--- a/dom/media/MediaDecoderStateMachine.cpp
+++ b/dom/media/MediaDecoderStateMachine.cpp
@@ -2008,18 +2008,22 @@ MediaDecoderStateMachine::OnMetadataRead
   // thus get the metadata). We could fix this if we could compute the start
   // time by demuxing without necessaring decoding.
   bool waitingForCDM =
 #ifdef MOZ_EME
     mInfo.IsEncrypted() && !mCDMProxy;
 #else
     false;
 #endif
-  mNotifyMetadataBeforeFirstFrame = mDuration.Ref().isSome() || waitingForCDM;
-  if (mNotifyMetadataBeforeFirstFrame) {
+
+  // Notify metadata if it is MSE. Otherwise we will delay the event until 1st
+  // frames are decoded to workaround the case where video size in the metadata
+  // is different from that of the 1st frame.
+  if (waitingForCDM || mReader->ForceZeroStartTime()) {
+    mNotifyMetadataBeforeFirstFrame = true;
     EnqueueLoadedMetadataEvent();
   }
 
   if (waitingForCDM) {
     // Metadata parsing was successful but we're still waiting for CDM caps
     // to become available so that we can build the correct decryptor/decoder.
     SetState(DECODER_STATE_WAIT_FOR_CDM);
     return;