I'm really into analog photography and I can think of several products that people (including myself) would pay for but that it's just not economically feasible for a big company to hire a bunch of people and pay them wages, health insurance, and 401ks to make, on top of the actual cost of the product. I'm looking to gradually put together a workshop to try and make some of them, and if I do them, I might as well sell them. Even if it's $100 out the door for a $5 piece of metal, there's actually a market for that in terms of hobby income, it's just not a market that will sustain full-time employees and mass production.
Between 3d printing, stamping, a CNC mill and lathe, casting, a laser cutter, and a vacuum oven, you can really do a tremendous amount of stuff in your garage, especially if you are willing to leverage these tools together. 3D print a part and then cast it in a durable metal, machine it to clean it up. CNC mill yourself a stamping die. Use the vacuum oven to cure things, or dehydrate your filament, etc. Like on paper that's pretty much a tool-and-die shop, given sufficient effort you can make things that will let you make whatever else you want - much like chemistry you're never more than a couple tools away from the thing you want, you just have to make the thing that will let you make the thing you want.
Optical lithography is not that hard either as long as you're not working at semiconductor scale. There's that kid that is making chips in his garage over in the UK or something. But you can use that as a manufacturing technique for other stuff. Or use resist etching like for PCBs.
In a lot of cases, really the only limit is when you bump into something that's restricted or too hazardous to keep around even if it's unrestricted. Like boy, mercury intensifier works great but... I like my nervous system the way it is. Even selenium intensifier or pyrocat developer are pretty yikes in terms of the MSDS, very much a "better have a fume hood" thing, or do it outside (in a daylight tank).
Incidentally, but, my most insane "I'd love to do that in my garage" is custom lenses. I know the accuracy is probably just not there compared to what you'd realistically need for good results, but it really seems like single-point diamond turning should be something that is achievable with a high-end setup (say $25k) in this era of solid CNC mill or lathe setups for half that price. Maybe it's something you could build out of a CNC setup but again, is it accurate enough to make it worth it (not sure of the tolerances required, at least 1/1000th, probably 1/10,000 is better, 1/100k or finer should do it, which I guess isn't too far out of what you can do with a lathe, it's all just end work and you have to be precisely optically centered and aligned). Coating is one where you'd need the vacuum oven for sure, assuming it wasn't too toxic (iirc coatings are fluoride based). Growing optics-grade fluorite or calcite crystals also might be possible for lens blanks (although again, maybe too nasty) - or glass casting too. You'd need an optical bench too of course.
There is definitely a market for that I think - all-fluorite lenses are excellent for wide-spectrum photography (UV to IR in the same lens without focus shift - see the Coastal Optics lenses f.ex) and people (companies) pay big bucks for those, like $50k is entry-level for something in that class if you go out and buy it new. And with single-point you can make aspheric lenses as easily as spheric, so you could do all-aspheric designs that aren't commercially viable for mass-market lenses... as well as super-high-quality repros of classic lenses that are obscure or just classics. People would pay for a neo-retro Hypergon or modern takes on sonnar/heliar/etc if you could produce good results. Or you could make tools that let you do it in the traditional fashion with spheric surfaces.
http://www.savazzi.net/photography/coastalopt_60.html
https://www.mir.com.my/rb/photography/companies/nikon/nikkor...
http://www.company7.com/nikon/lens/0105f4.5uv.html
https://jmcscientificconsulting.com/testing/asahi-pentax-ult...
Anyway though another place where some of this ends is "too difficult to make at home"... that's actually a more interesting question in some ways, a dedicated hobbyist with tool-and-diemaker level machinist skills fluency with modern CNC and 3D printing and the techniques enabled by that, and enough knowledge of chemistry/optical/electrical/RF to get yourself in trouble, can of course make an enormous amount of stuff. But things like single-crystal turbine blades for micro-turbine designs are difficult and without the "real deal" you are leaving performance on the table. In some cases (again, things that are too toxic to handle, or illegal) you do basically hit a wall, not all projects scale down to the hobbyist level.
Bikepacking bags (and to some extent ultra-light hiking) are good examples that come to mind where people started things as hobby businesses in their garage and they quickly expanded because of demand.
A trivial electronic gadget to turn an oil cooler pump on and off is a good start. Electronic assembly isn't noisy or stinky. At most you might need to bend sheet metal for a case.
Guitar players love audio processing gadgets they can turn on and off with their toes. Make something that can be adjusted to give them a unique sound. Distortion is easy to layer on.
This seems like the sort of work that was spun off from some kind of existing business relationship.