feat: video-to-video endpoints + v2m ducking/preserve_speech/output_format

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sonilo"
-version = "0.3.0"
+version = "0.4.0"
 description = "Official Python client for the Sonilo API"
 readme = "README.md"
 license = "MIT"

sonilo-video-kit/pyproject.toml

@@ -10,7 +10,7 @@ readme = "README.md"
 license = "MIT"
 requires-python = ">=3.9"
 authors = [{ name = "Sonilo AI" }]
-dependencies = ["sonilo>=0.3,<0.4"]
+dependencies = ["sonilo>=0.3,<0.5"]
 keywords = ["sonilo", "video", "music", "ffmpeg", "soundtrack", "ducking", "ai"]
 
 [project.urls]

src/sonilo/__init__.py

@@ -23,6 +23,7 @@
     SfxTask,
     StreamEvent,
     Track,
+    VideoResult,
 )
 
 __all__ = [
@@ -47,5 +48,6 @@
     "TaskFailedError",
     "TaskTimeoutError",
     "Track",
+    "VideoResult",
     "__version__",
 ]

src/sonilo/_async_client.py

@@ -13,6 +13,8 @@
 from sonilo.resources.text_to_sfx import AsyncTextToSfx
 from sonilo.resources.video_to_music import AsyncVideoToMusic
 from sonilo.resources.video_to_sfx import AsyncVideoToSfx
+from sonilo.resources.video_to_video_music import AsyncVideoToVideoMusic
+from sonilo.resources.video_to_video_sfx import AsyncVideoToVideoSfx
 from sonilo.types import StreamEvent
 
 
@@ -35,6 +37,8 @@ def __init__(
         self.video_to_music = AsyncVideoToMusic(self)
         self.text_to_sfx = AsyncTextToSfx(self)
         self.video_to_sfx = AsyncVideoToSfx(self)
+        self.video_to_video_music = AsyncVideoToVideoMusic(self)
+        self.video_to_video_sfx = AsyncVideoToVideoSfx(self)
         self.account = AsyncAccount(self)
         self.tasks = AsyncTasks(self)
 

src/sonilo/_client.py

@@ -14,6 +14,8 @@
 from sonilo.resources.text_to_sfx import TextToSfx
 from sonilo.resources.video_to_music import VideoToMusic
 from sonilo.resources.video_to_sfx import VideoToSfx
+from sonilo.resources.video_to_video_music import VideoToVideoMusic
+from sonilo.resources.video_to_video_sfx import VideoToVideoSfx
 from sonilo.types import StreamEvent
 
 DEFAULT_BASE_URL = "https://api.sonilo.com"
@@ -56,6 +58,8 @@ def __init__(
         self.video_to_music = VideoToMusic(self)
         self.text_to_sfx = TextToSfx(self)
         self.video_to_sfx = VideoToSfx(self)
+        self.video_to_video_music = VideoToVideoMusic(self)
+        self.video_to_video_sfx = VideoToVideoSfx(self)
         self.account = Account(self)
         self.tasks = Tasks(self)
 

src/sonilo/_requests.py

@@ -19,6 +19,23 @@ def build_t2m_data(
     return data
 
 
+def build_t2m_async_data(
+    prompt: str,
+    duration: int,
+    segments: Optional[List[Segment]],
+    mode: Optional[str],
+    output_format: Optional[str],
+) -> Dict[str, str]:
+    data = build_t2m_data(prompt, duration, segments)
+    resolved = mode or "async"
+    if resolved != "async":
+        raise SoniloError("text-to-music submit() requires mode='async'")
+    data["mode"] = resolved
+    if output_format is not None:
+        data["output_format"] = output_format
+    return data
+
+
 def normalize_video(video: Any) -> Tuple[str, Any, bool]:
     """Normalize a video input into (filename, httpx-uploadable, opened_here).
 
@@ -66,18 +83,31 @@ def build_v2m_parts(
     return data, files, opened
 
 
-def _resolve_music_mode(mode: Optional[str], isolate_vocals: Optional[bool]) -> str:
-    """isolate_vocals only works with async processing: auto-select mode
-    "async" when the caller didn't specify one, but fail fast (mirroring the
-    video/video_url XOR check above) if they explicitly asked for anything
-    else. Without isolate_vocals, submit() still needs an async response
-    (a task_id ack, not a stream), so "async" is also the default there.
+def _resolve_music_mode(
+    mode: Optional[str],
+    isolate_vocals: Optional[bool],
+    preserve_speech: Optional[bool] = None,
+    output_format: Optional[str] = None,
+    ducking: Optional[bool] = None,
+) -> str:
+    """isolate_vocals/preserve_speech/ducking/output_format='wav' only work
+    with async processing: auto-select mode "async" when the caller didn't
+    specify one, but fail fast if they explicitly asked for anything else.
+    submit() also needs an async response (a task_id ack, not a stream), so
+    "async" is the default regardless.
     """
-    if isolate_vocals:
-        if mode is not None and mode != "async":
-            raise SoniloError("isolate_vocals=True requires mode='async'")
-        return "async"
-    return mode or "async"
+    needs_async = (
+        bool(isolate_vocals)
+        or bool(preserve_speech)
+        or output_format == "wav"
+        or ducking is not None
+    )
+    if needs_async and mode is not None and mode != "async":
+        raise SoniloError(
+            "isolate_vocals/preserve_speech/ducking/output_format='wav' "
+            "require mode='async'"
+        )
+    return "async" if needs_async else (mode or "async")
 
