After one more day of fiddling with trims and bleeds, I had it published on Amazaon: https://www.amazon.com/Dorky-Dads-Knock-Book-Vegetables/dp/B...
The prompting of DALL-E, like any tool, took a bit of learning. My take-aways using it were: - The four variations initially generated are rarely enough. The variations can be quite diverse, and a better rendering can often be found just by generating again. After three times (12 variations), the returns get diminishing and it's usually better to either tweak the prompt phrasing, or drill down into the best so-far to get variations of it. - The ability to select and replace a section allows for a great deal of control. Usually the scene will end up more smoothly connected and related if you try to specify everything in the original prompt. However, where this is not understood properly by DAL-E, selecting an area and then using a new prompt in that sub-region works well. - DAL-E on occasion renders objects that you don't specify, ruining an illustration that would otherwise be perfect. There's unfortunately no "erase" option, but if you specify "empty" on the prompt it works to clear the object most of the time.
The capabilities are already amazing; it'll be even quicker in helping creativity with a few more tools on top of it, like being able to move objects into position. At any rate, much thanks to the OpenAI team for what I've already been able to do!