This could probably go into the Anki sticky, but it's more general than Anki specific, so I'm making a new thread.

Currently I use 2 fields: kanji (+okurigana) on the front, and kana + English translation (or picture) on the back. I use normal and reverse card (which makes 2 cards, one one way, another the other).

This method has a couple of problems.

1) It requires 2 card reviews, effectively doubling the amount of time spent on SRS. This also basically halves the amount of vocab I can put in at once without worrying about the reviews stacking too much (I seriously don't want to get to 300+ reviews a day as I did when working through RTK1 again).

2) The English and the reading are on the same side, which obviously gives away what word it is. This sort of screws up the recalling aspect (leaving only the Kanji to memory, not the word itself).

I've noticed that a lot of cards I see online have English on one side and both the Kanji and reading on the other. This seems weird to me, since it means that reading the Kanji is unable to be practised as the reading is right next to it.

So what's the best way to arrange vocab cards in an SRS? I'd ideally like to cut out having 2 cards for one word if possible, but am worried it might make me less able to read words when I see them.

I personally like having two cards for words so I can do kanji <> kana, which are two different recalls (but I have kanji>kana+English, and kana>kanji+English).
But one thing I read here that I wish I had done before (because having two cards is repetitive) was doing only recognition first (kanji>kana), and then doing production (kana>kanji) once the cards are mature (or even much later when you feel like it).

Of course it's easier to do with premade decks, but maybe you could suspend your production cards during the month, and then at the end of the month you add them as new, that'll be roughly the same.
But yeah, having two cards though self-reinforcing the words is repetitive, so more generally (as Nukemarine advocates in the Beginner thread I think) you would either do only production and read a lot on the side, or do only recognition (and if you don't need to produce anything, why bother, it's time saved). And that depends on your level/expectations with the language.