How I Transformed 3 Different 3D Printers From Frustrating to Fantastic: a Calibrated Approach Using AI to Improve Your 3D Prints

This Instructable is a bit different from the usual “here’s a thing I built” project. What I’m sharing here is a method to dial in any 3D printer so you can actually print the projects you want. These days AI gets used for everything, and most people see it as some magic box that replaces thinking. I don’t. As a maker I treat AI like any tool. Each has strengths and weaknesses. You don’t use a screwdriver as a chisel and expect a masterpiece, and you shouldn’t treat every AI model as equal either. Some help, some don’t. I use Prusa slicer, it's brilliant and the results I got here are with ChatGPT, because it worked for this task. Why? Because it can read 3MF files, you see, they are compressed Zip files essentially, DeepSeek, Gemini etc... just say sorry, cannot help.

Another reality is that unless you have money to burn on high-end machines, you need to understand the limits and quirks of cheaper printers. They can do great work, but you must understand them and calibrate them properly. I have four 3D printers:

1. A PRUSA RepRap i3 I built from scratch
2. An Ender 3 Pro I bought because I had more ideas than time
3. A PRUSA Bear upgrade that started as a RAMPS project and then turned into a genuine Prusa build when a friend gave me their parts after they upgraded to MMU
4. A Tronxy X5A that arrived as a dead machine and is now fully enclosed and alive again

I browse Instructables a lot, and something hit me one day. I look at some prints posted there, and I realise that even my prints I thought looked bad, are actually way better than many of the examples online. But that didn’t matter to me, because I want every single print to come out as clean and accurate as possible. If I design something, I want the end result in my hand to feel like a finished piece, not something that “sort of worked.”

So this isn’t showing off one single finished part. It’s sharing the exact method that took all of these printers from “good enough” to producing results that feel like proper master prints. And the best part is that the method works across completely different machines. If you follow the steps, you should be able to get there too. I’ll include photos of the calibration prints and results from each printer once we get into the details.

Supplies You’ll Need

This isn’t a crazy list.

We’re not building a fusion reactor -- just dialing in the perfect print.

Hardware:

1. A 3D printer (any FDM/FFF style)
2. The nozzle currently fitted (0.4mm default assumed, but any will work)
3. A set of calipers (digital preferably - we are measuring millimetres, not vibes)
4. A clean print bed
5. A stable surface to place the printer on

Filament:

1. The exact filament you want to calibrate
2. PLA, ABS, PETG, ASA - doesn’t matter
3. Just use the same roll throughout the calibration process

Software:

1. PrusaSlicer (or your slicer of choice, but I reference PrusaSlicer settings here, it's free too)
2. The XYZ Calibration Cube STL (free from Thingiverse)

Extras (Optional, but highly useful):

1. Phone or camera to document photos
2. Notebook or digital notes (for recording measurements and slicer settings)
3. USB cable/SD card for printing
4. Patience (yes, this is a supply, anyone who’s tuned printers knows why )

Before we start correcting anything, you need a reference print.

Not a Benchy.

Not a random gear.

Not that “perfect item” you saw someone flex on Facebook.

You need the Cube.

The Mission:

Print a single XYZ Calibration Cube using the filament you want to tune.

Model link: (Thingiverse) “XYZ 20mm Calibration Cube”

(thing:1278865 — and yes, even the photos there could use calibration... no jokes!)

Settings Don’t Matter. Yet.

Right now, it's not about choosing:

1. the right speed
2. the right temp
3. the right flow
4. the right retraction
5. or whether your walls are “ironed” or “vase mode spiritual”

WE DON"T CARE. Your current settings are your baseline. They're wrong, wonky or wandering, that's the whole point.

We’re going to tune from where you actually are, not where someone says you “should be.”

So whatever profile you used last… use that.

Whatever slicer defaults you pray to…

Just send the cube to print.

Why This Cube is Essential

That little 20mm cube is the MRI scan of your printer:

1. If Z-offset is wrong → it tattles
2. If extrusion is too high/low → it snitches
3. If temperatures are mismatched → it exposes
4. If mechanical wobble is present → it humiliates
5. If cooling is weak → it melts
6. If dimensional accuracy is off → it screams

It’s a diagnostic tool disguised as a toddler’s building block - Once you have this right PRINT 50x50 Cube it's just cool.

And Yes… Even the Example Photos Need Help

Go stare at the photos on Thingiverse.

Right?

Layer scars. Over-extrusion. Misalignment. Z-banding.

The cube shown on the listing could literally benefit from the exact tuning tips we're about to apply.

So don’t feel bad about your prints, the Calibration Cube humbles everyone. Just Take a photo and look at it, it's quite scary, maybe perfect to the naked eye, but a camera reveals all.

