diff --git a/core/workflow_analyzer.py b/core/workflow_analyzer.py
index 7b6d6a4..0de7da0 100644
--- a/core/workflow_analyzer.py
+++ b/core/workflow_analyzer.py
@@ -1451,8 +1451,15 @@ def get_node_model_info(
             if model_widget_category_hints
             else ([effective_category_hint] if effective_category_hint else None)
         )
+        # A generic widget name such as ``model_name`` is useful for choosing a
+        # category, but it is not proof that an arbitrary STRING value is a local
+        # model reference. Some metadata/search nodes use that name for human-
+        # readable titles. Only folder-backed schemas, known widget indexes, or a
+        # model-typed combo/output pairing are strong enough to admit an
+        # extensionless value.
         category_backed_model_widget = bool(
-            model_widget_category_hint
+            model_widget_folder_key_hints
+            or get_node_model_widget_category_hint(node_type, idx)
             or schema_output_category_hint
         )
         named_model_file_widget = bool(
diff --git a/tests/test_workflow_analyzer.py b/tests/test_workflow_analyzer.py
index 430f644..96bcd15 100644
--- a/tests/test_workflow_analyzer.py
+++ b/tests/test_workflow_analyzer.py
@@ -209,6 +209,46 @@ class WorkflowAnalyzerCategoryHintTests(unittest.TestCase):
 
         self.assertEqual([], refs)
 
+    def test_model_name_string_used_as_metadata_query_is_ignored(self):
+        workflow = {
+            "nodes": [
+                {
+                    "id": 153,
+                    "type": "Civitai Hash Fetcher (Image Saver)",
+                    "inputs": [
+                        {
+                            "name": "username",
+                            "type": "STRING",
+                            "widget": {"name": "username"},
+                        },
+                        {
+                            "name": "model_name",
+                            "type": "STRING",
+                            "widget": {"name": "model_name"},
+                        },
+                        {
+                            "name": "version",
+                            "type": "STRING",
+                            "widget": {"name": "version"},
+                        },
+                    ],
+                    "outputs": [{"name": "STRING", "type": "STRING", "links": []}],
+                    "widgets_values": [
+                        "latentheart",
+                        (
+                            "Krea2 [SFW / NSFW] Uncensored - Image-to-Prompt + "
+                            "Prompt Enhancer + 4K Upscaler + CivitAI Metadata"
+                        ),
+                        "",
+                    ],
+                }
+            ]
+        }
+
+        refs = analyze_workflow_models(workflow, available_models=[])
+
+        self.assertEqual([], refs)
+
     def test_upscale_model_loader_widget_named_model_name_stays_upscale(self):
         workflow = {
             "nodes": [