diff --git a/mobile/lib/services/machine_learning/face_ml/face_clustering/face_clustering_service.dart b/mobile/lib/services/machine_learning/face_ml/face_clustering/face_clustering_service.dart
index 3219fe7f9..c84059e39 100644
--- a/mobile/lib/services/machine_learning/face_ml/face_clustering/face_clustering_service.dart
+++ b/mobile/lib/services/machine_learning/face_ml/face_clustering/face_clustering_service.dart
@@ -854,7 +854,10 @@ class FaceClusteringService {
           .toList();
       final count = clusterIdToFaceInfos[clusterId]!.length;
       final Vector meanEmbedding = embeddings.reduce((a, b) => a + b) / count;
-      clusterIdToMeanEmbeddingAndWeight[clusterId] = (meanEmbedding, count);
+      final Vector meanEmbeddingNormalized =
+          meanEmbedding / meanEmbedding.norm();
+      clusterIdToMeanEmbeddingAndWeight[clusterId] =
+          (meanEmbeddingNormalized, count);
     }
 
     // Now merge the clusters that are close to each other, based on mean embedding
@@ -889,8 +892,10 @@ class FaceClusteringService {
         final count2 = clusterIdToMeanEmbeddingAndWeight[clusterID2]!.$2;
         final weight1 = count1 / (count1 + count2);
         final weight2 = count2 / (count1 + count2);
+        final weightedMean = mean1 * weight1 + mean2 * weight2;
+        final weightedMeanNormalized = weightedMean / weightedMean.norm();
         clusterIdToMeanEmbeddingAndWeight[clusterID1] = (
-          mean1 * weight1 + mean2 * weight2,
+          weightedMeanNormalized,
           count1 + count2,
         );
         clusterIdToMeanEmbeddingAndWeight.remove(clusterID2);