diff --git a/src/desktop/view/WLSurface.cpp b/src/desktop/view/WLSurface.cpp
index 1c3ac8aa1..4c2de539e 100644
--- a/src/desktop/view/WLSurface.cpp
+++ b/src/desktop/view/WLSurface.cpp
@@ -85,52 +85,52 @@ CRegion CWLSurface::computeDamage(const std::optional<CBox>& box) const {
     if (!m_resource->m_current.texture)
         return {};
 
-    CRegion damage = m_resource->m_current.accumulateBufferDamage();
-    damage.transform(Math::wlTransformToHyprutils(m_resource->m_current.transform), m_resource->m_current.bufferSize.x, m_resource->m_current.bufferSize.y);
-
     const auto BUFSIZE = m_resource->m_current.bufferSize;
     if (BUFSIZE.x <= 0 || BUFSIZE.y <= 0)
         return {};
 
-    const auto CORRECTVEC = correctSmallVecBuf();
-
-    if (m_resource->m_current.viewport.hasSource)
-        damage.intersect(m_resource->m_current.viewport.source);
-
     const auto SCALEDSRCSIZE =
         m_resource->m_current.viewport.hasSource ? m_resource->m_current.viewport.source.size() * m_resource->m_current.scale : m_resource->m_current.bufferSize;
     if (SCALEDSRCSIZE.x <= 0 || SCALEDSRCSIZE.y <= 0)
         return {};
 
-    damage.scale({BUFSIZE.x / SCALEDSRCSIZE.x, BUFSIZE.y / SCALEDSRCSIZE.y});
-    damage.translate(CORRECTVEC);
-
-    // go from buffer coords in the damage to hl logical
-
     const auto SURFSIZE = m_resource->m_current.size;
     if (SURFSIZE.x <= 0 || SURFSIZE.y <= 0)
         return {};
 
-    const Vector2D SCALE = SURFSIZE / m_resource->m_current.bufferSize;
-
-    damage.scale(SCALE);
+    std::optional<Vector2D> boxSize;
     if (box.has_value()) {
-        auto boxSize = box->size();
+        boxSize = box->size();
 
         if (m_view->type() == VIEW_TYPE_WINDOW) {
             const auto WINDOW = dynamicPointerCast<CWindow>(m_view.lock());
             if (!WINDOW)
                 return {};
 
-            boxSize = boxSize * WINDOW->m_X11SurfaceScaledBy;
+            boxSize = boxSize.value() * WINDOW->m_X11SurfaceScaledBy;
         }
 
-        if (boxSize.x <= 0 || boxSize.y <= 0)
+        if (boxSize->x <= 0 || boxSize->y <= 0)
             return {};
-
-        damage.intersect(CBox{{}, boxSize});
     }
 
+    const auto CORRECTVEC = correctSmallVecBuf();
+
+    CRegion    damage = m_resource->m_current.accumulateBufferDamage();
+    damage.transform(Math::wlTransformToHyprutils(m_resource->m_current.transform), BUFSIZE.x, BUFSIZE.y);
+
+    if (m_resource->m_current.viewport.hasSource)
+        damage.intersect(m_resource->m_current.viewport.source);
+
+    damage.scale({BUFSIZE.x / SCALEDSRCSIZE.x, BUFSIZE.y / SCALEDSRCSIZE.y});
+    damage.translate(CORRECTVEC);
+
+    // go from buffer coords in the damage to hl logical
+    damage.scale(SURFSIZE / BUFSIZE);
+
+    if (boxSize)
+        damage.intersect(CBox{{}, boxSize.value()});
+
     return damage;
 }
 
diff --git a/src/desktop/view/Window.cpp b/src/desktop/view/Window.cpp
index ff9e88942..ff98ee9a4 100644
--- a/src/desktop/view/Window.cpp
+++ b/src/desktop/view/Window.cpp
@@ -2785,6 +2785,8 @@ void CWindow::commitWindow() {
 
     const auto PMONITOR = m_monitor.lock();
 
