a thing that remains fascinating to me: it is impossible to do picture-in-picture with analog circuitry. doesn't matter how clever you get, it's just not a thing. and that doesn't feel like it should be true, but it is true.
in the context of pre-digital TV production, all video processing gear "raced the beam." all video operates on a strict and unchanging timebase that waits for no one, so while two video signals can be out of phase, they can't be transmitting at different rates. it's always exactly one frame every 60th of a second, no matter what. as a result there is very, very little you can do to modify video in the analog domain. every camera, VTR, everything outputs the exact same signal at the exact same speed and there are zero devices I'm aware of that allow you to change any of that.
want to squish an image? impossible. expand it? impossible. you can't do these things because, as signal #1 is being drawn, signal #2 is being drawn at the exact same time. to squish #2 horizontally, you'd have to move the "pixels" from the right edge of the image inwards, and display them earlier than they would normally appear in that scanline. you can't! the camera hasn't shot them yet! they don't exist!
if you are used to thinking digitally, this is crazy talk. the smallest unit of time in digital video is the frame; pixels, lines, rows, are simply convenient ways to think about the contents of a frame. they are, technically, being sent sequentially just like analog, but because digital systems don't allow you to look at the bitstream until an entire frame has been collected, it doesn't matter; you need a whole frame before you can do anything. as a result, nearly all digital video effects require an entire frame to be collected before it can be manipulated, imposing some minimum amount of delay on an entire production.
but in analog, you can't do that. "analog frame buffer" isn't really a thing. i suppose you could pipe a signal into a delay line, but that would delay the entire video stream. if you timed it right, you could cause one signal to appear a frame or two later than another. but that's it; you can't "address" the contents of that delay line, you wouldn't be able to "tap" off of it and pick out the middle of the delayed frame.
i suppose, if you wanted to superimpose two TV images, and then move the top "layer" in relation to the bottom, you could conceivably delay or advance the phase of the second input by modulating its genlock signal; in other words, making camera #2 take its pictures later or earlier than camera #1. i'm not sure anything ever did this, and it would require a very precise processor that could slice off the unused nonsense portions of the image that normally appear in the overscan region.
that's it though; that's the only trick you would ever be able to do. otherwise, you're strictly limited to effects that involve altering an image in-place (e.g. hue shift, contrast/brightness), or switching or blending between two images, without changing their position or size or shape.
an interesting corollary to this: if you simply set an analog TV in front of camera #1, and pipe camera #2 into it, bam, you'll have your PIP - because the phosphor screen acts as a framebuffer.
this makes me think, actually, that an analog PIP COULD have been possible, by leveraging several technologies that existed from the earliest days of TV, and most likely this didn't happen only because nobody thought of doing it before digital circuitry made it pointless.
early cameras used videotubes, which functioned sort of like a 60 FPS xerox machine: photons from the camera lens struck a "target" - a surface coated in a compound that converted them into electrons. this left an electrostatic "image" on the surface of the target, which was then scanned by a CRT electron beam. the charged areas scattered electrons that were collected by a pickup, and in this way the image was converted back to an electrical signal.
okay, so: imagine if you instead charged the target by sweeping another CRT beam across it. one electron gun deposits an image, then a second one sputters that image off the target. a special quality of these compounds is that their electron wells did not dissipate instantly when scanned, which is why bright objects in pre-80s live TV shows left trails. if the input CRT were massively amplified, it could leave an electron "picture" so strong that it would survive multiple scans. this would make the target behave like a persistent framebuffer, thus allowing the two guns to run at completely different rates and phases.
now, imagine if the target was only 1/8th the size, and the depositing gun scaled down along with it. instead of forming an electrostatic image that's, say, 4" corner to corner, it could instead make one that's only 1/2" across. now: scan it the same way, with the same "reader" gun, scanning the same area. it will mostly pick up nothing, except where it traces over the target - where it picks up a much tinier version of the second image. superimpose that, and you have your PIP.
i wish this has been invented, but if it had, there would be about three working ones left, all in museums in germany.
