diff --git a/README.md b/README.md
index c77c715..b0b57cc 100644
--- a/README.md
+++ b/README.md
@@ -71,9 +71,8 @@ FERESE_ENABLE_SCREENCOPY=1 cargo run -p ferese
 ```
 
 The current M5 implementation supports one-shot output and region capture into
-ARGB8888 shared-memory buffers. Captures currently include Ferese's software
-cursor even when a client does not request a cursor overlay; cursor-free
-capture remains part of the M5 interoperability acceptance work.
+ARGB8888 shared-memory buffers. Cursor overlays are included only when the
+capture client requests them.
 
 ### Direct DRM session
 
diff --git a/compositor/src/backends/direct.rs b/compositor/src/backends/direct.rs
index 81aa41a..55a5a25 100644
--- a/compositor/src/backends/direct.rs
+++ b/compositor/src/backends/direct.rs
@@ -38,7 +38,10 @@ use smithay::{
     wayland::presentation::Refresh,
 };
 
-use crate::{Ferese, winit::animated_window_elements};
+use crate::{
+    Ferese,
+    winit::{animated_window_elements, cursorless_window_elements, redraw_output},
+};
 
 pub struct DirectBackendState {
     pub session: LibSeatSession,
@@ -349,7 +352,27 @@ fn render_output(state: &mut Ferese, node: DrmNode, crtc: crtc::Handle) {
             &elements,
             [0.035, 0.04, 0.055, 1.0],
         )?;
-        if state.process_screencopies(&mut device.renderer, &framebuffer, &output.output) {
+        let cursorless_capture = state.has_pending_screencopy(&output.output, false);
+        let captured_with_cursor =
+            state.process_screencopies(&mut device.renderer, &framebuffer, &output.output, true);
+
+        if cursorless_capture {
+            let cursorless_elements =
+                cursorless_window_elements(state, &mut device.renderer, &output.output);
+            redraw_output(
+                &mut device.renderer,
+                &mut framebuffer,
+                &output.output,
+                &cursorless_elements,
+            )?;
+            state.process_screencopies(&mut device.renderer, &framebuffer, &output.output, false);
+            redraw_output(
+                &mut device.renderer,
+                &mut framebuffer,
+                &output.output,
+                &elements,
+            )?;
+        } else if captured_with_cursor {
             let _ = device
                 .renderer
                 .render(
diff --git a/compositor/src/handlers/screencopy.rs b/compositor/src/handlers/screencopy.rs
index 53b3dec..c4c6d49 100644
--- a/compositor/src/handlers/screencopy.rs
+++ b/compositor/src/handlers/screencopy.rs
@@ -28,6 +28,7 @@ const BYTES_PER_PIXEL: usize = 4;
 pub(crate) struct FrameData {
     output: Option<Output>,
     region: Rectangle<i32, Buffer>,
+    overlay_cursor: bool,
     used: Mutex<bool>,
 }
 
@@ -37,6 +38,7 @@ pub(crate) struct PendingScreencopy {
     buffer: WlBuffer,
     pub(crate) output: Output,
     region: Rectangle<i32, Buffer>,
+    overlay_cursor: bool,
     with_damage: bool,
 }
 
@@ -75,14 +77,14 @@ impl Dispatch<ZwlrScreencopyManagerV1, ()> for Ferese {
         match request {
             zwlr_screencopy_manager_v1::Request::CaptureOutput {
                 frame,
-                overlay_cursor: _,
+                overlay_cursor,
                 output,
             } => {
-                create_frame(data_init, frame, output, None);
+                create_frame(data_init, frame, output, None, overlay_cursor != 0);
             }
             zwlr_screencopy_manager_v1::Request::CaptureOutputRegion {
                 frame,
-                overlay_cursor: _,
+                overlay_cursor,
                 output,
                 x,
                 y,
@@ -94,6 +96,7 @@ impl Dispatch<ZwlrScreencopyManagerV1, ()> for Ferese {
                     frame,
                     output,
