diff --git a/backend/python/pocket-tts/backend.py b/backend/python/pocket-tts/backend.py
index d372d0d11..ef073ad0c 100644
--- a/backend/python/pocket-tts/backend.py
+++ b/backend/python/pocket-tts/backend.py
@@ -102,9 +102,17 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         self.default_voice_url = self.options.get("default_voice", None)
         self._voice_cache = {}
 
+        # Pocket TTS ships a separate model per language (english, french,
+        # german, ...) and load_model() falls back to English when no language
+        # is given, so the option has to reach the loader to pick another one.
+        load_kwargs = {}
+        language = self.options.get("language")
+        if language:
+            load_kwargs["language"] = str(language)
+
         try:
-            print("Loading Pocket TTS model", file=sys.stderr)
-            self.tts_model = TTSModel.load_model()
+            print(f"Loading Pocket TTS model (language: {load_kwargs.get('language', 'default')})", file=sys.stderr)
+            self.tts_model = TTSModel.load_model(**load_kwargs)
             print(f"Model loaded successfully. Sample rate: {self.tts_model.sample_rate}", file=sys.stderr)
 
             # Pre-load default voice if specified
diff --git a/backend/python/pocket-tts/test.py b/backend/python/pocket-tts/test.py
index 34efa1080..8f9b18ba1 100644
--- a/backend/python/pocket-tts/test.py
+++ b/backend/python/pocket-tts/test.py
@@ -6,6 +6,7 @@ import subprocess
 import time
 import os
 import tempfile
+from unittest import mock
 import backend_pb2
 import backend_pb2_grpc
 
@@ -139,3 +140,33 @@ class TestBackendServicer(unittest.TestCase):
             self.fail("TTS service with default voice failed")
         finally:
             self.tearDown()
+
+
+class TestLanguageOption(unittest.TestCase):
+    """
+    Checks that the language option reaches TTSModel.load_model without
+    downloading any weights.
+    """
+    def _load_model(self, options):
+        import backend
+
+        fake_model = mock.MagicMock()
+        fake_model.sample_rate = 24000
+        with mock.patch.object(backend.TTSModel, "load_model", return_value=fake_model) as load_model:
+            response = backend.BackendServicer().LoadModel(
+                backend_pb2.ModelOptions(Options=options), None
+            )
+        self.assertTrue(response.success, response.message)
+        return load_model
+
+    def test_language_option_selects_model(self):
+        load_model = self._load_model(["language:french"])
+        load_model.assert_called_once_with(language="french")
+
+    def test_versioned_language_option(self):
+        load_model = self._load_model(["language:english_2026-04"])
+        load_model.assert_called_once_with(language="english_2026-04")
+
+    def test_no_language_keeps_library_default(self):
+        load_model = self._load_model(["default_voice:azelma"])
+        load_model.assert_called_once_with()
diff --git a/docs/content/features/text-to-audio.md b/docs/content/features/text-to-audio.md
index 4b7140600..8d070e1e3 100644
--- a/docs/content/features/text-to-audio.md
+++ b/docs/content/features/text-to-audio.md
@@ -701,6 +701,19 @@ curl http://localhost:8080/tts -H "Content-Type: application/json" -d '{
    }' | aplay
 ```
 
+#### Languages
+
+Pocket TTS has a separate model for each language and loads the English one by default. Set the `language` option to load another:
+
+```yaml
+name: pocket-tts-french
+backend: pocket-tts
+options:
+  - "language:french"
+```
+
+Supported values depend on the installed `pocket-tts` version; current releases accept `english`, `french`, `german`, `italian`, `spanish`, `portuguese` and `dutch`, plus variants such as `french_24l` and `english_2026-04`. The built-in voice names are resolved against the selected language model.
+
 ### Qwen3-TTS
 
 [Qwen3-TTS](https://github.com/QwenLM/Qwen3-TTS) is a high-quality text-to-speech model that supports three modes: custom voice (predefined speakers), voice design (natural language instructions), and voice cloning (from reference audio).
diff --git a/gallery/pocket-tts.yaml b/gallery/pocket-tts.yaml
index f4edb130b..63c035f17 100644
--- a/gallery/pocket-tts.yaml
+++ b/gallery/pocket-tts.yaml
@@ -35,3 +35,6 @@ config_file: |-
       # If set, this voice will be loaded when the model loads for faster first generation
       - "default_voice:azelma"
       # - "default_voice:hf://kyutai/tts-voices/alba-mackenna/casual.wav"
+      # Language model to load (defaults to English), e.g. french, german,
+      # italian, spanish, portuguese, dutch
+      # - "language:french"