab. It comes from the launch video for PancakeSlicer, a new open-source slicer by Zimeng Xiong, who publishes as Alpaca Engineering. Tom’s Hardware and Hackaday both covered it on 15 November. The pitch is that non-planar printing has been “the next big thing” for a decade and almost nobody can use it, because no ordinary slicer supports it. So this post does two jobs: it explains what non-planar printing is and why it has been stuck, and it looks at what PancakeSlicer brings and how far I would trust it today.
What non-planar printing is
Every slicer you have used cuts a model into flat, horizontal slices. The nozzle draws one slice at a fixed height, the Z axis steps up once, and the next slice begins. Within a layer, Z does not move. My 5-axis post called this 2.5D printing, which is fair: three axes, used two at a time.
Non-planar printing drops that rule. Z moves during a line, so the nozzle can follow a curved or sloped surface the way a finger follows the top of a dome. On a normal three-axis printer the nozzle still points straight down. It cannot tilt. What changes is the path, not the machine.
In practice almost every three-axis approach works the same way. The bulk of the part is printed in flat layers as usual, and then the last few layers are laid as a curved skin over the top. That is the model Daniel Ahlers set out in his 2018 master’s thesis at the University of Hamburg, which produced the first usable non-planar slicer, a modified Slic3r. Only upward-facing surfaces get the treatment, and the toolpath is planned flat and then projected down onto the model’s surface.

The problem it solves: stair-stepping
Flat layers approximate a slope with a staircase. On a steep wall the steps are tiny and you do not see them. On a shallow surface they are wide and obvious, which is why the top of a dome or a gentle chamfer always looks worse than the sides. Ahlers’ thesis shows the step length rising sharply once a surface drops below about 20 degrees from horizontal.
The geometry is simple enough to work out. Each stair is as wide as the layer height divided by the tangent of the slope angle. These are my sums for a 0.2 mm layer.
| Slope from horizontal | Width of each stair |
|---|---|
| 45° | 0.20 mm |
| 30° | 0.35 mm |
| 20° | 0.55 mm |
| 10° | 1.13 mm |
| 5° | 2.29 mm |
A 0.4 mm nozzle draws a line about 0.4 mm wide. Below roughly 25 degrees, each stair is wider than the line that made it, and at 5 degrees it is more than five lines wide. The usual fix is thinner layers, which multiplies print time for the whole part to improve one surface.

What you gain
A smooth top surface. This is the main benefit and the one with the best evidence. If the nozzle follows the real surface, there is no staircase to see. Ahlers’ abstract says surface quality with non-planar layers was significantly better than with flat layers alone.
Thicker layers without the penalty. If the visible top is curved, the flat layers underneath can be coarse. PancakeSlicer’s own preview claimed a shorter print time for its sample than the standard slice, which I come back to below.
Some strength, in some shapes. A curved skin ties several flat layers together across their edges, and a body of research on curved-layer printing has measured gains. They are modest and they depend heavily on geometry. A 2026 review in the open literature lists one three-axis study where layers inclined at 10 degrees raised strength by 11% for coupons printed flat and 24% for coupons printed upright, and then notes that at 30 degrees the flat coupons were 21% weaker than planar ones. The often-quoted “6.35 times stronger” figure, which I cited in my 5-axis post, comes from a five-axis framework where the whole part is built from curved layers aligned with the load. A curved top skin on a three-axis printer is not that.
No new hardware, in principle. Any printer that can move X, Y and Z at once can do it. CNC Kitchen made this point in 2022: the hardware was never the bottleneck. The software was.
Why it is hard
If it is that good and needs no new parts, why are we not all doing it? Because six separate problems arrive together.
1. The print head hits the print
A print head is designed on the assumption that everything already printed is below the nozzle tip. Fan ducts, the heater block’s silicone sock and bed probes all sit within a few millimetres of that plane. Ask the nozzle to climb a slope and those parts are suddenly lower than the plastic beside them.
