fix(other-models): default downloads to a flat path, not {base_model}/{first_tag}

get_download_path_template() fell back to "{base_model}/{first_tag}" for any
unconfigured model type, so other-model downloads were silently nested under an
arbitrary CivitAI tag even though the settings UI exposes no template row for
"other" and priority_tags has no "other" entry (making {first_tag} resolve to
tags[0]).

Add DEFAULT_DOWNLOAD_PATH_TEMPLATES with other -> "" so unconfigured and
unknown types resolve to a flat layout under the already sub_type-scoped
default_other_roots; explicit settings.json values still win. Mirror the flat
default in the frontend DEFAULT_PATH_TEMPLATES and stop the download/move
default-path previews from rendering "/undefined" or a dangling slash.

py/services/settings_manager.py

@@ -25,6 +25,7 @@ from typing import (
 from platformdirs import user_config_dir
 
 from ..utils.constants import (
+    DEFAULT_DOWNLOAD_PATH_TEMPLATES,
     DEFAULT_ENABLED_OTHER_SUB_TYPES,
     DEFAULT_HASH_CHUNK_SIZE_MB,
     DEFAULT_PRIORITY_TAG_CONFIG,
@@ -2335,10 +2336,14 @@ class SettingsManager:
         """Get download path template for specific model type
 
         Args:
-            model_type: The type of model ('lora', 'checkpoint', 'embedding')
+            model_type: The type of model ('lora', 'checkpoint', 'embedding',
+                'other')
 
         Returns:
-            Template string for the model type, defaults to '{base_model}/{first_tag}'
+            Template string for the model type. Falls back to the per-type
+            default in ``DEFAULT_DOWNLOAD_PATH_TEMPLATES``; unknown model types
+            resolve to an empty string (flat layout) rather than silently
+            nesting downloads under an unconfigured subfolder.
         """
         templates = self.settings.get("download_path_templates", {})
 
@@ -2362,27 +2367,19 @@ class SettingsManager:
                 logger.warning(
                     f"Failed to parse download_path_templates JSON string: {e}. Setting default values."
                 )
-                default_template = "{base_model}/{first_tag}"
-                templates = {
-                    "lora": default_template,
-                    "checkpoint": default_template,
-                    "embedding": default_template,
-                }
+                templates = dict(DEFAULT_DOWNLOAD_PATH_TEMPLATES)
                 self.settings["download_path_templates"] = templates
                 self._save_settings()
 
         # Ensure templates is a dictionary
         if not isinstance(templates, dict):
-            default_template = "{base_model}/{first_tag}"
-            templates = {
-                "lora": default_template,
-                "checkpoint": default_template,
-                "embedding": default_template,
-            }
+            templates = dict(DEFAULT_DOWNLOAD_PATH_TEMPLATES)
             self.settings["download_path_templates"] = templates
             self._save_settings()
 
-        return templates.get(model_type, "{base_model}/{first_tag}")
+        return templates.get(
+            model_type, DEFAULT_DOWNLOAD_PATH_TEMPLATES.get(model_type, "")
+        )
 
 
 _SETTINGS_MANAGER: Optional["SettingsManager"] = None

py/utils/constants.py

@@ -251,6 +251,19 @@ DEFAULT_PRIORITY_TAG_CONFIG = {
     "embedding": ", ".join(CIVITAI_MODEL_TAGS),
 }
 
+# Default download path template for each model type. "other" defaults to a
+# flat layout (empty template) on purpose: other-model downloads are already
+# separated by sub_type roots (default_other_roots), and priority_tags has no
+# "other" entry, so {first_tag} would resolve to an arbitrary CivitAI tag and
+# scatter files into unstable folders. Users can still opt in to a template by
+# writing "other" into download_path_templates in settings.json.
+DEFAULT_DOWNLOAD_PATH_TEMPLATES: Dict[str, str] = {
+    "lora": "{base_model}/{first_tag}",
+    "checkpoint": "{base_model}/{first_tag}",
+    "embedding": "{base_model}/{first_tag}",
+    "other": "",
+}
+
 # baseModel values from CivitAI that should be treated as diffusion models (unet)
 # These model types are incorrectly labeled as "checkpoint" by CivitAI but are actually diffusion models
 DIFFUSION_MODEL_BASE_MODELS = frozenset(