diff --git a/project-dataset/src/globals.py b/project-dataset/src/globals.py
index 3fcb1fb..83907ff 100644
--- a/project-dataset/src/globals.py
+++ b/project-dataset/src/globals.py
@@ -43,7 +43,11 @@
 # region ui-consants
 deployed_nn_tags = ["deployed_nn"]
 inference_modes = ["full image", "sliding window"]
-add_predictions_modes = ["merge with existing labels", "replace existing labels"]
+add_predictions_modes = [
+    "merge with existing labels",
+    "replace existing labels",
+    "replace existing labels and save image tags",
+]
 # endregion
 
 # region caches
diff --git a/project-dataset/src/ui/inference_preview.py b/project-dataset/src/ui/inference_preview.py
index 2b6f781..f3a0896 100644
--- a/project-dataset/src/ui/inference_preview.py
+++ b/project-dataset/src/ui/inference_preview.py
@@ -392,6 +392,7 @@ def apply_model_to_datasets(
     else:
         all_image_infos = {image_info.id: image_info for image_info in image_infos}
     ann_info_batch = []
+    add_mode = settings.add_predictions_mode.get_value()
     for i in inference_session.inference_project_id_async(project_id, dataset_ids):
         ann_info_batch.append(i)
         if len(ann_info_batch) < batch_size:
@@ -409,7 +410,29 @@ def apply_model_to_datasets(
             res_ann_infos.append(ann_info)
 
         original_anns = None
-        add_mode = settings.add_predictions_mode.get_value()
+        if add_mode == "replace existing labels and save image tags":
+            img_infos_dict = {}  # dataset_id -> image_id -> [ImageInfo]
+            merged_anns = []
+            for ann_info in res_ann_infos:
+                img_info = all_image_infos[ann_info.image_id]
+                img_infos_dict.setdefault(img_info.dataset_id, []).append(img_info)
+
+            # download original ann infos
+            for dataset_id, ds_image_infos in img_infos_dict.items():
+                original_ann_infos = g.api.annotation.download_batch(
+                    dataset_id, [image_info.id for image_info in ds_image_infos]
+                )
+                original_anns_dict = {
+                    ann_info.image_id: ann_info for ann_info in original_ann_infos
+                }
+                for ann_info in res_ann_infos:
+                    original_ann_info = original_anns_dict[ann_info.image_id]
+                    orig_ann = sly.Annotation.from_json(
+                        original_ann_info.annotation, res_project_meta
+                    )
+                    pred_ann = sly.Annotation.from_json(ann_info.annotation, res_project_meta)
+                    pred_ann = pred_ann.clone(img_tags=orig_ann.img_tags)
+                    merged_anns.append(original_ann_info._replace(annotation=pred_ann.to_json()))
         if add_mode == "merge with existing labels":
             img_infos_dict = {}  # dataset_id -> image_id -> [ImageInfo]
             merged_anns = []
@@ -450,7 +473,29 @@ def apply_model_to_datasets(
             res_ann_infos.append(ann_info)
 
         original_anns = None
-        add_mode = settings.add_predictions_mode.get_value()
+        if add_mode == "replace existing labels and save image tags":
+            img_infos_dict = {}  # dataset_id -> image_id -> [ImageInfo]
+            merged_anns = []
+            for ann_info in res_ann_infos:
+                img_info = all_image_infos[ann_info.image_id]
+                img_infos_dict.setdefault(img_info.dataset_id, []).append(img_info)
+
+            # download original ann infos
+            for dataset_id, ds_image_infos in img_infos_dict.items():
+                original_ann_infos = g.api.annotation.download_batch(
+                    dataset_id, [image_info.id for image_info in ds_image_infos]
+                )
+                original_anns_dict = {
+                    ann_info.image_id: ann_info for ann_info in original_ann_infos
+                }
+                for ann_info in res_ann_infos:
+                    original_ann_info = original_anns_dict[ann_info.image_id]
+                    orig_ann = sly.Annotation.from_json(
+                        original_ann_info.annotation, res_project_meta
+                    )
+                    pred_ann = sly.Annotation.from_json(ann_info.annotation, res_project_meta)
+                    pred_ann = pred_ann.clone(img_tags=orig_ann.img_tags)
+                    merged_anns.append(original_ann_info._replace(annotation=pred_ann.to_json()))
         if add_mode == "merge with existing labels":
             img_infos_dict = {}  # dataset_id -> image_id -> [ImageInfo]
             merged_anns = []
@@ -644,6 +689,15 @@ def get_inference_progress(inference_request_uuid):
     add_mode = settings.add_predictions_mode.get_value()
 
     original_anns = None
+    if add_mode == "replace existing labels and save image tags":
+        original_anns = g.api.annotation.download_batch(dataset_id, image_ids)
+        original_anns = [
+            sly.Annotation.from_json(ann_info.annotation, g.project_meta)
+            for ann_info in original_anns
+        ]
+        merged_anns = []
+        for ann, pred in zip(original_anns, res_anns):
+            merged_anns.append(pred.clone(img_tags=ann.img_tags))
     if add_mode == "merge with existing labels":
         original_anns = g.api.annotation.download_batch(dataset_id, image_ids)
         original_anns = [
diff --git a/project-dataset/src/ui/output_data.py b/project-dataset/src/ui/output_data.py
index 1bd1074..63b94d7 100644
--- a/project-dataset/src/ui/output_data.py
+++ b/project-dataset/src/ui/output_data.py
@@ -30,7 +30,9 @@
 card.collapse()
 
 
-def apply_model_ds(src_project, dst_project, inference_settings, res_project_meta):
+def apply_model_ds(
+    src_project, dst_project, inference_settings, res_project_meta, save_imag_tags=False
+):
     import time
 
     timer = {}