The alphabets that are created for nanoid seems to be a valid choice too: https://github.com/CyberAP/nanoid-dictionary

It's not strictly necessary to remove all the similar characters, as you can treat any of the similar inputs as being the same. eg. you could include 8 or B in the output, as long as you accepted 8 and B as being the same character you chose for output at input.

If you want to avoid the possibility of accidental profanity in your "random" strings, it is definitely worthwhile removing the vowels, although that may not apply to other languages than English!

Until someone in design or marketing changes the font on you.

Which is why I only mentioned lowercase l. 1 in some (especially serif) fonts looks like l and in some others (especially sans serif) like I. Definitely not okay in general.