A sitting KAWS figure I downloaded off the internet [16] came off the Mars 5 wearing a thick crust of over-cured resin the designer never drew, welded to the build plate hard enough that getting it loose took real force and a running hope it would not crack before the plate let go. The exposure was fine. §M4a had done its job. What I had gotten wrong was everything holding the model up, which on that print was nothing at all.
Printing directly on the plate
After calibration I did the obvious-seeming thing. I placed models flat on the build plate, no supports, base touching the plate directly, and hit print.
Printing that way does something specific. Remember from §M4a that the first layers of any print get the 30-second bottom exposure instead of the normal 2.6 s, so the print welds to the plate. When the model itself sits on the plate, the model's own first layers get that treatment. Over-curing makes those layers wider, denser, and harder than intended. The light bleeds sideways, layer after layer fuses into a dense slab, and the bottom few millimeters of the model turn into something the designer never drew (the community calls the flared version of this "elephant's foot"). On top of that, the whole footprint of the model is now welded to the plate, a large surface area, fully cured at high exposure, stuck flat on aluminum.
The print itself worked, in the sense that a KAWS-shaped object existed at the end. Everything was just stuck. Scraping off a print you spent hours on is a good teacher.
What supports are for
I did some reading afterward and had to update my mental model. I had thought supports were for overhangs, basically a necessary evil. They're actually doing three jobs.
- They lift the model off the plate. The 30-second layers all land in the support raft, which gets cut away. The model itself is built entirely at the calibrated 2.6 s exposure, so its dimensions stay honest from the first real layer. This is exactly why §M4a could say the bottom-exposure setting "doesn't affect part quality" in my workflow. The supports are what make that true.
- They make removal a non-event. You pop the raft off the plate and clip the model free, instead of prying the model itself off aluminum.
- They let you orient the model for quality instead of stability. This one turned out to matter most.
Orientation, the part nobody tells you about
Once the model hangs from supports, you can print it at any angle. Two rules came out of my testing.
Put the faces you care about facing up, or at least away from the supports. Support tips touch the model, and every touch point leaves a small dimple when you clip it off. On a functional surface a dimple might not matter. On a face, a curve, or any surface people will look at, it does. So the important details go up, and the supports attach to the back or bottom where nobody looks.
Tilt the model, somewhere around 45°. Layer height doesn't change when you tilt, so the layers are never thinner. What changes is how a given surface gets sliced. A surface lying nearly flat in the build is crossed by only a few layers and each one leaves a wide step, which is the stair-stepping you see across the top of a dome printed flat-side-down. Tilt that dome and the same surface is crossed by many more layers, so each step takes a smaller bite out of the curve. Tilting also shrinks the cross-sectional area of every layer, which lowers the peel force as the layer separates from the film, standard practice and one more reason tilted prints fail less.
What supports cost
Supports aren't free, and I want the disadvantages on the record too.
- Resin. Supports are printed from the same resin as the model, so a heavily supported plate yields fewer models per bottle.
- Dimples. Every support tip leaves its mark. For pieces where smoothness or aesthetics are the point, support placement becomes careful, slow work, and a badly placed tip can ruin the surface that mattered.
There's also a real exception to "always use supports." For some functional parts, printing directly on the plate is a feature, not a mistake. The over-cured bottom becomes a strong, rigid base, and if the part is a flat structural piece, that's exactly what you want. In this project the deliberate plate-direct prints have been the calibration-cone plates themselves and some of the flexure bodies [builder, 2026-08-06]. The flexure bodies are the uncomfortable case and I would rather flag it than smooth it over. §M4a called out flexure hinges a few hundred micrometers thin as exactly the features that live or die on dimensional accuracy, and nothing here measures what a 30-second bottom exposure does to a flexure body's first layers. Until something does, treat the plate-direct call on those parts as a working choice and not a validated one. Display parts are the opposite case. Definition and surface quality rule, so they get tilted, supported, and accepted as slower and more wasteful to print.
Islands
Supports also introduced me to a failure mode I didn't know existed. Parts of the KAWS figure's head simply didn't print on a couple of attempts. Wrong orientation plus wrong support placement had created what resin printing calls an island, a region of a layer with nothing below it (not model, not support) to build from. The printer exposes that region anyway, but the cured resin has nothing to attach to, so it stays on the film or floats off into the vat, and the layers above it inherit the hole. The result is a print that's fine everywhere except the missing chunk, which on a figure tends to be somewhere very visible, like the head.
The fix is the same discipline as everything else in this chapter: orient first, then support, then run the slicer's island check before committing the plate. Most slicers will highlight unsupported islands if you ask. I learned to ask.
Current state of the printing pipeline
By the end of this stretch the workflow had settled into the shape it still has: calibrate exposure per resin (§M4a), orient important faces up, tilt near 45°, support from the non-critical side, check for islands, print, clip, cure. The KAWS figure eventually came out clean, thick-base-free, with its face intact. You can inspect the model in the chapter's 3D viewer (stl_viewer.html). Drag the STL in, and see [16] for where to get it, since it isn't mine to redistribute.
That print was the qualification run. The next print was the real test of all of it at once, a much less forgiving model with fine details in every direction. That print, the Evangelion, is §M4c.