                     Some(Rectangle::new((x, y).into(), (width, height).into())),
+                    overlay_cursor != 0,
                 );
             }
             zwlr_screencopy_manager_v1::Request::Destroy => {}
@@ -161,6 +164,7 @@ impl Ferese {
             buffer,
             output,
             region: data.region,
+            overlay_cursor: data.overlay_cursor,
             with_damage,
         });
         crate::backends::direct::render_all(self);
@@ -171,6 +175,7 @@ impl Ferese {
         renderer: &mut R,
         framebuffer: &R::Framebuffer<'_>,
         output: &Output,
+        overlay_cursor: bool,
     ) -> bool
     where
         R: ExportMem,
@@ -180,7 +185,9 @@ impl Ferese {
         let captures = std::mem::take(&mut self.pending_screencopies);
 
         for capture in captures {
-            if &capture.output != output {
+            if &capture.output != output
+                || !cursor_overlay_matches(capture.overlay_cursor, overlay_cursor)
+            {
                 remaining.push(capture);
                 continue;
             }
@@ -198,6 +205,17 @@ impl Ferese {
         self.pending_screencopies = remaining;
         framebuffer_binding_changed
     }
+
+    pub(crate) fn has_pending_screencopy(&self, output: &Output, overlay_cursor: bool) -> bool {
+        self.pending_screencopies.iter().any(|capture| {
+            capture.output == *output
+                && cursor_overlay_matches(capture.overlay_cursor, overlay_cursor)
+        })
+    }
+}
+
+fn cursor_overlay_matches(requested: bool, rendered: bool) -> bool {
+    requested == rendered
 }
 
 fn create_frame(
@@ -205,6 +223,7 @@ fn create_frame(
     frame: New<ZwlrScreencopyFrameV1>,
     wl_output: smithay::reexports::wayland_server::protocol::wl_output::WlOutput,
     requested: Option<Rectangle<i32, Logical>>,
+    overlay_cursor: bool,
 ) {
     let Some(output) = Output::from_resource(&wl_output) else {
         let frame = data_init.init(
@@ -212,6 +231,7 @@ fn create_frame(
             Arc::new(FrameData {
                 output: None,
                 region: Rectangle::from_size((1, 1).into()),
+                overlay_cursor,
                 used: Mutex::new(true),
             }),
         );
@@ -224,6 +244,7 @@ fn create_frame(
             Arc::new(FrameData {
                 output: Some(output),
                 region: Rectangle::from_size((1, 1).into()),
+                overlay_cursor,
                 used: Mutex::new(true),
             }),
         );
@@ -235,6 +256,7 @@ fn create_frame(
     let data = Arc::new(FrameData {
         output: Some(output),
         region,
+        overlay_cursor,
         used: Mutex::new(false),
     });
     let frame = data_init.init(frame, data);
@@ -392,4 +414,12 @@ mod tests {
             capture_region_for_geometry(mode, 2.0, Transform::Normal, Some(requested),).is_none()
         );
     }
+
+    #[test]
+    fn cursor_overlay_requests_use_only_the_matching_render_pass() {
+        assert!(cursor_overlay_matches(false, false));
+        assert!(cursor_overlay_matches(true, true));
+        assert!(!cursor_overlay_matches(false, true));
+        assert!(!cursor_overlay_matches(true, false));
+    }
 }
diff --git a/compositor/src/winit.rs b/compositor/src/winit.rs
index 5930952..8d1ad9b 100644
--- a/compositor/src/winit.rs
+++ b/compositor/src/winit.rs
@@ -6,10 +6,10 @@ use std::{
 use smithay::{
     backend::{
         renderer::{
-            Frame, ImportDma, Renderer,
+            Color32F, Frame, ImportDma, Renderer,
             damage::OutputDamageTracker,
             element::{
-                AsRenderElements, Kind as RenderElementKind,
+                AsRenderElements, Kind as RenderElementKind, RenderElement,
                 memory::MemoryRenderBufferRenderElement,
                 render_elements,
