At the beginning of January 2024, I published an article on LinkedIn titled “What’s next for FDM / FFF 3D printers?” in which I wondered where the further development of this technology in its desktop incarnation might be heading.
Well, I wasn’t particularly generous with ideas... I only came up with two: replacing stepper motors with servo motors and converting filament printing to pellets.
Today, I look back on that article with mixed feelings. It’s one of those examples of a text that hasn’t aged particularly well. Broadly speaking, my predictions didn’t come true. Despite the 2.5 years that have passed since publication, none of the major players has pursued either of those ideas.
Well, Bambu Lab did introduce servo motors, but in its extruders rather than for moving the print head or build plate.
And as for pellet printing, not only has nobody really adopted it, but in the meantime I’ve “discovered” a number of shortcomings of the approach when it comes to desktop 3D printing, and I’m no longer nearly as enthusiastic about it as I was 2–3 years ago.
That doesn’t mean the ideas were bad. The market simply went in a different direction and ignored those solutions.
Today, however, I’d like to play around with this format again, so I’ve prepared a list of ten potential directions in which the desktop FFF sector could develop in the future. Let’s say I’m creating a baseline for another self-review article in 2–3 years.
At the moment, pretty much every manufacturer of desktop FFF 3D printers is going down the multi-toolhead and multicolor 3D printing path. Eventually, we’re also supposed to get the ability to work with multiple materials within a single print, but for now that remains more of a promise than a reality.
Anyway, it seems to me that there are incomparably more possibilities for development.
But two important notes before we start:
some of the ideas discussed here already exist on the market in some basic, early form, and my proposal is to develop them significantly further
every idea here is my own; I’m not basing this on proprietary knowledge from any other companies or individuals; this is 100% my own thinking.
So where could FFF technology go over the next 2-3 years? Here are my thoughts.
#10 Belt kinematics
I’ll start with an idea that the market has already tried once and then put back on the shelf: belt printing, i.e. printing on a moving belt instead of a rigid build plate.
The first was Dutch company Blackbelt, but the best-known product was Creality’s CR-30, also known as the 3DPrintMill. The project was funded on Kickstarter in late 2020, with Naomi Wu as its public face. Creality raised more than $1 million, with the printer starting at $538 and eventually expected to retail for $999. The build area was 200 × 170 mm horizontally, with “infinite” length along the belt axis. The print head was tilted at a 45-degree angle.
On paper, it sounded great. You could print objects that were “infinitely” long. Or you could run batch production, with finished models automatically rolling off the belt one after another while the machine kept working without an operator. A small factory on your desk.
In reality, it didn’t quite work that way. For a beginner, it wasn’t a great first machine. There was eight-point calibration, the belt’s texture showed up on the first layer, and print speeds were typical of conventional desktop printers, which meant batch production took forever because the angled geometry made a difference. The CR-30 really only made sense for long, narrow objects, such as cosplay swords.
The somewhat surprising thing about the project was that Naomi Wu openly said the machine was iterative: nobody had invented anything from scratch; they had simply taken a proven concept and made it cheaper and at a larger scale. At least that was an honest way of presenting it.
The CR-30 sliced models using an old, modified version of Cura. And Cura was never designed to lay down filament at a 45-degree angle on a moving belt. So we ended up with fresh hardware being forced onto software designed around completely different printing physics.
Then there was the limitation inherent in the geometry itself. The layers bonded at a 45-degree angle, the contact area between them was relatively small, and strength across that boundary was reduced.
But...
If you approached the concept again, calmly, with all the benefits of today’s desktop printers like auto-calibration and input shaping, and most importantly, with a slicer designed from the ground up specifically for angled belt printing, then things start to get interesting.
I mainly see it being useful for objects that are long but low-profile. Insoles, for example, or footwear made from soft TPU, where interlayer bonding behaves differently than it does in rigid plastics and angled deposition might actually be beneficial. In fact, most 3D-printed footwear today is already produced at an angle because of the geometry and the way supports are distributed. A belt wouldn’t be particularly exotic in that context.
And if you added the solutions from points #2 and #3 on this list, the belt would stop being a curiosity for cosplayers and start looking like the foundation for an entirely different additive manufacturing segment.
PS: I just saw that Mosaic Manufacturing has released the exact same printer I wrote about - Orion. It is based on belt kinematics and is specifically designed for printing insoles.
So yeah… 😎
#9 DIW / paste toolhead for silicone, ceramics, or adhesives
Again, this is nothing new, but it’s a topic manufacturers have largely abandoned. Instead of melting plastic, the toolhead extrudes a paste. This is known as Direct Ink Writing (DIW): you load it with silicone, ceramic paste, or pretty much any other printable material.
Prusa showed a silicone toolhead developed by Filament2. On the industrial side, there’s French company Lynxter, which works with silicones, ceramics, polyurethanes and interchangeable toolheads.
So the basic framework already exists. What’s missing? To make this kind of toolhead easy to use. Plug it in, calibrate it, and print it.
