Most sound cards are AC coupled - to build a laser scanner using sound card control you need to bypass the DC blocking caps. I have no idea how he plans to overcome that with an integrated sound card (people normally use a modified USB sound card for this). Along the same lines, what about feedback? Are the mirrors open loop? I've never seen an open loop laser scanner with any kind of reasonable performance. I've seen people build their own galvos, but still using closed loop control. Ah - I've just watched their demo video. That thing is *slow*. I still have no idea how they overcome the DC blocking issue though, considering the frequencies are definitely in the 1Hz range there (vs. more like 5-10kHz for a real galvo system). AM modulation with a cheapie detector, maybe?
The design has the same problem that extrusion based printers have: high X, Y, Z resolution, but a large "nozzle diameter" (or beam diameter in this case). You can't collimate a laser beam into a very thin beam (it won't stay thin for long, since the divergence becomes more pronounced the smaller you try to collimate the beam). And with varying distance between the scanner and the current resin level, you can't just use fixed focus. It doesn't matter that you can move your beam by 1um if it's 1mm in diameter. It also doesn't matter that you can move your Z axis by 1um if the beam is curing a 0.5mm deep resin layer.
Galvos move a beam to a given angle - to turn that into X,Y coordinates on the build volume, you need to use trig. That means you need to carefully calibrate the distance to the object or else you'll end up with trapezoid prints. Also, there are a bajillion variables here that would require calibration in order to build dimensionally accurate prints, and too many things that are likely to be unstable over time. Extrusion based 3D printers use belts and steppers that are very repeatable. Good photolithographic printers use constant distance to the print and/or a sturdy framework that provides repeatability. I don't think a simple jury-rigged design like this can manage that, which means at best it'll be good for decorative parts, not parts requiring accurate dimensions.
With uncontrolled Z feed, there's a limit to how complex a layer can be before you run out of time to print it. And you waste time on simple layers. All of their test objects seem to be very simple. Also, how do they turn the beam off to move between disjoin sections in a slice? Or do they just move it quickly and hope for the best? Their "Dave" print seems to show that they cannot print layers with disjoint parts.
This does feel way too cheap. I'm sure it'll print, but it's going to be at least as finicky as an extrusion based printer, for the same quality, if not more, and that's just for decorative parts. When I built an Ultimaker my very first print (with zero tuning) was crap by its standards, but it already looked better than the prints shown on that kickstarter page.