This is because the ADC (analog to digital converter) right after can only handle so many bits of data (like 12-16ish in consumer cameras). You want to “center” the data “spread” so when the “ends” get cut off, it’s not so bad. Adjusting the ISO moves this spread around. In addition, even if you had an infinite bitrate ADC, noise gets added between the gain circuit and the ADC so you want to raise the base signal above the “noise floor” before it gets to the ADC.
Gain is not great — it amplifies noise too. You want as low ISO as possible (lowest gain), but the goal is not actually to lower gain; your goal is to change the environment so you can use a lower gain. If you have the choice between keeping the lights off and using higher ISO versus turning on the lights and using a lower ISO, the latter will always have less noise.
Most photo cameras have one gain circuit that has to cover both dark and light scenes. Some cameras like a Sony FX line actually have two gain circuits connected to each photosite and you can choose, with one gain circuit optimized for darker scenes and the other optimized for brighter scenes. ARRI digital cinema line cameras have both and both are actually running at the same time (!).
...or integration time.
> … also while fact checking this comment, I stumbled across Canon's KB stating to use as low an ISO as possible, which the video rails against. They should talk to Canon, I guess?
Isn't ISO last the same as setting it as low as possible? Obviously it's always set to something, so I thought 'doing it last' means start with it low, set exposure & shutter, increase as necessary?
(Shutter speed being dictated by subject and availability of tripod, essentially it's just exposure & ISO which becomes about how much light there is and how it's distributed, I suppose.)
I'm not really into photography though, so perhaps that's all nonsense/misunderstanding.