Synilay Launches 3D Printed Artificial Intelligence Shoes

• 1
  How AI reaches a printable 3D printed artificial intelligence shoes geometry
• 2
  Why a lattice midsole behaves the way it does
• 3
  Which materials hold a printed midsole
• 4
  Where the process stops being economical
• 5
  Combining printed lattices with CNC components
• 6
  Process fit for a footwear component

Synilay's 3D printed artificial intelligence shoes are not printed by a robot that thinks. A generative model proposes shapes. A human designer rejects most of them. The surviving geometry then has to be converted into a mesh that a printer can actually deposit, layer by layer, without collapses or trapped material.

In the Synilay workflow, an image model generates early form studies. A designer redraws them by hand during the sketch phase. A second tool converts the sketch into a 3D surface model, and a third adds surface texture and internal structure. Each handoff loses information unless the file stays parametric, so the team keeps a master model and treats the AI output as reference art, not as production data.

The useful part is not the image. It is the density map. A generative loop can propose where material should be thick and where it can be hollowed into a lattice, based on pressure zones under the foot. That is a structural decision, and it is the same decision a machinist makes when lightening a bracket.

Once the mesh is fixed, the shoe stops being a design problem and becomes a manufacturing problem. Wall thickness, overhang angle, and minimum feature size all have hard limits. If the lattice strut is thinner than the nozzle can deposit, the model is fiction.

• 1
  AI output is reference geometryTreat it as art direction until a parametric mesh exists.
• 2
  Density map matters more than shapeMaterial placement drives comfort and weight, not the outline.
• 3
  Printer limits are the real constraintStruts below the nozzle diameter cannot be built.

A lattice midsole is a spring network. Each strut bends under load, stores energy, and returns it when the foot lifts. The stiffness of the whole structure comes from three variables: strut diameter, cell angle, and the number of cells stacked in the load path. Change any one and the ride changes.

That is why 3D printed artificial intelligence shoes can be tuned without changing the outer shell. The generative model shifts strut thickness in the heel and forefoot independently. A heel strike zone might run 1.2 mm struts, while the forefoot runs 0.8 mm for a softer push-off.

The catch is fatigue. Elastomer struts in a midsole see hundreds of thousands of compression cycles. A lattice that feels right on day one can sag after 50 km if the strain per strut is too high. Engineers usually target a strain range that keeps the material below its yield point across the full gait cycle.

Lattice structures also trap debris and water. Open cells drain poorly. Closed cells trap air and stiffen under compression. Neither is wrong, but the choice should be made early, because it changes both the feel and the cleaning routine.

• 1
  Stiffness = strut diameter × cell angleBoth variables are tunable per zone in the same print.
• 2
  Fatigue sets the floorKeep strut strain below yield across the gait cycle.
• 3
  Open vs closed cellsOpen drains, closed stiffens. Pick before tooling.

Most printed midsoles use thermoplastic polyurethane or a comparable elastomer. TPU prints well, bonds between layers, and has enough rebound for a shoe. It also creeps under sustained load, which is why storage matters. A pair left compressed in a hot car will take a set.

For higher rebound, some teams move to a photopolymer with a higher resilience. These resins print finer lattices, down to 0.4 mm struts, but they are more brittle and UV-sensitive. A resin lattice can crack at a thin strut after repeated impact if the formulation is not toughened.

Carbon-filled filaments add stiffness without much weight, but they wear nozzles and reduce layer adhesion. They suit a structural shank or a heel counter more than a full midsole. Mixing materials in one part means mixing failure modes, so keep the interface generous.

The choice is really about the load case. Walking shoes favor damping and creep resistance. Sprint plates favor stiffness and low hysteresis. A single material rarely wins both, which is why many designs use a stiff plate under a soft lattice.

• 1
  TPU is the defaultGood layer bonding, moderate rebound, prone to creep.
• 2
  Resins print finerDown to 0.4 mm struts, but brittle and UV-sensitive.
• 3
  Carbon fill is for platesStiff, abrasive to nozzles, weak layer adhesion.

Printing an entire shoe makes sense when the geometry is impossible to mold. A lattice with graded density and internal channels cannot be pulled from a two-part tool. That is the real argument for additive, not the AI label.

It stops making sense when the part is a simple solid. A flat insole, a shank, or a heel wedge can be machined or molded faster and cheaper. Additive wins on complexity, not on volume. At 10,000 units, a molded midsole will usually beat a printed one on unit cost.

There is also a size limit. Most elastomer printers build within a 300 mm envelope. A men's size 12 midsole is around 320 mm long, which pushes against the edge of many machines. Larger builds need either a bigger platform or a split part with a bonded seam.

Finally, surface finish is coarse. Printed elastomer comes out with visible layer lines, typically Ra 6–12 μm. If the visible surface must be smooth, it needs post-processing, and most elastomers resist sanding and polishing.

• 1
  Additive wins on complexityGraded lattices and internal channels cannot be molded.
• 2
  Molding wins on volumeSimple solids are cheaper at high unit counts.
• 3
  Build envelope is the ceilingMost elastomer printers top out near 300 mm.

Few high-performance shoes are fully printed. Most combine a printed lattice with a machined or molded plate, a metal shank, or a CNC-trimmed carbon heel. The printed section handles cushioning. The machined section handles stiffness and dimensional control.

This is where a CNC shop enters the picture. A carbon fiber plate needs a clean edge, a consistent thickness, and holes that line up with the lattice pockets. Those features sit in the ±0.05 mm range, which is well inside what a 3-axis mill can hold.

At GreatLight we machine plates from carbon fiber, 7075 aluminum, and titanium for footwear and wearable prototypes. We hold ±0.005 mm on metal parts and Ra 0.8–1.6 μm on mating surfaces. The printed lattice comes from a separate process and we work to the interface dimensions the designer sets.

The practical rule: print the soft, complex, low-load parts. Machine the stiff, simple, high-precision parts. Then check the assembly tolerance stack before either process starts, because printed elastomer shrinks and the plate does not.

• 1
  Print the cushioningLattices, channels, and graded density go additive.
• 2
  Machine the stiffnessPlates, shanks, and interfaces go subtractive.
• 3
  Check the stack earlyPrinted parts shrink; machined parts do not.