What You Do Now:

1. Print the cube
2. Don’t sand it
3. Don’t smooth it
4. Don’t fix anything
5. Don’t reprint it yet

Just bring your flawed little truth-teller into Step 2.

Now we’ve printed the Cube of Truth.

We’ve stared at it.

We’ve touched it.

We’ve judged it harder than a bake-off judge staring at raw dough.

But here’s the deal:

That cube isn’t “ugly”… it’s honest. not to our eyes, it tells you exactly where your printer is misbehaving.

What we look for (and what I found)

When you upload the photo to ChatGPT, don’t just dump it without context. You describe what you see, because, ChatGPT is clever, but it isn’t a mind reader and it doesn’t have your physical cube in its hands

In my case, I said:

“The sides on the Cube are dimensionally all correct, measuring 20mm on all Axis, but what it not correct it looks like there is still under extrusion on the Z letter on the Cube. You can see this in the photo attached. The three perimeter lines still have a gap between those, can we correct that?”

That alone already pointed us in the right direction, as we are telling ChatGPT what is correct and what is not, by uploading the image, it can see where things are not right too, maybe something you missed, looks at the ChatGPT response in this instance (see below). By Telling ChatGPT, your environment, what you are using to print, the actually thickness, it can cross reference that with the 3MF file from PRUSA slicer, on other agents your can upload a photo with the parameters, that is likely to work too.

But what I learned from the response is that we should add more CONTEXT:

"I am printing on an Ender 3 Pro, with 0.4mm nozzle, there is still gaps between top infill and the perimeter, as well as small gaps between the perimeter walls. We are printing with a PLA Filament, the actual diameter of the PLA is 1.72mm"

Else the agent would not have confirmed with the following in a conversation:

I QUOTE:

"

🧪 If you want, share:

1. Material (PLA, PETG?)
2. Layer height
3. Print temperature
4. Slicer (Cura/Prusa/etc.)
5. A small photo

And I will pinpoint exactly which setting is the culprit.

🔥 Ender-3s can make perfect sealed surfaces — this is 100% fixable."

After that, this was ChatpGPT's response.

🧰 CAUSES & FIXES

🔧 1) Line Width / Flow too low

Even if extrusion multiplier is “100%”, slicers often default to 0.40 mm line width with a 0.40 mm nozzle, and that is just barely enough to seal.

✔️ Fix:

Set line width manually:

🔧 2) Top Layer Count Too Low

If you can see infill pattern, add more top layers.

✔️ Fix:

1. For 0.2 mm layer height → at least 6 top layers
2. For 0.16 mm → 7–8 top layers
3. For 0.12 mm → 8–10 top layers

🔧 3) Flow Rate / Extrusion Multiplier Needs Tuning

Under-extrusion causes tiny air gaps.

✔️ Fix:

Increase flow 3–5%

1. Start with +2%
2. Print a calibration cube
3. Increase again if needed

Typical good values: 102–105 %

🔧 4) Print Temperature Too Low

Cold plastic doesn't fuse to the perimeter.

✔️ Fix:

Increase nozzle temp by:

1. +5 to +10 °C

Example:

1. If printing PLA at 200 °C → try 205–210 °C

🔧 5) Top Infill Overlap

This is the big one.

The slicer needs to intentionally push infill into the walls.

✔️ Fix:

Set Infill–Perimeter Overlap to:

🔧 6) Wall Ordering

Cura has an option Wall Ordering: Inside to Outside

This can leave small gaps.

✔️ Fix:

Set:

🔧 7) Speed too high

Fast infill or walls reduce fusing.

✔️ Fix:

1. Top layer speed: 20–30 mm/s
2. Wall speed: 30–40 mm/s

🔧 8) Nozzle Pressure Issues / Z Seam

If the seam is under-extruding, you see gaps.

✔️ Fix:

Change Z-seam to a hidden location

OR

Set seam = random

--------------------------------------------------------------------------------

🔍 QUICK TROUBLESHOOT LIST (IN ORDER OF IMPACT)

1. Increase overlap: 25%
2. Solid Infill Line Width: 0.44 mm
3. Increase extrusion: +3%
4. Increase temp: +5–10°C
5. Increase top layers: 6+
6. Reduce top speed: 20–30 mm/s

--------------------------------------------------------------------------------

As you can see ChatGPT can be very thorough, But is was giving answer for Cura, now that is proof that it can get somethings wrong, as I gave it a 3MF file, which is PRUSA Slicer, so I reconfirmed (CONTEXT) that we are using PRUSA Slicer, this time ChatGPT gave a slightly different result, and it was a lot more revealing.

