Related: Here's a C program that draws random dungeons sort of like you use in a roguelike dungeon crawler:

    #include <time.h> //  Robert Nystrom
    #include <stdio.h> // @munificentbob
    #include <stdlib.h> //     for Ginny
    #define  r return    //    2008-2019
    #define  l(a, b, c, d) for (i y=a;y\
    <b; y++) for (int x = c; x < d; x++)
    typedef int i;const i H=40;const i W
    =80;i m[40][80];i g(i x){r rand()%x;
    }void cave(i s){i w=g(10)+5;i h=g(6)
    +3;i t=g(W-w-2)+1;i u=g(H-h-2)+1;l(u
    -1,u+h+2,t-1            ,t+w+2)if(m[
    y][x]=='.'                  )r;i d=0
    ;i e,f        ;if(!s){l(      u-1,u+
    h+2,t-    1,t+w+2){i s=x<t     ||x>t
    +w;i    t=y<u||           y>    u+h;
    if(s    ^t&&              m[      y]
    [x    ]=='#'    ){d++;    if(g    (d
    )     ==0)    e=x,f=y;    }}if    (d
    ==    0)r;    }l(u-1,u    +h+2    ,t
    -1    ,t+w    +2){i s=    x< t    ||
    x>    t+w;    i t= y<u    ||y>    u+
    h;    m[y]      [x]= s    &&t?   '!'
    :s^t    ?'#'                    :'.'
    ;}if    (d>0)m                  [f][
    e]=g(2    )?'\'':'+';for(i j=0;j<(s?
    1:g(6)        +1);j++)m[g(h)+u][g(w)
    +t]=s?'@'                 :g(4) ==0?
    '$':65+g(62)              ;}i main(i
    argc, const char* argv[]) {srand((i)
    time(NULL));l(0, H, 0,W)m[y][x]=' ';
    for(i j=0;j<1000;j++)cave(j==0);l(0,
    H,0,W) {i c=m[y][x]; putchar(c=='!'?
    '#':c);if(x==W-1)printf("\n");}r 0;}

Sadly neither version works here with an older clang on OS X. Both variants build fine with 9 warnings each. But the old variant dies with a "Bus Error: 10", and the new variant with "Segmentation fault: 11". Same with gcc (albeit only 8 warnings.)

/edit

OK, just wrong user input. You gotta feed it a number, and not a "foobar" or another random string.

Now I should fix the link.

To have mazes look more human drawn, cells need to be irregular and the inner walls need naturally follow the contours of the outer shape.

If you actually want more aesthetic freedom, you can compose with an arbitrary diffeomorphism of the square to itself. But I think that might usually look worse.

† - That is, preserving all angles, including right angles. The terminology stems from the output angles conforming (???) to the input angles.

Is this intended to be able to be used on a phone? I saw the instructions about guiding a yellow ball but what I experienced was about 10 seconds of chiptune music with a maze that had some icons I couldn't interact with.

After the 10 seconds a series of green spheres and vertical lines washed across the screen but I wasn't able to control them.

I'm using the duck duck go browser on Android if that helps.

https://news.ycombinator.com/item?id=45549017

Wilson’s Algorithm gives the most pleasing visual results for me.

https://news.ycombinator.com/item?id=23131983

https://www.astrolog.org/labyrnth/psych.htm

I think there is a difference if you want to make it only expensive to solve using popular maze solver algorithms, vs to make it difficult for a human to solve. Many of the recommendations on that page are for how to do things that can make a maze more difficult for humans to solve, but will not always matter to an algorithm that just mechanically tries solutions in some order.