                 surface::{WaylandSurfaceRenderElement, render_elements_from_surface_tree},
@@ -19,6 +19,7 @@ use smithay::{
                 },
             },
             gles::GlesRenderer,
+            utils::draw_render_elements,
         },
         winit::{self, WinitEvent},
     },
@@ -143,7 +144,22 @@ pub fn init(event_loop: &mut EventLoop<Ferese>, state: &mut Ferese) -> Result<()
                             &elements,
                             [0.035, 0.04, 0.055, 1.0],
                         )?;
-                        if state.process_screencopies(renderer, &framebuffer, &output) {
+                        let cursorless_capture = state.has_pending_screencopy(&output, false);
+                        let captured_with_cursor =
+                            state.process_screencopies(renderer, &framebuffer, &output, true);
+
+                        if cursorless_capture {
+                            let cursorless_elements =
+                                output_elements(state, renderer, &output, false);
+                            redraw_output(
+                                renderer,
+                                &mut framebuffer,
+                                &output,
+                                &cursorless_elements,
+                            )?;
+                            state.process_screencopies(renderer, &framebuffer, &output, false);
+                            redraw_output(renderer, &mut framebuffer, &output, &elements)?;
+                        } else if captured_with_cursor {
                             let _ = renderer
                                 .render(
                                     &mut framebuffer,
@@ -203,13 +219,34 @@ pub(crate) fn animated_window_elements(
     state: &Ferese,
     renderer: &mut GlesRenderer,
     output: &Output,
+) -> Vec<AnimatedWindowRenderElement> {
+    output_elements(state, renderer, output, true)
+}
+
+pub(crate) fn cursorless_window_elements(
+    state: &Ferese,
+    renderer: &mut GlesRenderer,
+    output: &Output,
+) -> Vec<AnimatedWindowRenderElement> {
+    output_elements(state, renderer, output, false)
+}
+
+fn output_elements(
+    state: &Ferese,
+    renderer: &mut GlesRenderer,
+    output: &Output,
+    include_cursor: bool,
 ) -> Vec<AnimatedWindowRenderElement> {
     let Some(output_geometry) = state.space.output_geometry(output) else {
         return Vec::new();
     };
     let scale = output.current_scale().fractional_scale();
 
-    let mut elements = cursor_elements(state, renderer, output_geometry, scale);
+    let mut elements = if include_cursor {
+        cursor_elements(state, renderer, output_geometry, scale)
+    } else {
+        Vec::new()
+    };
     elements.extend(layer_elements(
         renderer,
         output,
@@ -264,6 +301,34 @@ pub(crate) fn animated_window_elements(
     elements
 }
 
+pub(crate) fn redraw_output<R, E>(
+    renderer: &mut R,
+    framebuffer: &mut R::Framebuffer<'_>,
+    output: &Output,
+    elements: &[E],
+) -> Result<(), R::Error>
+where
+    R: Renderer,
+    R::TextureId: 'static,
+    E: RenderElement<R>,
+{
+    let mode = output.current_mode().expect("output has a mode");
+    let transform = output.current_transform().invert();
+    let damage = Rectangle::from_size(transform.transform_size(mode.size));
+    let mut frame = renderer.render(framebuffer, mode.size, transform)?;
+
+    frame.clear(Color32F::new(0.035, 0.04, 0.055, 1.0), &[damage])?;
+    draw_render_elements(
+        &mut frame,
+        output.current_scale().fractional_scale(),
+        elements,
+        &[damage],
+    )?;
+    let _ = frame.finish()?;
+
+    Ok(())
+}
+
 fn layer_surfaces(output: &Output) -> Vec<LayerSurface> {
     layer_map_for_output(output).layers().cloned().collect()
 }