Ahlers put numbers on this for an Ultimaker 2. Treating the whole print head as the obstacle, the steepest safe surface was 8 degrees, on features up to 50 mm tall. Treating only the nozzle tip as the obstacle allowed 45 degrees, but only on features up to 7.5 mm tall. Large and shallow, or small and steep; not both. CNC Kitchen found that on a Prusa the fan shroud and bed sensor sit so close to the nozzle plane that barely any non-planar printing was possible, and reached a practical ceiling of about 30 degrees only after rebuilding a tool around a long nozzle.

2. The nozzle tip is flat
A nozzle does not end in a point. It ends in a small flat ring around the hole, which is what irons each line down. Ahlers measured that flat at 2.5 times the nozzle diameter on an E3D nozzle, so about 1 mm across for a 0.4 mm nozzle. Tilt the surface and one edge of that ring digs in. His formula for how far is half the flat’s diameter times the sine of the slope angle. By my sum, a 1 mm flat on a 30 degree slope digs in 0.25 mm, which is deeper than a 0.2 mm layer. The nozzle ploughs through what it has just laid. CNC Kitchen reported exactly this as the cause of failures at steeper angles. My nozzle types guide covers the shapes involved.
3. Extrusion has to change along the line
On a flat layer every line has the same height, so flow is constant. On a curved layer the gap under the nozzle varies, and the amount of plastic has to vary with it. CNC Kitchen’s G-code bending script rescales flow section by section for this reason. Ahlers found that projecting a flat pattern onto a curve over-fills hollows and under-fills crests. Hackaday reports that PancakeSlicer’s second attempt, which varied layer thickness continuously, left layers too thin in narrow regions and under-extruded there.
4. Only the top benefits
A vertical nozzle can only reach surfaces that face upward. Undersides and steep walls are printed exactly as before. And where a top surface is too steep or too enclosed for the head to reach, the slicer has to fall back to flat layers, which can leave a visible change in texture across one face.
5. The Z axis was not built for this
This one is reasoning, not test data. In normal printing the Z axis makes one tiny move per layer. In non-planar printing it moves continuously for the whole of the top skin. A commenter on Tom’s Hardware’s forum raised the worry that the Z motor is harder to drive than X and Y and could overheat, and asked that developers test for it. I have seen no measurements either way. On machines with lead screws, as on the A1 series, it is at least a duty the axis was not designed around.
6. The slicing problem itself
Deciding which surfaces can be curved, checking every move against a three-dimensional model of the print head, and blending curved layers into flat ones is far more work than cutting a model into slices. Ahlers’ Slic3r fork did it and then stopped being maintained. Reports from the time describe having to build it from source in a Linux virtual machine. That is the gap PancakeSlicer is aimed at.
How we got here
| Year | Project | What it did | Why it did not go mainstream |
|---|---|---|---|
| 2018 | Slic3r non-planar branch, Daniel Ahlers | Curved top layers on a stock three-axis printer, with a cone-shaped collision model | Research code, hard to build, no longer maintained |
| 2021 | RotBot and conical slicing, ZHAW | A 45 degree nozzle on a rotating head, printing overhangs without supports | Needs a custom fourth axis |
| 2022 | G-code bending, CNC Kitchen | A script that bends finished G-code along a curve | Limited shapes; by its author’s account buggy and without a proper interface |
| 2025 | S4 Slicer, Joshua Bird | Deforms the model, slices it flat, then reverses the deformation; printed a Benchy upside down | Runs in a Jupyter notebook and suits special printers |
| 2026 | Z anti-aliasing in OrcaSlicer | Small Z adjustments within a layer to smooth shallow tops | Shipping, but deliberately limited and with no collision handling |
| 2026 | PancakeSlicer, Zimeng Xiong | A full slicer with curved-top modes and per-printer head models | Released this month; experimental |
There is also a work-in-progress OrcaSlicer fork with non-planar features on GitHub, and hand-written toolpath tools of the kind I covered in the OGcode post. Hackaday’s summary is fair: a universal non-planar slicer still does not exist.
What PancakeSlicer is
PancakeSlicer is a complete slicer, not a script. It is built on OrcaSlicer’s slicing engine and released under the same AGPL-3.0 licence. There are desktop builds for Windows, macOS and Linux, and it can run in Docker on a home server or NAS with a web interface, which is the “self-hostable” part of the name. The README describes the non-slicing features as covering roughly what PrusaSlicer and SuperSlicer did in 2021, for one nozzle and one filament.