I quote:

"Absolutely — I’ve studied the photo and the (uploaded) 3MF.

The symptoms you see around the Z perimeter on the top of the cube are classic signs of top layer under-extrusion / missing infill contact.

🔍 What We're Seeing

✔ Small voids/gaps around the raised perimeter of the Z

✔ Top infill lines don't close fully against the contour

✔ Top surface isn’t completely “melted together”

IMPORTANT: This isn’t a geometric issue — it’s a material flow + top-layer coverage issue.

🎯 Causes (Ranked Most Likely → Least)

1. Not enough top solid layers
2. Infill too low to support the top skin
3. Extrusion a little low
4. Print speed too high on top layers
5. Cooling too strong during top skin

🛠 Fixes in PRUSA SLICER (for Ender 3 Pro PLA)

✅ 1) Increase Top Solid Layers

You're printing at 0.2 mm layers →

Minimum recommended top thickness = 1.0–1.2 mm

So:

(6 × 0.2 = 1.2 mm)

✅ 2) Increase Infill % (just a little)

To avoid unsupported gaps below the final layer:

✅ 3) Slow Down Top Solid Layers

Reduces filament stretching when bridging next to the perimeter:

✅ 4) Increase Extrusion Slightly on Top

In PrusaSlicer:

✅ 5) Extra Perimeters (Optional but Helps)

For raised lettering:

OPTIONAL: Ironing the Top

If you want a perfectly fused surface:

🔧 Printer Side Checks (Quick)

These are quick wins and shouldn’t be skipped:

✔ Calibrate E-steps properly

If extrusion is 2–3% low, the symptoms match EXACTLY.

✔ Filament Diameter

Measure with calipers and set in slicer if not actually 1.75.

⚡ Conclusion

The gaps around the Z perimeter are caused by too few top layers + slight under-extrusion.

The most effective cure is:

Top Solid Layers = 6, Extrusion Multiplier = 1.03, Top Speed = 25mm/s

After this your top should be perfect — fullness and smooth closure around the perimeter.

"

In the photo what ChatGPT Saw (The Good, the Bad and the less obvious to me)

It confirmed what I suspected:

- Good:

1. Perfect dimensions
2. Good layer adhesion
3. Clean corners

- Needs tuning:

1. Tiny gaps between the top infill lines
2. Slight spacing between perimeters and infill

- Quick steps:

1. Calibrate E-steps properly
2. Filament Diameter

All of that means, the top layers aren’t laying down enough material relative to the walls. It's not a slicer catastrophe, not a mechanical failure, just a calibration mismatch. ChatGPT can do the "MATHS" so to speak and fix it. But see these things don't happen in Isolation and this is where using AI can make a huge difference, it's calculating based on, one input change how it may affect another,

But this still didn't fix the issue, so I decided maybe those Quick Steps are correct, I have already checked the filament diameter, so I know that was correct, but I never checked the E-steps. I assumed the E-Steps was still correct, after I replaced the plastic tension arm to a Aluminium tension arm, I didn't replace the extruder gears, so it shouldn't have changed, but after I measured, turned out it was under extruding by 14mm!! So I asked ChatGPT to fix that.

I QUOTE:

"

Absolutely — here’s exactly what you need to fix your E-steps based on a measured under-extrusion of 14 mm.

🧮 Quick Confirmation of the Adjustment

You told me the printer extrudes 14 mm less than commanded.

You most likely commanded 100 mm, so extrusion was 86 mm.

So the correction factor is:

So you need to increase by ~16.3%.

🧾 STEP 1 — Show (Read) the Current E-steps

This reports the ORIGINAL setting stored in EEPROM:

Look for this line in the output:

🧾 STEP 2 — Set the Corrected Value

Let’s say the current E-steps are (example) 93.

Calculation:

Rounded to:

You will insert YOUR measured number, not 93.

🧾 STEP 3 — Save E-steps to EEPROM

✔ Final “Working Example” G-code Block

👉 To read the original value:

👉 To set the new corrected value:

👉 To save:

🔍 Example Using Typical Ender 3 Values

If your original M503 shows this:

Then your corrected setting becomes:

⚡ After Updating, Re-test Calibration

1. Mark 120 mm
2. Command 100 mm
3. Should now extrude 100 mm dead on

If off by <1 mm, you're perfect.

"

That sorted the issues, and the printer is near perfect, and now I can live with that.

And the crazy part?

These exact issue exists on more prints in the world than bad selfies.

But nobody fixes it, because few now how to read the cube, but also sometime you cannot see, if so enlarging it with a photograph, makes it visible. Read what ChatGPT is saying, Make the tweak, reprint the Cube, Upload the new pic, and we push it even closer to perfect. Then repeat the steps, till you are happy.

