Anet A8: Modded to a Point
The Anet A8 is a cheap, boxy, entry-level FDM printer. Mine keeps the original frame — and almost nothing else. It's the printer where "mod" stops meaning a plate and starts meaning a project.
What's actually on it
| Part | What it's running |
|---|---|
| Controller | Klipper, running on a BIGTREETECH SKR Mini E3 V3 |
| Stepper drivers | TMC2209 on all four axes, UART mode |
| Build plate | 220×220 mm PEI spring steel |
| Sensor | BLTouch — auto bed level, saved nozzle offset around 0.5 mm |
| Power | 350 W PSU (the stock one was a prayer) |
| Belts | Fiberglass, with upgraded tensioners |
| Camera | OpenSCAD-printed mount — the part that holds the camera is itself printed |
No enclosure, no resonance sensor, and no pretense that this is a fast machine. It's a 220×220×240 mm box that prints better than the box it shipped in.
The config is the mod
Klipper turns a printer into a configuration problem, which is exactly the kind of problem I like. The interesting bits:
- TMC2209 currents tuned per axis — extruder runs hottest, X/Y in the middle, Z lightest — all in stealthChop. Quiet enough to run in the same room the kids sleep in.
- Kinematics — cartesian, 300 mm/s max velocity, 3500 mm/s² acceleration. Aggressive for this frame; the fiberglass belts are the reason it holds up.
- BLTouch — single sample, 5 mm/s touch-down, z-offset saved into the config. "First layer" stopped being a daily negotiation.
The whole thing lives on the Raspberry Pi — the same box that runs the home lab — reachable from anywhere in the house over klipperpi.local through Moonraker and Fluidd.
The workflow
- Slice in Cura 5.11 with a profile I rebuilt by recovering the original printer's settings from its old G-code files. The printer predated my configs, but the G-code remembered what the stock profile was —
SETTING_3and friends are an archaeology site. - Print test cubes and the elephant-foot test. The local model looks at the photos and the logs and tells me whether the first layer is actually good, which is the only first-layer judge I trust.
- Watch live over the camera and Moonraker logs. The agent can babysit a print and report back — closer to what you'd want from a printer than the manual ever offered.
- Post-mortem on defects. When a print comes out wrong, the model works from photos + G-code + live logs to find which variable moved. Elephant foot, adhesion, layer shift — each has a paper trail.
Next on the bench: a small print-quality stats app that scores completed prints, so "is it getting better?" stops being a vibe. (OpenSCAD is how the camera mount started — see the OpenSCAD docs.)
Why bother
A few hundred dollars of printer, a few hundred more of parts, free firmware, and a few hundred hours of tinkering. The total still costs less than one good printer — and the knowledge is the actual product.