diff --git a/DOCS/interface-changes.rst b/DOCS/interface-changes.rst
index 23e71b3a2d..7fc9b0b5c5 100644
--- a/DOCS/interface-changes.rst
+++ b/DOCS/interface-changes.rst
@@ -59,6 +59,14 @@ Interface changes
       setting the properties to non-existing tracks may report it as selected
       track for a small time window, until it's forced back to "no". The exact
       details how this is handled may change in the future.
+    - x11 VOs now update --geometry at runtime according to the current window
+      position, and will react to writes to the option/property at runtime and
+      resize the window accordingly. This changes behavior when playing multiple
+      videos with different size: the first video initializes the window size
+      and thus overwrites --geometry, and the second video will obviously use
+      the size provided in --geometry, so the second video will not change the
+      window size. You can use --reset-on-next-file=geometry to get the old
+      behavior.
  --- mpv 0.30.0 ---
     - add `--d3d11-output-format` to enable explicit selection of a D3D11
       swap chain format.
diff --git a/DOCS/man/options.rst b/DOCS/man/options.rst
index f31173a8ee..7ec39afad8 100644
--- a/DOCS/man/options.rst
+++ b/DOCS/man/options.rst
@@ -2703,6 +2703,12 @@ Window
     The coordinates are relative to the screen given with ``--screen`` for the
     video output drivers that fully support ``--screen``.
 
+    Some backends update this option at runtime (like all options, it's exposed
+    as property, in this case ``geometry``), and resize their window if the user
+    sets the property. This means the window size is not reset when playing a
+    video with different file. Use ``--reset-on-next-file=geometry`` to get
+    the old behavior.
+
     .. note::
 
         Generally only supported by GUI VOs. Ignored for encoding.
diff --git a/video/out/x11_common.c b/video/out/x11_common.c
index 38fdca73df..dd5ce0fcdf 100644
--- a/video/out/x11_common.c
+++ b/video/out/x11_common.c
@@ -1730,14 +1730,32 @@ static void vo_x11_update_geometry(struct vo *vo)
     Window dummy_win;
     Window win = x11->parent ? x11->parent : x11->window;
     x11->winrc = (struct mp_rect){0, 0, 0, 0};
+    x11->last_geometry = (struct m_geometry){0};
     if (win) {
-        XGetGeometry(x11->display, win, &dummy_win, &dummy_int, &dummy_int,
+        int r_x = 0, r_y = 0;
+        XGetGeometry(x11->display, win, &dummy_win, &r_x, &r_y,
                      &w, &h, &dummy_int, &dummy_uint);
         if (w > INT_MAX || h > INT_MAX)
             w = h = 0;
         XTranslateCoordinates(x11->display, win, x11->rootwin, 0, 0,
                               &x, &y, &dummy_win);
         x11->winrc = (struct mp_rect){x, y, x + w, y + h};
+        if (!x11->window_hidden) {
+            // Ah, the pain of X11.
+            long params[4] = {0};
+            x11_get_property_copy(x11, win, XA(x11, _NET_FRAME_EXTENTS),
+                                  XA_CARDINAL, 32, params, sizeof(params));
+            x11->last_geometry = (struct m_geometry){
+                .x = x - params[0],
+                .y = y - params[2],
+                .w = w,
+                .h = h,
+                .xy_valid = 1,
+                .wh_valid = w > 0 && h > 0,
+            };
+            x11->opts->geometry = x11->last_geometry;
+            m_config_cache_write_opt(x11->opts_cache, &x11->opts->geometry);
+        }
     }
     double fps = 1000.0;
     for (int n = 0; n < x11->num_displays; n++) {
@@ -1844,6 +1862,24 @@ static void vo_x11_minimize(struct vo *vo)
     }
 }
 
+static void vo_x11_set_geometry(struct vo *vo)
+{
+    struct vo_x11_state *x11 = vo->x11;
+    struct m_geometry geo = x11->last_geometry;
+
+    if (x11->window_hidden || !geo.wh_valid || !geo.xy_valid)
+        return;
+
+    int s_w = mp_rect_w(x11->screenrc), s_h = mp_rect_h(x11->screenrc);
+    int nx = geo.x, ny = geo.y, nw = geo.w, nh = geo.h;
+    m_geometry_apply(&nx, &ny, &nw, &nh, s_w, s_h, &x11->opts->geometry);
+
+    MP_VERBOSE(vo, "Setting geometry: %d:%d %dx%d -> %d:%d %dx%d\n",
+               geo.x, geo.y, geo.w, geo.h, nx, ny, nw, nh);
+    if (nx != geo.x || ny != geo.y || nw != geo.w || nh != geo.h)
+        XMoveResizeWindow(x11->display, x11->window, nx, ny, nw, nh);
+}
+
 int vo_x11_control(struct vo *vo, int *events, int request, void *arg)
 {
     struct vo_x11_state *x11 = vo->x11;
@@ -1877,6 +1913,8 @@ int vo_x11_control(struct vo *vo, int *events, int request, void *arg)
                 vo_x11_minimize(vo);
             if (opt == &opts->window_maximized)
                 vo_x11_maximize(vo);
+            if (opt == &opts->geometry)
+                vo_x11_set_geometry(vo);
         }
         return VO_TRUE;
     }
diff --git a/video/out/x11_common.h b/video/out/x11_common.h
index fffc5be294..ae038bc3cf 100644
--- a/video/out/x11_common.h
+++ b/video/out/x11_common.h
@@ -24,6 +24,7 @@
 #include <X11/Xlib.h>
 #include <X11/Xutil.h>
 
+#include "options/m_option.h"
 #include "osdep/atomic.h"
 #include "osdep/semaphore.h"
 
@@ -100,6 +101,7 @@ struct vo_x11_state {
     // Current actual window position (updated on window move/resize events).
     struct mp_rect winrc;
     double current_display_fps;
+    struct m_geometry last_geometry;
 
     int pending_vo_events;