I clocked an SN74HC174 hex flip flop with the 4FSC clock signal, connected the flip flops in series with each other, and connected the pixel switch to the first input. Using the 5th flip flop output gives a delayed switch signal which is neither early nor late. So there is no more black band before or after the overlay.
Although some haloing left and right of the white bands is still present, I don’t think this is caused by the switch. The image is quite noisy, you can see a noise artefact in the magenta band. Some of the connections on the breadboard are a bit loose and the circuit is quite finnicky. I have to fiddle with it for several minutes to produce an output.
I now have a working OSD prototype, which works for NTSC only. I still haven’t come up with a method that supports both NTSC and PAL. Also, the circuit is quite complex: 7 ICs and supporting passive components.
The current draw is 143mA at 5V, which includes the STM32 discovery board with its on-board debugger, a much more economical usage of power than my previous attempt which drew 550mA.
I noticed a few things. Although the last 2 white bands are 2 pixels and 1 pixel wide, the 2-pixel band looks about 3 times as wide as the 1-pixel band. Also, colours don’t really show up in less than 2 pixels. I tried rendering the white bands in various colours, and the 1-pixel band was still mostly white. This seems to be a limitation of the AD724, as previously when using digital video ASICs I was still able to render single pixels in colour.