Is it really "mad science"? The new Image Signal Processor (ISP) has a pretty good chance of being as large, [in terms of transitors], as the CPU + GPU cores (minus cache and other stuff) of the first three (maybe four) iPhones. If you throw 10's of millions of transistors at some fixed-function logic (does just one calculation), then specific things can go quite fast (e.g., disk controllers like the T2 encrypt and decrypt in real time with no read/write overhead).... All those pics will have to be written to RAM until the magic moment that the user presses the record button ... Never mind the “mad science” of stitching those images together afterwards, dealing with issues such as parallax error and so on. This type of post-processing (esp. with a 10-bit vs. an 8-bit pipeline) is non-trivial, esp. in a small embedded application like this one, with a limited battery budget and the consumer expectation that lots of pics can be taken quickly.
[The ISP is] probably in the same range with the Neural Engine – about as big a budget as the CPU+GPU logic section nine years ago but only covering a smaller computation problem.
The faster CPU+GPU in the A13 makes the camera GUI run better, but the real issue is the millions of transistors that Apple throws at camera sensor processing that do absolutely nothing when you are not using the camera sensor(s). The total transistor budget is so large that there are billions more to go around, and they work hard at being able to put unused subsections of the system on a chip to sleep when not in use.
The A13 has probably at least all the specialized camera "digital horsepower" as higher-end DSLR or video cameras, as a 'sidecar' to the main section of the chip. I'm sure there is some processing portion that flows over to the CPU+GPU, but that would be in the 'additional effects' zone.
It is more about how to capture stills from a video camera at this point (even for higher-end "mirrorless" DSLRs.)
Selling tens of millions of A13's per years helps Apple be several generations ahead in process tech vs. the older stuff the camera makers sit on to save money (because they don't generate the revenues through volume), so Apple ends up doing more with more (more transistors, do more work, get more done).