To analyse a failed 3D print from a photo, start by working out when it failed, not what it looks like. A single photo carries more diagnostic information than most people use, but only if you read it in the right order. First-layer problems — not sticking, warping at the base, elephant’s foot — are almost always bed-related: levelling, Z-offset, surface cleanliness or bed temperature. Mid-print failures such as layer shifting, spaghetti and delamination are usually mechanical or thermal. Surface defects — stringing, blobs, rough overhangs, ringing — are nearly always slicer settings. Identifying the failure phase from the photo cuts your diagnosis roughly in half before you’ve looked at a single specific symptom.
That phase-first heuristic is the backbone of this guide. Below, each common FDM defect is broken down by its visual signature in a photo, the mechanism that causes it, the fix in the order that actually works, and the photo-traps that routinely send people chasing the wrong problem. If you’d rather hand the image straight to a vision model, our Diagnose tool identifies defects from a photo and returns concrete slicer changes — but it helps to understand what it’s reading.
Read the phase before the symptom
There are only six root causes behind the vast majority of FDM failures: poor bed adhesion, wrong temperature, wet filament, a nozzle clog, a mechanical motion fault, or a mismatched slicer profile. Each one tends to express itself in a particular phase of the print, which is why the phase triage is so powerful.
| Failure phase | What you’ll see in the photo | Where to look first |
|---|---|---|
| First layer | Not sticking, base warp, elephant’s foot, strands in air from the start | Z-offset, bed cleanliness, bed temperature, levelling |
| Mid-print | Layer shift, spaghetti/bird’s nest, delamination, cracks | Mechanical (belts, skipping steppers) and thermal |
| Surface | Stringing, blobs, rough overhangs, ringing | Slicer settings: temperature, retraction, cooling, speed |
Stringing: fine wisps between features
Stringing happens when molten filament leaks from the nozzle as the head travels between sections. In a photo it reads as hair-thin wisps spanning gaps — a surface-phase, slicer-settings problem. Tune it in this order: temperature → travel speed → retraction distance → retraction speed.
Start by lowering the nozzle 5–10 °C: a cooler melt is more viscous and weeps less during travel. For reference, Prusa’s PLA profile runs 215 °C first layer / 210 °C after, and PETG runs 230 °C / 240 °C. Then set retraction sensibly for your kinematics — direct-drive extruders typically need 0.5–1.5 mm, Bowden setups 3–6 mm depending on tube length. Increase in 0.5 mm increments until the strings disappear. The correct value is the minimum distance that eliminates stringing, not the maximum you can set.
Two common mistakes. First, over-cranking retraction: pushing it to 10 mm can kill stringing but adds roughly 30% to print time and wears the extruder gear. On a Bowden setup, retraction speeds above 60 mm/s let the gear strip the filament instead of pulling it back — the stepper skips, actual retraction drops to zero, and you’ll hear clicking. Second, the photo-trap: wet filament mimics stringing. Damp PLA pops and sizzles at the nozzle, firing tiny plastic droplets that look exactly like strings. If you hear crackling, dry the spool before touching retraction.
Under-extrusion: gaps, thin walls, swiss-cheese tops
In a photo under-extrusion shows as gaps between perimeters, walls thinner than they should be, and holes in top surfaces. The reflex is to raise the slicer flow rate — and that’s usually wrong. The most common actual cause is a partial nozzle clog that no temperature change will fix. A partial clog extrudes about 80% of commanded volume: enough to keep printing, not enough to be strong. Unlike a full clog (zero extrusion), it produces weak, rough prints that complete.
Use this diagnostic split: if walls are consistently thinner than nozzle diameter, you have a genuine flow issue — raise flow by the ratio (target ÷ measured). If walls are inconsistent, you have a clog or feeder slip, not a flow problem. Flow rate should sit between 95–105%; if you need 110% or more, the printer is broken, not miscalibrated. Dialling flow to 110% is a band-aid that masks the mechanical fault and produces inconsistent results across the print.
