diff --git a/tests/test_multipart_list_fields.py b/tests/test_multipart_list_fields.py
new file mode 100644
index 00000000..a472f3af
--- /dev/null
+++ b/tests/test_multipart_list_fields.py
@@ -0,0 +1,165 @@
+"""Regression tests for list-of-primitive multipart form fields.
+
+Generator ``fern-python-sdk`` 4.46.9 started serializing every non-scalar
+multipart form field with ``json.dumps(jsonable_encoder(...))``. ``List[str]``
+fields are not scalars, so ``speech_to_text.convert(keyterms=["a", "b"])`` went
+over the wire as a single ``["a", "b"]`` string instead of two ``keyterms``
+form fields, and the API validated the whole JSON array against its per-item
+limits (``invalid_keyword_length``).
+
+See https://github.com/elevenlabs/elevenlabs-python/issues/819 and the fix in
+https://github.com/elevenlabs/elevenlabs-python/pull/825. The fields live in
+Fern-generated ``raw_client.py`` files and the fix is re-applied on every
+regeneration by Fern Replay, so these tests guard against a regeneration
+silently reverting it.
+"""
+
+import email
+import email.message
+import json
+import typing
+
+import httpx
+import pytest
+
+from elevenlabs.client import AsyncElevenLabs, ElevenLabs
+
+KEYTERMS = ["hello world", "ElevenLabs"]
+TAGS = ["ambient", "lo-fi"]
+GENRES = ["fiction", "sci-fi"]
+LOCATORS = ['{"pronunciation_dictionary_id": "abc", "version_id": "def"}']
+
+
+class _RequestCaptured(Exception):
+    """Raised from the mock transport once the outgoing request is recorded."""
+
+    def __init__(self, request: httpx.Request) -> None:
+        super().__init__("request captured")
+        self.request = request
+
+
+def _capture_sync(call: typing.Callable[[ElevenLabs], typing.Any]) -> httpx.Request:
+    """Run ``call`` against a client whose transport records and aborts the request."""
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        request.read()
+        raise _RequestCaptured(request)
+
+    client = ElevenLabs(api_key="sk-test", httpx_client=httpx.Client(transport=httpx.MockTransport(handler)))
+    with pytest.raises(_RequestCaptured) as excinfo:
+        call(client)
+    return excinfo.value.request
+
+
+async def _capture_async(call: typing.Callable[[AsyncElevenLabs], typing.Awaitable[typing.Any]]) -> httpx.Request:
+    """Async counterpart of :func:`_capture_sync`."""
+
+    async def handler(request: httpx.Request) -> httpx.Response:
+        await request.aread()
+        raise _RequestCaptured(request)
+
+    client = AsyncElevenLabs(api_key="sk-test", httpx_client=httpx.AsyncClient(transport=httpx.MockTransport(handler)))
+    with pytest.raises(_RequestCaptured) as excinfo:
+        await call(client)
+    return excinfo.value.request
+
+
+def _form_fields(request: httpx.Request) -> typing.List[typing.Tuple[str, str]]:
+    """Parse a multipart request body into an ordered list of (name, value) pairs."""
+    content_type = request.headers["content-type"]
+    assert content_type.startswith("multipart/form-data"), content_type
+
+    message = email.message_from_bytes(b"Content-Type: " + content_type.encode() + b"\r\n\r\n" + request.content)
+    fields: typing.List[typing.Tuple[str, str]] = []
+    for part in typing.cast(typing.List[email.message.Message], message.get_payload()):
+        name = part.get_param("name", header="content-disposition")
+        payload = part.get_payload(decode=True)
+        if name is None or not isinstance(payload, bytes):
+            continue
+        fields.append((str(name), payload.decode("utf-8")))
+    return fields
+
+
+def _values(request: httpx.Request, name: str) -> typing.List[str]:
+    return [value for field_name, value in _form_fields(request) if field_name == name]
+
+
+def _stt_convert(client: ElevenLabs) -> typing.Any:
+    return client.speech_to_text.convert(model_id="scribe_v1", keyterms=KEYTERMS)
+
+
+async def _stt_convert_async(client: AsyncElevenLabs) -> typing.Any:
+    return await client.speech_to_text.convert(model_id="scribe_v1", keyterms=KEYTERMS)
+
+
+def test_speech_to_text_keyterms_are_repeated_form_fields():
+    """The exact reproduction from #819: keyterms must not be one JSON-array field."""
+    request = _capture_sync(_stt_convert)
+    assert _values(request, "keyterms") == KEYTERMS
+
+
+def test_speech_to_text_keyterms_are_not_json_encoded():
+    """A single field holding ``'["hello world", "ElevenLabs"]'`` is the bug."""
+    request = _capture_sync(_stt_convert)
+    assert json.dumps(KEYTERMS) not in [value for _, value in _form_fields(request)]
+
+
+async def test_speech_to_text_keyterms_are_repeated_form_fields_async():
+    request = await _capture_async(_stt_convert_async)
+    assert _values(request, "keyterms") == KEYTERMS
+
+
+def test_speech_to_text_object_fields_stay_json_encoded():
+    """Only lists of primitives changed; object fields are still JSON blobs."""
+    request = _capture_sync(
+        lambda client: client.speech_to_text.convert(
+            model_id="scribe_v1",
+            keyterms=KEYTERMS,
+            entity_detection={"entity_types": ["person"]},
+        )
+    )
+    assert _values(request, "entity_detection") == [json.dumps({"entity_types": ["person"]})]
+
+
+def test_dubbing_project_keyterms_are_repeated_form_fields():
+    request = _capture_sync(
+        lambda client: client.dubbing.project.create(source_url="https://e.com/a.mp4", keyterms=KEYTERMS)
+    )
+    assert _values(request, "keyterms") == KEYTERMS
+
+
+def test_music_video_to_music_tags_are_repeated_form_fields():
+    def call(client: ElevenLabs) -> typing.Any:
+        with client.music.with_raw_response.video_to_music(videos=[("clip.mp4", b"video")], tags=TAGS):
+            pass
+
+    request = _capture_sync(call)
+    assert _values(request, "tags") == TAGS
+
+
+def test_music_finetunes_tags_are_repeated_form_fields():
+    request = _capture_sync(
+        lambda client: client.music.finetunes.create(name="My finetune", primary_genre="rock", tags=TAGS)
+    )
+    assert _values(request, "tags") == TAGS
+
+
+def test_studio_projects_list_fields_are_repeated_form_fields():
+    request = _capture_sync(
+        lambda client: client.studio.projects.create(
+            name="My project",
+            genres=GENRES,
+            pronunciation_dictionary_locators=LOCATORS,
+            voice_settings=['{"stability": 0.5}'],
+        )
+    )
+    assert _values(request, "genres") == GENRES
+    assert _values(request, "pronunciation_dictionary_locators") == LOCATORS
+    assert _values(request, "voice_settings") == ['{"stability": 0.5}']
+
+
+def test_audio_native_pronunciation_dictionary_locators_are_repeated_form_fields():
+    request = _capture_sync(
+        lambda client: client.audio_native.create(name="My project", pronunciation_dictionary_locators=LOCATORS)
+    )
+    assert _values(request, "pronunciation_dictionary_locators") == LOCATORS