The problem is brutally physical. Paste doesn’t behave like filament. There is no such thing as retraction in the sense we’re used to. Two-component silicones have to be mixed in precise proportions and deposited before they start cross-linking, and after the job is finished, everything has to be cleaned before it hardens completely.
All of this requires a different flow model in the slicer, different toolpaths and different travel movements.
The toolhead itself is only half the job. Without dedicated software that understands paste rather than treating it as filament with a different diameter, such a toolhead remains an expensive dispenser for hobbyists.
#8 Conductive ink, paste dispenser and copper wire placement
If the toolhead can extrude paste, why not make it extrude paste that conducts electricity? Or, in the hardcore version, have it lay down thin copper wire directly inside the plastic during printing.
The result? A plastic enclosure comes out of the printer with an antenna, touch sensor or simple circuit already built into it. No improvised soldering and no gluing a PCB on at the end.
Again, this is something that already exists or has existed. Voltera has been selling the V-One for years, a desktop machine that prints conductive traces with silver ink, dispenses solder paste and drills vias. There was Voxel8, which combined FFF with silver-ink dispensing and paused the print so you could insert a component. There’s nano3Dprint, with a toolhead that lays down conductive paste alongside conventional filament, specifically for antennas and sensors. Copprint has been working on inexpensive copper ink as an alternative to expensive silver.
The reality? Conductive inks need to be cured or sintered, their resistance still loses out to solid copper, and inserting the actual components is still often something a human has to do. So for now, we’re talking about simple circuits, antennas and sensors. A laptop motherboard remains out of reach.
But for producing small, clever devices where the electronics and enclosure are created in a single operation, this could be fantastic. Provided the slicer learns to design the geometry and the circuit together, in a single file, instead of treating them as two separate worlds that have to be awkwardly glued together.
#7 Functional materials: conductive, magnetic, thermochromic, and piezoelectric
Filament doesn’t have to just hold its shape. It can do something more.
It can conduct electricity, react to a magnet, change color when heated (thermochromic), or turn pressure into an electrical signal, effectively acting as a sensor.
All of these materials can already be bought on a spool today. The catch is that they can be brittle and difficult to bond. Conductive PLA can handle a touch sensor or an LED running at 3 V. For a power circuit, it’s simply not good enough.
But I see a proper family of functional materials here that can print reliably on an ordinary machine. Then a single print could sense pressure, light up, react to temperature, or have a built-in touch button. Combine that with the conductive traces from #8 and the smart slicer from #5, and you end up with a little building block for creating intelligent objects.
But again: the material is only one part of the equation. Without software that understands that “this part of the geometry is a sensor, while that part is the enclosure,” we’re left with a curiosity for the shelf.
#6 Foaming filaments with controlled density
This is one of those ideas that already works, but just in a basic way.
There are filaments that foam inside the nozzle: ColorFabb’s LW-PLA, varioShore TPU, eSun’s TPU-LW, or Siraya TPU Air. They contain a foaming agent that releases gas above a certain temperature, causing the material to expand into tiny bubbles. The higher you turn up the temperature, the lighter and softer the print becomes. LW-PLA can reduce density by roughly half.
The problem is that today, you control this rather crudely. You set the temperature and flow for the entire print, struggle with calibration, fight stringing, and watch strength drop faster than weight.
Now imagine being able to control density locally, on the fly, directly from the slicer. One shoe: a hard, dense sole at the bottom and a soft, foamed insole at the top, printed in a single material. Or a drone: rigid motor mounts and a lightweight, foam-like fuselage. Or a seat that’s firm under the sit bones and softer under the thighs.
This requires two things at the same time. A hotend that can change temperature extremely quickly, because today heating and cooling the nozzle takes too long to control the foaming process at a microscopic scale. And a slicer that maintains a density map for the model rather than giving you one slider for everything.
#5 Generative, stress-aware slicing and pre-print simulation
From an engineering point of view today’s slicers are pretty dumb. They cut a model into flat slices, lay down infill in a grid or honeycomb pattern, and has absolutely no idea what the object is actually going to do. Where it will be loaded, where it will break, where there’s too much material and where there isn’t enough.
But this is already starting to change. In August 2025, Bambu Lab integrated Helio Additive’s Dragon engine directly into its slicer. Dragon performs voxel-based calculations step by step, tracks heat exchange between neighboring sections of the model, and estimates before the print even starts, where layers will bond poorly, where something will overheat, and where deformation may occur.
You get a “Thermal Index” overlaid on the model: too cold, just right, too hot, along with suggested settings such as speed, temperature and layer time. The goal is simple: get a good print on the first try, without going through rounds of trial and error.
It’s essentially a digital twin - a simulation of the entire print before you hit START. You can see that warping is going to lift a corner and fix it on screen, rather than discovering the problem eight hours later with a pile of wasted filament.
But that’s only one part of the story - the thermal side. The other, still mostly confined to laboratories, is even more interesting: slicers that calculate stress distributions using finite element analysis and lay down toolpaths along the lines of force, combined with non-planar layers, that tackle the age-old problem of weak interlayer bonding.
