diff --git a/gst/isomp4/qtdemux.c b/gst/isomp4/qtdemux.c
index e36e4be760..082da382a8 100644
--- a/gst/isomp4/qtdemux.c
+++ b/gst/isomp4/qtdemux.c
@@ -8803,6 +8803,9 @@ qtdemux_parse_trak (GstQTDemux * qtdemux, GNode * trak)
       }
     }
 
+    if (stream->caps)
+      gst_caps_unref (stream->caps);
+
     stream->caps =
         qtdemux_video_caps (qtdemux, stream, fourcc, stsd_data, &codec);
     if (G_UNLIKELY (!stream->caps)) {
@@ -9533,6 +9536,9 @@ qtdemux_parse_trak (GstQTDemux * qtdemux, GNode * trak)
       GST_WARNING_OBJECT (qtdemux, "unknown audio STSD version %08x", version);
     }
 
+    if (stream->caps)
+      gst_caps_unref (stream->caps);
+
     stream->caps = qtdemux_audio_caps (qtdemux, stream, fourcc,
         stsd_data + 32, len - 16, &codec);