The non-planar side has three ideas worth understanding.
It models real print heads. Where Ahlers used a simple cone, PancakeSlicer checks its curved paths against a 3D model of each supported printer’s head. That matters, because clearance is the whole game and every head is different. Printers not on the list are checked against a generic, oversized head, which the README warns may limit what is possible and may be wrong.
It finds the top by looking down. According to Hackaday, the slicer traces rays downward through the part to find the surfaces a curved skin can cover.
It took three attempts to get the layers right. Hackaday describes the progression. First, a flat base with a few curved layers on top, where some staircase still showed through. Second, a morphed structure with continuously varying layer thickness, which removed the staircase but under-extruded in thin regions. Third, “conformal” slicing: a flat base, a morphed transition zone, and several curved layers of constant thickness on top, falling back to flat layers wherever the head cannot reach.
The README says there are four non-planar modes and then, endearingly, admits the author was too tired to document them. The demo shows the names: Curved tops, Smooth morph, Slant and Conformal layers, alongside ordinary Flat layers. An “Auto” option was greyed out in the build I used.
Which printers it supports
| Status in the README | Printers |
|---|---|
| Curved tops supported | Bambu Lab A1, A1 mini, A2L, P1P, P1S, P2S, X1, X1 Carbon, X1E, H2S; Prusa MK4, MK4S, CORE One; Creality Ender-3 V3 family, K1 and K2 families; Voron 0.1, 2.4, Trident, Switchwire; RatRig V-Core 3 and 4; Sovol SV06 and SV08 families; Anycubic Kobra 2 and 3 families; Elegoo Neptune 4 family; several Flashforge models; Qidi Q1 Pro and X-Max 3 |
| Very limited, not recommended | Bambu Lab H2D, H2D Pro, H2C, X2D (a second nozzle sharply reduces clearance); Prusa MINI, MINI IS, MK3S (the probe sits 1.5 mm above the nozzle) |
| Not listed | Still slices, using a generic oversized head. The H3C is not mentioned |
That second row matches what Ahlers wrote in 2018: dual-nozzle printers are generally unsuitable, because the idle nozzle sits at the same height as the working one. It is a small irony that the cheapest Bambu printers are better placed for this than the flagships.
Trying the demo
The demo at demo.pancake.cooking runs in a browser with no install. You cannot load your own model, but there are six samples. It opened with a Bambu Lab A1 mini and Bambu PLA Basic already selected, and “Curved tops” as the slicing method.
The settings are fewer than I expected. Under Curved tops there are top layer count, top direction and a half-height walls option. Under Head clearance there are four: ignore cooling duct, collision tolerance (1 mm by default), duct graze limit, and maximum surface angle, which defaults to 30 degrees. That default lines up with CNC Kitchen’s practical ceiling four years ago.
I sliced the sample called “Serpentine waves 10°” at 0.2 mm and switched on the built-in comparison with a standard slice.
| Measure, as reported by the slicer | Standard | Curved tops |
|---|---|---|
| Top surface error | 57 µm | 12 µm |
| Share of top surface curved | 0% | 99% |
| Print time | 22 min 2 s | 18 min 32 s |
| Filament | 7.07 g | 7.03 g |
| Head hits | not applicable | 0 |
Treat those numbers with care. They are the slicer’s own estimate of its own output, on a sample chosen by the developer at a gentle 10 degrees, and the panel carried a note reading “Print head not calibrated yet”. They are not measurements of a printed part. What they do show is that the tool reports the right things: how much of the top it managed to curve, how far the result sits from the true surface, and whether the head model predicts a strike. The side-by-side preview makes the difference easy to see, with the standard slice showing the contour bands and the curved one showing continuous lines across the surface. As a piece of interface design it is the most approachable non-planar tool I have seen.