Question, I bet will be does it work. Proof as the saying goes is always in the Pudding. Lets examine the Prusa Bear Upgrade in the next Step, I started building this printer as a Bear Upgrade to be completed with Ramps boards (as I had), coincidence had it that a friend was gifted Genuine PRUSA parts, from another chap we know, who did a full MMU upgrade, so now my Bear Upgrade, became a genuine PRUSA BEAR Upgrade.

This is what ChatGPT had to say after looking at a Print, then follow the images to see the result, The Orange it PETG and the Yellow Cube is a 50x50mm print in ABS, two very difficult materials to print

🔍 What I See (and what it means)

✔️ SURFACE QUALITY

The side walls are very smooth — this tells me:

1. Extruder steps are correctly calibrated ✔️
2. Layer flow is consistent ✔️
3. Z-motion is properly aligned ✔️

So your extrusion mechanics are now dialled in.

❗ Issues I see on the tops and corners

1) Stringing / Wisps between features

Visible especially near the Z inset.

Cause: PETG likes to ooze.

Fixes:

1. Increase retraction:
2. Direct drive: ~0.6–1.2 mm
3. Bowden: ~3.5–6.0 mm
4. Increase retraction speed to ~35–45 mm/s
5. Lower print temperature: 240°C → 235°C (Try 235 first)
6. Enable “combing not in skin”
7. Enable “wipe on retract”

2) Blobbing on corners (micro-zits)

Likely due to PETG pressure buildup at ends of perimeter loops.

Fixes:

1. Reduce linear advance / pressure advance (if enabled)
2. Reduce print speed slightly on external perimeters
3. Add coasting (PrusaSlicer calls it "Extra length on restart")

Try:

3) Minor top-infill fusion issues

The Z face and X face recess show some “not merged” surface between the infill paths.

Fixes:

1. Increase top solid layers to 6
2. Increase infill overlap to 15%. (Default is 10%)
3. Lower top surface speed to 15–20 mm/s

4) First layer elephant-foot

The cube bases show a slight outward flare.

Cause: Bed temp too high or Z-offset slightly too low.

PETG fix:

1. First layer Z offset: +0.05 mm higher than PLA
2. Bed temp: 80°C → try 75°C

🎯 Ideal PETG Settings for Prusa (your target tuning)

SettingValue

Nozzle Temp

235–240°C

Bed Temp

75–80°C

Extrusion multiplier

0.98 to start

Layer height

0.2

Perimeter Speed

35–40 mm/s

Retraction

0.8–1.2 mm (Prusa direct)

Retraction Speed

40 mm/s

Top layers

6

Infill overlap

15%

Cooling

20–30% (PETG hates wind)

💡 Why your result is close to excellent

Your issues now are PETG-specific, not mechanical.

PLA prints perfect → PETG requires:

1. Less fan
2. Lower print temp
3. More controlled motion

You're at about 95% dialled in.

🔧 Quick Win Changes (in order of priority)

If you do ONLY these, prints will improve a lot:

1. Lower temp to 235°C
2. Retract 0.8 mm @ 40 mm/s
3. Fan at 25%
4. Top solid layers 6

🤘 You're very close!

Send me your:

1. Retraction settings
2. Nozzle temp
3. First layer Z offset

…and I’ll fine tune them to perfection."

If you look at the Dust shoe, the evidence shows it for itself, the prints are near perfect in my opinion, with one of the more difficult filaments to print.

What It Means If You Own One (Or Plan To)

Think of the X5A like a project car. It’s powerful, flexible, and attractive but you won’t drive it off the lot and win a race. It needs work. It needs tinkering. It needs constant attention.

If you’re someone who likes to:

1. tweak settings
2. tune mechanics
3. experiment with filaments
4. and don’t mind “wasted prints” while dialing it in

then the X5A can be a rewarding, albeit frustrating printer. It can produce usable, flexible, large-format prints with potential. Turns out that the Gifted TronXY has an issue with the Extruder, as it keeps getting blocked, so I will replace it with a E3D V6, then we know that is the new baseline to start from and no hardware issues. I figured out the ABS is find as the Teflon tube inside is stable, but when trying to print PETG. I will order a new a new print head with with a 100% heat break. that should solve the getting stuck in extruder as I think someone has damaged the Teflon tube in the heat break.

But the machine can produce a very credible working part in ABS, as can be seen with the Yellow attached dust cover for my CNC machine I am building

Lastly thanks ChatGPT because your workshop helper guided you to better print results, even though, like me you are not a professional, the print look professional.

Thanks for reading if you made it to here and I really hope this improves your 3D printer tuning experience