Toy Workshop · round 2
Printed battle tops with a creature in the saddle from the very first spin. Whoever’s rider hangs on longest wins.
Mode A · rider from the launch
Pick a rider and a stilt. Then do what the real launch asks of you: a straight, fast pull, and a smooth slide sideways so the launcher doesn’t clip the saucer. A league is three rounds. The clock runs at 4× speed, and sound stays off unless you switch it on. For the youngest players, the one-click launch lets a grown-up do the pull and the slide.
The Lightning Lizard is in the saddle on a 5 mm stilt. The top stands on its tip at the bottom of the bowl, and the launcher holds it by the pinion.
| Round | Rider and stilt | Time | Points |
|---|---|---|---|
| League total | 0 | ||
Best round: – Best league: –
The referee box from the full version: ESP32-C3, a TM1637 display, two big buttons and a piezo. Here it keeps time for you.
Wobble. Red is where this rider tips over: about 5.5°.
A rough model built from estimates. A good pull gives about 40 s of spin without a rider (a comment on model 1698197 says so), and the saddle cap and rider take about 10% off. By the jury’s estimate, a rider with its center of gravity 8 mm above the foot stays on up to a tilt of about 9°. Every millimeter of stilt lowers that limit, so the lizard on 10 mm tips at about 4°. The multipliers are the ones from the rules, before any real calibration. Records stay in this browser.
And what a round looks like
The launcher slides away and the lizard is still up there, turning with the top. A spinning top with a passenger is a new sight even for a kid who owns six tops. Then the passenger starts to lean, and the whole kitchen goes quiet.
My 9-year-old sets the Lightning Lizard, on its 5 mm stilt, into the saddle on top of the spinning top. He slides the top sideways through the slot into the launcher and lowers it until the tip taps the bottom of the bowl.
He pulls the ripcord, slides the launcher away, and the lizard spins along with the top. It wobbles more and more and falls off at second 38. With the ×1.5 multiplier, that makes 57 points.
Round 1 · Lizard · 5 mm · 38 s × 1.5 = 57
Example output: the creature card
A sample creature card. The record is the example round from the design notes; the height and weight are made up to fit the rules. The full version prints cards with measured records.
And what the referee box shows during that round
Fits on the inside of the lid
The rider is seated before the launch. The clock runs from the moment the launcher slides away until the rider falls off or the top stops spinning. Points are seconds times the stilt multiplier.
The dimple is always 1.5 mm deep. The stilt height alone sets the difficulty.
Before printing the whole set, three rides per height with one top. The multipliers get tuned until the average score is about the same for every height, which turns the choice into a pure bet. If the heights barely differ, a 15 mm stilt joins the set.
Once the top settles in the middle of the bowl, you release the rider from a dropper no more than 5 mm above the saucer. The funnel guides its foot into the dimple. Landing keeps its own record table.
Two tops in one bowl. If your rider falls off, or your top gets knocked out of the bowl, you lose the round, even if the top is still spinning.
Three rounds, points added up. You pick the stilt before each round, so there is some strategy. Bank a safe round first and gamble later, or the other way around.
She plays with the big rider, under supervision. It stands on a Ø38 × 2 mm PETG base with a neck at least 6 mm thick, and the stilt is fixed at 0 mm. When it tilts, the rim of the base rests on the Ø30 mm saucer, so it hardly ever falls off. That is her built-in handicap, and in the demo above it’s enough to beat her brother now and then.
Replay value
The heart of it
A kid describes a creature in one sentence, and that evening it rides a top, balanced well enough to play fair. Before, that sentence ended up as a drawing on the fridge, or as an evening of mine in CAD making something that wouldn’t balance anyway.
It never draws the creature. It reads the description and picks from a tested OpenSCAD library: body, head, ears, wings, tail, pattern, colors and height. Free-form 3D from a language model tends to be unprintable. A template always prints, and two creatures can be crossed later by mixing their genes.
A balancing script (Python with the trimesh library) finds the center of mass, moves it over the middle of the ball foot, tops up the weight into the 3–6 g range and cuts a flat base. Any figure that passes through it becomes a fair rider.