+    // damageSurface consumes damage, so snapshot it for the tearing check below
+    const bool HADDAMAGE = !wlSurface()->resource()->m_current.damage.empty() || !wlSurface()->resource()->m_current.bufferDamage.empty();
     g_pHyprRenderer->damageSurface(wlSurface()->resource(), m_realPosition->goal().x, m_realPosition->goal().y, m_isX11 ? 1.0 / m_X11SurfaceScaledBy : 1.0);
 
     if (!m_isX11) {
@@ -2794,16 +2796,12 @@ void CWindow::commitWindow() {
 
     // tearing: if solitary, redraw it. This still might be a single surface window
     if (PMONITOR && PMONITOR->m_solitaryClient.lock() == m_self.lock() && canBeTorn() && PMONITOR->m_tearingState.canTear && wlSurface()->resource()->m_current.texture &&
-        !PMONITOR->isTearingBlocked()) {
-        CRegion damageBox{wlSurface()->resource()->m_current.accumulateBufferDamage()};
-
-        if (!damageBox.empty()) {
-            if (PMONITOR->m_tearingState.busy) {
-                PMONITOR->m_tearingState.frameScheduledWhileBusy = true;
-            } else {
-                PMONITOR->m_tearingState.nextRenderTorn = true;
-                g_pHyprRenderer->renderMonitor(PMONITOR);
-            }
+        !PMONITOR->isTearingBlocked() && HADDAMAGE) {
+        if (PMONITOR->m_tearingState.busy) {
+            PMONITOR->m_tearingState.frameScheduledWhileBusy = true;
+        } else {
+            PMONITOR->m_tearingState.nextRenderTorn = true;
+            g_pHyprRenderer->renderMonitor(PMONITOR);
         }
     }
 }
diff --git a/src/output/Monitor.cpp b/src/output/Monitor.cpp
index fc397a63b..4097916b2 100644
--- a/src/output/Monitor.cpp
+++ b/src/output/Monitor.cpp
@@ -2207,7 +2207,13 @@ bool CMonitor::attemptDirectScanout() {
 
     PSURFACE->presentFeedback(Time::steadyNow(), m_self.lock());
 
-    m_output->state->addDamage(PSURFACE->m_current.accumulateBufferDamage());
+    // the commit path already added the damage into the damagering
+    CRegion scanoutDamage = m_damage.getBufferDamage(1);
+    // the ring is in transformed space, the fb we hand to KMS is in pixel space
+    scanoutDamage.transform(Math::wlTransformToHyprutils(Math::invertTransform(m_transform)), m_transformedSize.x, m_transformedSize.y);
+    // expand to not miss pixels from rounding, being a bit over is safe, going below is stale pixels.
+    scanoutDamage.expand(1).intersect(CBox{{}, m_pixelSize});
+    m_output->state->addDamage(scanoutDamage);
 
     // multigpu needs a fence to trigger fence syncing blits and also committing with the recreated dgpu fence
     if (g_pHyprRenderer->explicitSyncSupported() && isMultiGPU()) {
@@ -2235,6 +2241,9 @@ bool CMonitor::attemptDirectScanout() {
 
     scanoutCommitted = true;
 
+    // the flip has used the damage, rotate it.
+    m_damage.rotate();
+
     if (m_lastScanout.expired()) {
         m_lastScanout = PCANDIDATE;
         Log::logger->log(Log::DEBUG, "Entered a direct scanout to {:x}: \"{}\"", rc<uintptr_t>(PCANDIDATE.get()), PCANDIDATE->m_title);
@@ -2268,6 +2277,7 @@ void CMonitor::handleDSleave() {
 
     m_drmFormat   = m_prevDrmFormat;
     m_blurFBDirty = true;
+    m_damage.damageEntire();
 }
 
 bool CMonitor::canAttemptDirectScanoutFast() const {
diff --git a/src/protocols/types/SurfaceState.cpp b/src/protocols/types/SurfaceState.cpp
index 80189ed36..4605a64f8 100644
--- a/src/protocols/types/SurfaceState.cpp
+++ b/src/protocols/types/SurfaceState.cpp
@@ -17,24 +17,32 @@ Vector2D SSurfaceState::sourceSize() {
     return trc / scale;
 }
 
