Not only did the CGA have composite out, but by playing with the signal timings it was possible for a CGA to output more colors than the four nasty "CGA palette" colors it's known for (no matter which palette you use, shit's nasty) to a composite monitor.
The demoscene prod "8088MPH" demonstrates this to spectacular effect.
It was also very simple to just use the 640x480 graphics mode on a composite display. US NTSC color cycles 160 times in that display area, which yields 4 pixels per cycle.
That's a 16 color, any color any pixel display 160x200.
Some PC games offered this option. It should have been in the CGA spec. For memory reasons, 1 and 2 bits per pixel were common. However, the same amount of RAM offers the full color set at a reasonable, if modest, resolution. (For the period)
16 colors on a 160x200 display was considered "nice" at that time, and having it be official would have improved early PC gaming considerably.
The drawback to this mode is it looks like ass on anything but a composite display -- which not a lot of people actually hooked up their CGAs to.
There was also a hack to set the display to text mode and reprogram the character height to fit 100 characters vertically on the screen, then use chopped-off block graphics characters to yield, in effect, a 160x100, 16-color, any color any monitor (composite or RGB) pseudo-graphics mode that compared favorably to, say, the 128x48 mono pseudo-graphics mode of the TRS-80. More PC games made use of this mode; one of the more notable recent ones is Paku Paku, a (remarkably good) Pac-Man clone.
Yes, all tradeoffs. Of course, it was not too difficult to have multiple sets of art.
I like 160x100 presentations and thought they were a good use of the CGA personally. A lot of game can be done at that resolution.
Really, the thing for IBM to have done was 16 color graphics of some kind. 160x200 was the obvious choice in that it would have worked with their memory scan scheme.
The demoscene prod "8088MPH" demonstrates this to spectacular effect.