There’s also a melt-rate ceiling people hit without realising. A standard 0.4 mm brass nozzle on a stock hotend handles roughly 10–12 mm³/s volumetric flow; push past that on dense infill and perimeters and you’ll under-extrude from sheer speed. Slow the outer wall to 40–50 mm/s and infill to 80–100 mm/s as a starting point.
To clear a partial clog, a cold pull works: heat to 170 °C for PLA, cool to 90 °C, then pull the filament firmly. If you clean the tip, insert a 0.35 mm cleaning needle at printing temperature — never run a drill bit through a hot nozzle, as it damages the interior. And find the cause: a new nozzle fitted without addressing dirty filament can clog again within about ten hours.
Bed adhesion and spaghetti: the bird’s nest
Spaghetti is the most dramatic failure in any photo — the nozzle keeps following its path but extrudes strands into thin air. It’s almost always a downstream symptom of one upstream cause: the first layer never actually bonded. One analysis puts detachment at around 80% of these cases. So treat spaghetti as a first-layer problem even though the mess appears mid-print.
Among adhesion failures, an incorrect Z-offset (nozzle too high) accounts for roughly 28%, and a dirty or greasy build plate about 18%. Both are readable from a photo of the first layer. The tell: a good first layer looks like flat ribbons, not round strings. A correct Z-offset compresses the first layer so extrusion width is roughly 110–125% of nozzle diameter — the lines merge rather than sit as individual strands. Fine-tune in 0.02 mm increments until they fuse.
For cleanliness, oils from fingers, dust and filament residue all cut adhesion. Wipe with isopropyl alcohol (IPA) before every print; for stubborn residue, warm water and dish soap. One important exception to flag: a QIDI Cool Plate should be wiped dry or slightly damp and washed with warm water and mild detergent for heavy dirt — never IPA or acetone, which damage the coating. Match your cleaner to your surface.
If you can catch spaghetti risk before you print, you never see the bird’s nest. Our G-code pre-flight checker flags first-layer and profile mismatches up front, and the pre-flight tool explainer covers what the 40+ rules look for.
Warping and elephant’s foot: the flared base
Elephant’s foot is a specific, readable defect: the bottom few layers bulge outward past the rest of the model. In a photo, look for a clear width difference at the base-to-wall transition — a flare, not general unevenness. If the entire first layer looks uniformly over-squished rather than flared, that’s a Z-offset problem, not elephant’s foot. Getting that distinction right from the photo saves you adjusting the wrong setting.
The cause is usually the nozzle sitting too close to the bed, the bed running too hot, or an over-extruded first layer. When the bed is too hot the filament is almost liquid as it’s laid down; the soft layer can’t set and gets pushed outward as the head passes over it. Fix order: raise Z-offset slightly (0.05 mm increments until the base matches the rest), drop the bed temperature after the first layer has bonded, and apply elephant’s foot compensation in the slicer. Reference bed temperatures: PLA and TPU around 55–65 °C; high-temperature materials like ABS 100–110 °C.
Delamination and cracking: layers that never welded
Delamination shows in a photo as visible splits between layers or cracks running up a wall, most often on ABS. The mechanism is simple: two layers never fused into one weld, and the shrinkage force of cooling plastic pulled them apart. The most common cause across materials is print temperature too low — the plastic exits too cold to reflow the layer below. Increase temperature by 5 °C and retest.
The second cause is cooling, and this is where copying a PLA profile to another material goes wrong. Cooling is material-dependent:
| Material | Part cooling fan | Note |
|---|---|---|
| PLA | 100% | Aggressive cooling is fine |
| PETG | 40% (range 40–70%) | Never 100% |
| ABS | 0–20% | Near-zero; enclosure helps |
| Nylon | 0% | Zero, in an enclosed chamber |
ABS delaminates by physics: its large thermal expansion coefficient means layers contract significantly as they cool, and a steep thermal gradient tears them apart. Minimise the gradient with an enclosure and minimal fan, and never carry PLA cooling settings across to it.
