You can always use XKCD's password methodology ;)
https://www.eff.org/deeplinks/2016/07/new-wordlists-random-p...
His "solution" is to use shorter passwords. The XKCD method is good if you add separators, padding, etc; as expressed featured on xkpasswd.net
I highly recommend generating a password and then adding something unique to it.
For instance, a password I might generate would be:
$66=mine=BODY=spot=STOP=23$-d1j1t
It's memorable enough, and I highly doubt it's easily crackable. Certainly no less than 'tlpw2m'.
However, as you point out, it provides no argument against passphrases, aside from referring to them as a "trick". I still don't know how people look at the XKCD explanation (where Randall Munroe actually does a pretty good job of correctly and succinctly detailing the strength of the two password styles [1]), and call it a trick. The only trick is that your mind has an easier time remembering passphrases than it does remembering a similar strength random string.
[1] Specifically, Randall already assumes that the cracker knows that the password is a passphrase, and has the 1000-word list it was picked from.
You still definitely have to accept at least four words actually randomly generated (this is important, else the scheme falls apart horribly). Accepting 6 random words is pretty secure IMHO.