And that’s why I’m betting that the next few years in desktop FFF will be decided by simulation software that may not even be visible to the user.
The slicer will simply become smarter.
#4 A self-cleaning, self-changing nozzle, and a printer that can react to its own failure
The nozzle is the most disappointing component of an FFF 3D printer. It clogs, it wears out, especially when working with abrasive-filled materials.
On the one hand, this area is already being handled by control software, AI and cameras monitoring the print. But so far, that still means nothing more than detecting a failure, alerting the user and calling them in to take action.
What I dream about is a printer that can fix the failure itself. It clears the nozzle or replaces it with a new one from a magazine, and keeps going.
This is the foundation of truly unattended operation. A farm of 1000 printers makes no sense if every single one of them needs an operator.
Again: cheap sensors are the hardware, but the software is what actually does all the work by automating the operation.
#3 Pellet extrusion
Pellet is back, but in a much more humble version than I was imagining two years ago.
Back then, I saw it as a technology that would kill filament. Today, I look at it differently. Pellets are worse at maintaining surface quality, screw extruders are heavier and more difficult to control, and recycled blends can be inconsistent. And pellets don’t offer an easy way to add color. That would have to happen in a separate process, such as... turning them into filament.
But there is one scenario where pellets make more sense than almost anything else: working with waste materials that cannot be turned into filament and using them to produce ultra-cheap, disposable applications.
Such as... kamikaze drones.
Drones where nobody cares whether they come back. Where making the material as cheap as possible is one of the fundamental factors determining the success of the entire project.
At the moment, drones are manufactured from low-quality filament where color and dimensional accuracy are no concerns. Producing drone components from plastic waste could drive the material cost down to practically zero.
On top of that, you could start mixing plastic with other types of waste, such as mineral wool. Polish company IVE LABS has successfully tested 3D printing with this kind of material using its proprietary pellet extruders, with a regular Ender 3 serving as the 3D printer.
So once again, we’re not dealing with some profound breakthrough emerging from the bottom of a laboratory. It’s something that already works - it’s just extremely niche.
And if we added 3D printing technology from pellets made from leftover plastic, reinforced with non-standard materials, and onto a belt... We have a truly super-efficient production platform in a low-budget version.
#2 Variable-diameter nozzle and adaptive layer height on the fly
The entire FFF process is built around one nasty compromise. A small nozzle gives you detail, but prints slowly. A large nozzle prints quickly, but loses detail. You have to choose one or the other.
But what if the nozzle could change its diameter on the fly? Increase its diameter when printing infill and thick walls, then narrow down when it reaches edges and fine details. All in a single pass.
It sounds absurd, and yet it’s already being explored. In 2024, Science Advances described AN3DP, a nozzle with eight controlled pins arranged around an elastic membrane that can actually change the diameter and shape of the orifice during printing. It’s still a laboratory prototype, with a minimum diameter measured in millimeters, and transitions require a brief pause. But the concept is alive.
Now add adaptive layer height: thin layers on curves, thick layers on flat walls. Combine variable nozzle diameter and variable layer height under the control of the slicer, and you get genuine two-scale printing. Fast where nobody is looking, precise where it matters.
Of course, the mechanical miracle of a variable nozzle is useless without a slicer capable of planning when it should get wider and when it should narrow down.
#1 Automatic part ejection for truly unattended operation
Number one is the least spectacular idea on the entire list, which is exactly why I gave it the number one spot.
The concept is dead simple. The printer automatically ejects the finished part from the build plate, takes the next job from the queue and keeps printing. Over and over, throughout the night, without a human.
This is the foundation on which the entire consumer market is being built today. I’ve written about this many times: the real money in desktop FFF is being made by farms - thousands of printers churning out parts that people actually want to buy. A farm without automation is just a hall full of people spending all day removing prints from build plates and clicking “Start.” Expensive and stupid.
The idea is as old as the technology itself. But it still hasn’t become a fully adopted, standardized solution. That’s why I’m betting that automatic part ejection will eventually become a permanent feature of desktop printers, built natively into the machine itself. Whoever gets it right first, properly and cheaply, will hold the cards in distributed manufacturing.
So where is all of this heading?
Every idea I’ve described is some piece of hardware: a new toolhead, a new nozzle, a new material, a new mechanism. And almost every one of them runs into the same problem.
Hardware alone isn’t enough.
Let’s go back for a moment to the belt printer from the beginning of this list. Creality built a decent printer, cheaply and at scale. And it still didn’t work, because it was running on an old, modified version of Cura that had never been designed for 45-degree angled printing. New hardware was married to old software, with the expectation that somehow it would all work.
It didn’t.
Hardware will continue to evolve. I have absolutely no doubt about that. But every one of these ideas will only truly take off when software capable of driving it is developed in parallel.
A slicer that understands stress. Firmware that recognizes a clogged nozzle. A system that manages an entire print farm and closes the material loop.
Hardware and software have to move forward hand in hand.
That’s the one principle that is really worth taking away from all of this.