Fan models downloaded from Printables go through the same script. So do models from a text-to-3D service like Meshy, which only a parent runs. The game itself never needs AI. It only comes in when a new creature is born.
This page can’t call Claude, so the four answers are written in advance, in the shape the real one has to take. In the real pipeline, an answer that doesn’t fit the template’s schema gets rejected by code before anything reaches the printer. The printed creature goes to the MAKE button of a companion toy idea, or straight to PrusaSlicer.
Printed parts, one remix, one script
With the original launchers, the head of the top sits inside the launcher at launch, so a rider has nowhere to sit. We measured the downloaded STLs. The stlDenise launcher’s Ø12 mm hole is closed at the top, and on model 1749722 only a rider narrower than 8 mm would fit through the 8 × 15 mm window.
The fix is a remix of 1749722 in OpenSCAD: import the STL, then cut two things away.
Loading. The top, saucer and all, slides into the launcher sideways through the slot. The Ø10 mm pinion, the Ø11.5 mm collar and the Ø11 mm post all fit through 14 mm. Loading from below, like the original, doesn’t work with a Ø30 mm saucer.
Launch. The rider sits in the dimple and the tip stands on the bottom of the bowl. One hand holds the launcher so the pinion stays engaged, the other pulls. Then the launcher slides away horizontally in the direction of the pull, the post leaves through the slot, and the saucer passes over the plate.
| Other way to get a rider on board | Verdict | Why |
|---|---|---|
| Through hole, just opened at the top | no | Once released, the top drops through the plate, so the rider would have to pass through a Ø12 mm hole. |
| Drive on the rim, rider inside a ring gear | not for the prototype | A ring gear with a ~22 mm radius instead of ~4 mm gives about five times lower RPM and ~1/30 of the energy from the same pull. Only a gear train of our own design could make up for that. |
| Spin-up from below, pinion near the tip | no | Before the pull the top stands on its tip alone. With 1 mm of play it tilts by about 11°, and the rider falls off before the launch. |
| Side exit and launch from the floor | yes | The pinion stays where it is and the launcher slides off horizontally after the pull, so it never passes over the rider. |
In summer 2026, Printables and Open Sauce ran the “3D Printed Spinning Battle Top” contest. The reference top by stlDenise set shared rules: fully printed in plastic, a ripcord of at most 273 mm, the blade 25 mm above the ground and at most 91 mm across. That’s why parts from different designers fit together, and why this toy starts from downloads instead of from scratch.
| Model | License | Downloads / makes | What the comments say | Role here |
|---|---|---|---|---|
| 1698197 Spinning Battle Tops (stlDenise) | CC-BY-NC-SA | 328 / 9 | “Prints well”, ~40 s of spin with the basic design. In PETG the teeth on the top bent, in PLA “perfect”. Left-hand thread. | The top |
| 1749722 Modern Battle Top Launcher | CC-BY-SA | 177 / 4 | “Really great launcher”, oil helped. It jams if you push the top in over the teeth. Print the ripcord in PETG. | Launcher, remixed with a side exit |
| 1749046 MEGA Collection | CC-BY-SA | 192 / 4 | In PLA “printed really splendid”. A dad who ran a tournament with his kids would prefer a center screw. Parts get split in the slicer. | Variations: 24 tops, 47 discs, 8 tips |
| 1638345 Pull Start Launcher | CC-BY-NC | 496 / 3 | In PETG “works very, very well”. One maker scaled the rack to about 95% in Z. | Backup, its own system |
| 1183606 Customizable BeyStadium (OpenSCAD, BOSL2) | CC-BY-NC | 81 / 0 | No makes yet, like every arena we found. | Arena for the full version |
Also looked at and skipped: 1750350 (blades fell out), 1769320 (no makes), 77930 (needs screws and the original launcher). Data from the Printables GraphQL API, 2026-09-25. All the chosen models are free for non-commercial printing at home. If shared, the saddle-cap remix of the top falls under CC-BY-NC-SA.
