I was forced to resort to attaching an external clock source, as the internal oscillator isn't entirely stable, which was causing the picture to wobble. ...and, as long as I was attaching the 14.31818 MHz crystal oscillator, I figured I should try to do color.
That's just the ATtiny13a, a 14.31818 MHz crystal oscillator, five 1000 ohm resistors, and a power source. ...and a television. Can't do that without a T.V. The picture is much smoother in real life. Apparently we're seeing interference between the T.V.'s pixels and my digital camera's pixels. IRL there's just some annoying shit at the vertical edges of the color bars, but it's otherwise smooth and stable.
It took a while to figure out why it wasn't working at first. Seems you have to invert the polarity of the color signal for each successive scan line, because each scan line is 227.5 cycles of the color carrier. If you don't do that, the television just ignores the color, I guess because every time it tries to sync with the color burst, it finds itself to be entirely out of phase. Perhaps it assumes it better to not even try to display color in that case.
Interestingly, the image also shows up on my LCD television, which refused to display my previous attempts years ago. It even did so when I screwed up the timing of some of the scan lines, which makes me wonder if its entire problem with the signal had nothing to do with the timing accuracy but instead with the missing equalization pulses. Now it seems to try to display the signal no matter how much I screw it up. It'll even display the wobbly image generated when I use the ATtiny's internal oscillator, whereas the image from what I built years ago, despite using analog timers, didn't wobble at all.