diff --git a/media/gpu/windows/d3d12_video_encode_h264_delegate.cc b/media/gpu/windows/d3d12_video_encode_h264_delegate.cc
index f6365db802d73..04997a08228a6 100644
--- a/media/gpu/windows/d3d12_video_encode_h264_delegate.cc
+++ b/media/gpu/windows/d3d12_video_encode_h264_delegate.cc
@@ -648,7 +648,13 @@ EncoderStatus D3D12VideoEncodeH264Delegate::InitializeVideoEncoder(
                                  "support manual reference control, got %u",
                                  picture_control_support_h264.MaxDPBCapacity)};
     }
-    max_num_ref_frames_ = picture_control_support_h264.MaxDPBCapacity;
+    // Manual reference buffers map directly onto long-term reference indices,
+    // so the number of DPB slots used must not exceed what the driver can
+    // track as long-term references plus the slot reserved for frame_num gap
+    // handling.
+    max_num_ref_frames_ =
+        std::min(picture_control_support_h264.MaxDPBCapacity,
+                 picture_control_support_h264.MaxLongTermReferences + 1);
   }
 
   if ((config.bitrate.mode() == Bitrate::Mode::kConstant ||
diff --git a/media/gpu/windows/d3d12_video_encode_h264_delegate_unittest.cc b/media/gpu/windows/d3d12_video_encode_h264_delegate_unittest.cc
index 457d49b68f2fb..8613f758f5632 100644
--- a/media/gpu/windows/d3d12_video_encode_h264_delegate_unittest.cc
+++ b/media/gpu/windows/d3d12_video_encode_h264_delegate_unittest.cc
@@ -72,8 +72,9 @@ class D3D12VideoEncodeH264DelegateTest
             return E_INVALIDARG;
           }
           picture_control->PictureSupport.pH264Support->MaxLongTermReferences =
-              1;
-          picture_control->PictureSupport.pH264Support->MaxDPBCapacity = 16;
+              kMaxLongTermReferences;
+          picture_control->PictureSupport.pH264Support->MaxDPBCapacity =
+              kMaxDPBCapacity;
           return S_OK;
         });
     ON_CALL(*video_device3_.Get(),
@@ -195,6 +196,8 @@ class D3D12VideoEncodeH264DelegateTest
 
   static constexpr D3D12_VIDEO_ENCODER_LEVELS_H264 kMaxLevel =
       D3D12_VIDEO_ENCODER_LEVELS_H264_31;
+  static constexpr UINT kMaxLongTermReferences = 4;
+  static constexpr UINT kMaxDPBCapacity = 16;
   Microsoft::WRL::ComPtr<D3D12DeviceMock> device_;
   Microsoft::WRL::ComPtr<D3D12VideoDevice3Mock> video_device3_;
 };
@@ -471,6 +474,21 @@ TEST_F(D3D12VideoEncodeH264DelegateTest,
   }
 }
 
+TEST_F(D3D12VideoEncodeH264DelegateTest,
+       ManualReferenceBuffersLimitedByLongTermReferences) {
+  static_assert(kMaxLongTermReferences < kMaxDPBCapacity);
+  VideoEncodeAccelerator::Config config = GetDefaultH264Config();
+  config.manual_reference_buffer_control = true;
+  ASSERT_TRUE(encoder_delegate_->Initialize(config).is_ok());
+  // Manual reference buffers are implemented as long-term references, so the
+  // number of DPB slots used must respect the driver's long-term reference
+  // limit rather than its full DPB capacity.
+  EXPECT_EQ(encoder_delegate_->GetMaxNumOfRefFrames(),
+            kMaxLongTermReferences + 1u);
+  EXPECT_LE(encoder_delegate_->GetMaxNumOfManualRefBuffers(),
+            static_cast<size_t>(kMaxLongTermReferences));
+}
+
 TEST_F(D3D12VideoEncodeH264DelegateTest, EncodeWithManualReferenceControl) {
   VideoEncodeAccelerator::Config config = GetDefaultH264Config();
   config.manual_reference_buffer_control = true;