-CRegion SSurfaceState::accumulateBufferDamage() {
-    if (damage.empty())
-        return bufferDamage;
+bool SSurfaceState::consumeBufferDamage() const {
+    return buffer && !buffer->isSynchronous();
+}
 
-    CRegion surfaceDamage = damage;
-    if (viewport.hasDestination) {
-        Vector2D scale = sourceSize() / viewport.destination;
-        surfaceDamage.scale(scale);
+CRegion SSurfaceState::accumulateBufferDamage() {
+    if (!damage.empty()) {
+        CRegion surfaceDamage = damage;
+        if (viewport.hasDestination) {
+            Vector2D scale = sourceSize() / viewport.destination;
+            surfaceDamage.scale(scale);
+        }
+
+        if (viewport.hasSource)
+            surfaceDamage.translate(viewport.source.pos());
+
+        Vector2D trc = transform % 2 == 1 ? Vector2D{bufferSize.y, bufferSize.x} : bufferSize;
+
+        bufferDamage = surfaceDamage.scale(scale).transform(Math::wlTransformToHyprutils(Math::invertTransform(transform)), trc.x, trc.y).add(bufferDamage);
+        damage.clear();
     }
 
-    if (viewport.hasSource)
-        surfaceDamage.translate(viewport.source.pos());
+    auto taken = bufferDamage;
+    if (consumeBufferDamage())
+        bufferDamage.clear();
 
-    Vector2D trc = transform % 2 == 1 ? Vector2D{bufferSize.y, bufferSize.x} : bufferSize;
-
-    bufferDamage = surfaceDamage.scale(scale).transform(Math::wlTransformToHyprutils(Math::invertTransform(transform)), trc.x, trc.y).add(bufferDamage);
-    damage.clear();
-    return bufferDamage;
+    return taken;
 }
 
 CRegion SSurfaceState::effectiveInputRegion() const {
@@ -181,10 +189,19 @@ void SSurfaceState::updateFrom(SSurfaceState& ref) {
     }
 
     if (ref.updated.bits.damage) {
-        damage       = ref.damage;
-        bufferDamage = ref.bufferDamage;
-    } else {
-        // damage is always relative to the current commit
+        if (consumeBufferDamage()) {
+            // what we still hold is only consumed on rendering.
+            // dropping it here would leave those pixels stale.
+            damage.add(ref.damage);
+            bufferDamage.add(ref.bufferDamage);
+        } else {
+            damage       = ref.damage;
+            bufferDamage = ref.bufferDamage;
+        }
+    } else if (ref.updated.bits.buffer || !consumeBufferDamage()) {
+        // damage is relative to the buffer, shm drops the buffer on commit
+        // dmabuf doesnt drop it until we recieve a new one. and we cant clear
+        // the dmabuf damage until the renderer has actually consumed it.
         damage.clear();
         bufferDamage.clear();
     }
diff --git a/src/protocols/types/SurfaceState.hpp b/src/protocols/types/SurfaceState.hpp
index 08da928f2..cb8a63a51 100644
--- a/src/protocols/types/SurfaceState.hpp
+++ b/src/protocols/types/SurfaceState.hpp
@@ -119,6 +119,7 @@ struct SSurfaceState {
 
     // helpers
     CRegion accumulateBufferDamage();       // transforms state.damage and merges it into state.bufferDamage
+    bool    consumeBufferDamage() const;    // whether accumulateBufferDamage() takes the damage rather than leaving it
     CRegion effectiveInputRegion() const;   // materializes the input region clipped to the current surface size
     void    updateFrom(SSurfaceState& ref); // updates this state based on a reference state.
     void    reset();                        // resets pending state after commit