What a photo can and can’t tell you
Photo-based analysis is the most intuitive way in, especially for less experienced users, because the visual signature of a failure is the fastest route to its category. Narrow vision models are genuinely strong on narrow tasks — tuned CNNs have classified single defects like cracks or pores at close to 100% accuracy on controlled datasets, and detection models have found more than 90% of cracks and pores in test images. The honest caveat: those are lab datasets, not messy consumer photos under kitchen lighting, so treat such figures as a ceiling rather than a promise.
General-purpose vision models now power consumer tools, and the better ones add context beyond the image — the 3MF, G-code clues, printer settings and known fixes — because that’s what separates “that looks like stringing” from “your retraction is 1 mm on a Bowden tube, raise it”. Good tools also abstract their confidence: academic work found numeric probabilities confuse users, so bands like Highly Likely (75–100%) and Unlikely (25–50%) communicate certainty better. For a fuller breakdown of what holds up in practice, see our guide to the best AI tool to analyse failed prints and the how-to on diagnosing from a photo.
Take a photo worth analysing
- Light it flat and even. Side lighting exaggerates ringing and layer lines; diffuse light shows true surface texture.
- Shoot the transition. For base defects, frame the base-to-wall edge so a flare is distinguishable from an over-squished first layer.
- Get one macro and one wide shot. The macro shows texture (stringing vs droplets); the wide shot shows whether the failure is localised or global.
- Note the material and nozzle. A photo plus “PETG, 0.4 mm, Bowden” is worth far more than the photo alone.
Running more than one printer?
If you’re printing to sell or across a fleet, catching failures after the fact isn’t enough — you want them caught live and costed. Printer Farm runs the whole job end to end: a web order from WooCommerce, Shopify or your own custom website (connected through the farm’s shop API — a token, no plugin) comes in, stock is used first, one click raises a works order for the shortfall, cloud slicing in the browser uses the real OrcaSlicer engine, and the queue dispatches each job to the first free compatible printer with the right filament loaded. Cameras with AI failure detection watch each print and alert you, good prints bank into stock with a true cost per part, and despatch raises a numbered delivery note. It needs a small agent on a Windows or Linux computer that stays on; your files stay on your own PC, and slicing runs there through ATN Slicer (or on our servers on Workshop and above). It’s in open beta with a free 14-day Workshop trial — start the trial here, no card.
FAQ
Can AI accurately diagnose a failed 3D print from a single photo?
For categorising the defect type, yes — visual signatures are distinctive and both narrow and general-purpose vision models do this well. Accuracy improves sharply when you add context: the material, nozzle size, extruder type and ideally the 3MF or slicer settings. A photo tells you what failed; the context tells you why and which setting to change.
Why does my print look like it’s stringing when the filament is dry?
Check for crackling at the nozzle. Wet filament produces micro-steam explosions that eject tiny droplets which photograph almost identically to retraction stringing. If the spool is damp, dry it before adjusting temperature or retraction — otherwise you’ll tune settings that were never the problem.
Should I just raise the flow rate to fix under-extrusion?
Only if walls are consistently too thin. If the walls are inconsistent, the cause is a partial clog or feeder slip, and raising flow masks it. Keep flow within 95–105%; needing 110%+ means something mechanical is wrong, not miscalibrated.
Why do my ABS prints crack but my PLA prints don’t?
ABS has a much larger thermal expansion coefficient, so cooling layers contract hard and a steep thermal gradient tears them apart. Print it hotter, run near-zero part cooling (0–20%), and use an enclosure. Never copy PLA’s 100% fan setting to ABS.
Related: How to Start a 3D Print Farm: A Step-by-Step Plan From One Printer to a Profitable Fleet