The fastest high speed camera that I am aware of is the Hadland Imacon 200. This camera is capable of recording 16 sequential frames at a rate of (approximately) 200 Mfps. That means it takes one picture every 5 ns - during which time light only travels 1.5 m! This is a "true" camera in the sense that it can acquire a burst of 2D images; not to be confused with the sampling cameras that require accurately pulsed laser sources to build up an image of a repeating event, this is the kind of camera that will record a one-time event, like a bullet hitting a target. Note that a fast bullet (2000 m/s) will travel 10 µm in 5 ns - less than the width of a human hair. It is indeed about as fast as you ever need to go to film macroscopic one-time phenomena.
I found the following blurb on the Photonics Spectra website
The Imacon 200 offers a simultaneous framing and streak recording option for analysis of complex, ultrafast transient events such as laser ablation. This digital imaging system from DRS Hadland Ltd. has a patented multichannel beamsplitter and a retriggerable CCD chip that extends the number of discrete images captured to 16. One or two streak channels can be incorporated into the camera. The Imacon 200 is capable of data acquisition rates of up to 200 million fps when it is referenced to a 200-MHz quartz crystal clock.
A more detailed description can be found on the vision-systems.com website
The key to this system is the image intensifier that amplifies the signal: for such short frame rates, it is hard to get sufficient light from the sample. One typically uses a flash gun that is triggered just before the event of interest, then fires the camera during the event. It is mind boggling to think that a 1 µs flash (extremely fast... most consumer flash guns are 1/1000 to 1/20,000 of a second, so 50 - 1000x slower) has to be "turned on, then wait a bit, then take the pictures". It reminds you how crazy fast this is.
A slightly older series of Imacon cameras had a single phosphor, and the electrons from this were accelerated and deflected electronically. But such phosphors are not fast enough at these rates, which is why they have gone to the optical beamsplitter followed by image intensifiers (MCP - multi-channel plates).
(image from the above-linked article on vision-systems.com)
Older, slower film cameras with stationary film (like the C4 used to image the first atomic bomb tests) use a rotating mirror with a pair of relay lenses (one per frame) to get to their frame rate.
The insides of such a camera had to be evacuated (or at least depressurized) in order to get the mirror up to speed and prevent pressure waves from distorting the image. You can see a description here - this is also the source of the above picture. The film used in these cameras is not "off the shelf", because it has a slight curve in it in order to accommodate the optical path. It therefore was cut from huge rolls of (ordinary, wide) film used for aerial photography - the only film stock wide enough to permit the cutting of an arced segment. After a single shot exposure, you had to develop the film... and had to hope all the triggering happened in the right sequence!