Non-planar, ZAA, ironing or five axes?
| Approach | What it does | Hardware | Collision risk | Where to get it |
|---|---|---|---|---|
| Ironing | A second hot pass smooths the top | Any printer | None | Every mainstream slicer; flat tops only |
| Z anti-aliasing | Small Z shifts within a layer to follow shallow slopes | Any printer | Low, but not checked | OrcaSlicer 2.4.0 alpha and later |
| Non-planar tops | Whole curved layers over a flat core | Any three-axis printer with enough clearance | Real; depends on the head model | PancakeSlicer, experimental |
| Four and five axis | Tilts the nozzle or the part | Special machines | Handled by the machine’s own software | RotBot, TOP.E R1 and others |
My Z anti-aliasing post called ZAA a specialised, cheap form of non-planar printing, and that still reads right. ZAA nudges; PancakeSlicer commits. If you already run OrcaSlicer, ZAA is the low-risk way to get some of the benefit today.
Should you try it?
The developer’s own warning is the place to start. The README says non-planar mode is experimental, that there is no guarantee it will not permanently damage your printer, and that you should review the G-code preview and stay with the print. I would take that at face value. A head strike at speed on a bed-slinger is not a cosmetic problem.
Some context on maturity. The first public release, 0.4.1, is dated 9 November, and 0.4.8 followed on the 12th, so every release so far landed in four days. The repository shows nearly 700 commits. It is single-extruder only, with multi-nozzle printing untested, and 3MF support is partial. That is fast, energetic development, and it is also a tool that is a week old.
If you want to try it, this is the order I would go in.
- Play with the browser demo first. It costs nothing and shows what the modes do.
- Check your printer is on the supported list. If it is not, the generic head model is a guess.
- Start with a small, gentle shape and leave the maximum surface angle at its default.
- Scrub through the preview and look at the “Head hits” count before sending anything.
- Watch the whole of the curved section with a hand near the stop control.
- Do not use it with a second nozzle, an aftermarket duct or a bulky sock the head model does not know about.
Both of my printers are on the supported list, and I have not printed with it yet. When I do, it will be on a small part I do not mind losing, and I will report what happens, including to the Z axis.

What happens next
The interesting question is whether this stays one person’s project. Because PancakeSlicer is built on OrcaSlicer’s engine under the same licence, its non-planar code could in principle be adopted upstream, the way ZAA moved from a fork into OrcaSlicer after Bambu Lab declined it. Hackaday also points to hardware that would widen the limits, such as cooling that blows from the sides of the bed so the duct beside the nozzle can go.
My view: non-planar printing has always been held back by tooling more than by physics, and PancakeSlicer is the first tool I have seen that an ordinary printer owner could plausibly use. The physics has not gone away. Clearance still caps the angle at around 30 degrees, the flat nozzle tip still ploughs, and only the top of a part benefits. For a domed lid, a contoured grip or the roof of a model car, that could be worth it. For most functional prints it will not be. I would rather it existed than not, and I would give it a few months and a few thousand other people’s test prints before trusting it with a long job.
Where the evidence is thin
- I have not printed a non-planar part. The demo figures are slicer estimates.
- I found no independent test of PancakeSlicer’s output: no measured surface roughness, no report of head strikes or their absence, no long-term use.
- The Z-axis concern is one forum comment and my own reasoning. Nobody has published temperatures or wear data.
- The four modes are undocumented. My one-line descriptions come from their names, the demo and Hackaday’s summary of the video, which I could not watch.
- I read about three-quarters of Ahlers’ thesis: the method and limits, not the chapter of measured results.
- Strength figures come from studies on specific test shapes and will not transfer directly to a curved skin on an everyday part.
Sources and further reading
- PancakeSlicer on GitHub
- PancakeSlicer releases
- PancakeSlicer browser demo
- Alpaca Engineering: launch video
- Tom’s Hardware: PancakeSlicer coverage
- Hackaday: Printing Smooth Upper Surfaces With a Non-Planar Slicer
- Daniel Ahlers: 3D Printing of Nonplanar Layers for Smooth Surface Generation (2018)
- CNC Kitchen: Non-Planar 3D Printing by Bending G-Code
- CNC Kitchen: The RotBot and conical slicing
- S4 Slicer on GitHub
- OrcaSlicer non-planar fork (work in progress)
- Flexural response of planar and non-planar FFF (2026 review)
- 3D Printing Industry: the 6.35x five-axis strength result
- Modelling of a non-planar slicer for improved surface quality (arXiv)