| First playable prototype | Filament | CZK | |
|---|---|---|---|
| 2 tops (1698197) | ~100 g | ~4 h | ~70 |
| 2 saddle caps | ~6 g | ~40 min | ~4 |
| Side-exit launcher and PETG ripcord | ~50 g | ~3 h | ~35 |
| 2 riders and 6 stilts | ~14 g | ~2 h | ~10 |
| Dropper for Landing | ~10 g | ~30 min | ~7 |
| Silicone oil for the launcher (at home; otherwise ~100 CZK, estimate) | – | – | 0 |
| Total | ~180 g | ~10 h | ~130 CZK (about €5) |
Filament at 0.7 CZK/g (estimate). The big rider for our 6-year-old adds ~10 g, so ~7 CZK.
| Full version adds | CZK |
|---|---|
| ESP32-C3 (LaskaKit) | 148 |
| 2 big buttons (estimate) | ~60 |
| TM1637 4-digit display (LaskaKit) | 34 |
| Piezo | ~10 |
| Filament for the referee box | ~30 |
| Referee box | ~280 (about €10) |
| Arena from model 1183606: ~600 g, ~12 h of printing (estimate) | ~420 (about €15) |
| Total added | ~700 CZK (about €30) |
Also in the full version, no price yet: creature cards with measured records, and hex saddles with a drive lug so the rider spins up right at launch (to be tested against the round dimple).
On a phone, scroll the diagram sideways.
The part printed on the side of the box
The small-parts test cylinder measures Ø31.7 × 57.2 mm, and anything that fits inside it in any position fails. A rider up to 30 mm fits. The US legal ban covers toys for kids under three; our family rule of no small parts for the 6-year-old is stricter, and it stays.
So she gets the big rider on a Ø38 mm base, which doesn’t fit the cylinder in any position, and plays under supervision. If the creature snaps off its base, the broken piece is a small part. The riders have no magnets and no metal; a magnet version was considered for older players and dropped.
The game runs without a Mac, a phone or a network. The prototype needs a bowl and a stopwatch, and the referee box in the full version is a standalone ESP32-C3.
The internet only comes in when a new creature is made, and a parent sits there for that. Meshy is parent-only because of its age limits and because its models often won’t print. Fan models (Pokémon, Brawl Stars) are fine to download and print at home. Don’t share or sell them; the characters are protected.
Move your piece one space at a time
One remixed launcher, one top with a saddle cap, one rider. Does the launcher slide off without hitting the stem or the saucer? Which way does the ripcord pull?
Two tops with saddle caps, the side-exit launcher, a PETG ripcord, two riders with 0, 5 and 10 mm stilts, a dropper for Landing, a drop of oil in the launcher hole. Plays without a Mac or a network.
Spin times without the cap, with it and with a rider. Three rides per stilt height, then set the multipliers. Add a 15 mm stilt if the heights barely differ.
The creature template, the balancing script and Claude filling in genes. The big rider for our 6-year-old.
The ESP32-C3 referee with a “3, 2, 1, let go!” countdown and a dual stopwatch, the 1183606 arena, creature cards with measured records, hex saddles with a drive lug.
Status: an idea. Nothing has been printed or tested yet.
Blind AI jury, round 5
Round 5 total, down from 22 in round 4.
If the first test launch works, Build goes up to 4 and the total to 21.
“A strong game loop (the saddle as a bet, collecting creatures, the league) for ~130 CZK. But the rider physics, meaning landing on a spinning top and staying in a round dimple, is untested so far, and the dimple probably doesn’t control difficulty the way the card assumes.”
“This toy drops from 22 to 20. The uncertain rider physics pulls Build down (4 → 3), and landing the rider pulls the first 30 seconds down (5 → 4). The game loop, on the other hand, is stronger (tenth time 4 → 5).”
Both were scored before the revised design, which moved the difficulty to stilt height and put the rider on the top from the launch. Reviewing the revised design, the jury kept the total at 20 and added three catches: the slot opposite the pull, sideways loading and more vertical clearance. All three are in this page.
One of sixteen finalist toy ideas in the Toy Workshop, generated by AI agents, picked by a dad.
Made by Ondra Malacka with Claude agents · how the agents worked · malacka.cz