 
 def build_v2m_async_parts(
@@ -87,17 +117,53 @@ def build_v2m_async_parts(
     segments: Optional[List[Segment]],
     mode: Optional[str],
     isolate_vocals: Optional[bool],
+    preserve_speech: Optional[bool] = None,
+    output_format: Optional[str] = None,
+    ducking: Optional[bool] = None,
 ) -> Tuple[Dict[str, str], Optional[Dict[str, tuple]], bool]:
-    """Like build_v2m_parts, plus the async-only `mode`/`isolate_vocals`
-    fields used by the video-to-music submit()/generate_async() path."""
-    resolved_mode = _resolve_music_mode(mode, isolate_vocals)
+    """Like build_v2m_parts, plus the async-only fields for the
+    video-to-music submit()/generate_async() path."""
+    resolved_mode = _resolve_music_mode(
+        mode, isolate_vocals, preserve_speech, output_format, ducking
+    )
     data, files, opened = build_v2m_parts(video, video_url, prompt, segments)
     data["mode"] = resolved_mode
     if isolate_vocals is not None:
         data["isolate_vocals"] = "true" if isolate_vocals else "false"
+    if preserve_speech is not None:
+        data["preserve_speech"] = "true" if preserve_speech else "false"
+    if output_format is not None:
+        data["output_format"] = output_format
+    if ducking is not None:
+        data["ducking"] = "true" if ducking else "false"
     return data, files, opened
 
 
+def build_v2v_music_parts(
+    video: Any,
+    video_url: Optional[str],
+    prompt: Optional[str],
+    preserve_speech: Optional[bool],
+    isolate_vocals: Optional[bool],
+) -> Tuple[Dict[str, str], Optional[Dict[str, tuple]], bool]:
+    data, files, opened = build_v2m_parts(video, video_url, prompt, None)
+    if preserve_speech is not None:
+        data["preserve_speech"] = "true" if preserve_speech else "false"
+    if isolate_vocals is not None:
+        data["isolate_vocals"] = "true" if isolate_vocals else "false"
+    return data, files, opened
+
+
+def build_v2v_sfx_parts(
+    video: Any,
+    video_url: Optional[str],
+    prompt: Optional[str],
+    segments: Optional[List[Segment]],
+) -> Tuple[Dict[str, str], Optional[Dict[str, tuple]], bool]:
+    # build_v2m_parts JSON-serializes `segments` — fine for SFX start/end segments too.
+    return build_v2m_parts(video, video_url, prompt, segments)
+
+
 def build_sfx_t2s_data(
     prompt: str, duration: int, audio_format: Optional[str]
 ) -> Dict[str, str]:

src/sonilo/_version.py

@@ -1 +1 @@
-__version__ = "0.3.0"
+__version__ = "0.4.0"

src/sonilo/resources/tasks.py

@@ -6,7 +6,15 @@
 from urllib.parse import quote
 
 from sonilo.errors import SoniloError, TaskFailedError, TaskTimeoutError
-from sonilo.types import MusicAudioMedia, MusicResult, MusicTitle, SfxMedia, SfxResult, SfxTask
+from sonilo.types import (
+    MusicAudioMedia,
+    MusicResult,
+    MusicTitle,
+    SfxMedia,
+    SfxResult,
+    SfxTask,
+    VideoResult,
+)
 
 if TYPE_CHECKING:
     from sonilo._async_client import AsyncSonilo
@@ -106,6 +114,7 @@ def parse_music_result(body: Dict[str, Any]) -> MusicResult:
             audio=_music_audio_list_from(body.get("audio")),
             vocals=_media_from(body.get("vocals")),
             mux=_music_audio_list_from(body.get("mux")),
+            ducked=_music_audio_list_from(body.get("ducked")),
             title=_music_title_from(body.get("title")),
             duration_seconds=body.get("duration_seconds"),
             cost=body.get("cost"),
@@ -116,6 +125,24 @@ def parse_music_result(body: Dict[str, Any]) -> MusicResult:
         raise SoniloError(f"Malformed task response: missing {e.args[0]!r}") from e
 
 
+def parse_video_result(body: Dict[str, Any]) -> "VideoResult":
+    """Map a GET /v1/tasks/{id} body for a video-to-video task to VideoResult;
+    unknown fields are ignored."""
+    try:
+        return VideoResult(
+            task_id=body["task_id"],
+            status=body["status"],
+            type=body.get("type"),
+            video=_media_from(body.get("video")),
+            duration_seconds=body.get("duration_seconds"),
+            cost=body.get("cost"),
+            error=body.get("error"),
+            refunded=body.get("refunded"),
+        )
+    except KeyError as e:
+        raise SoniloError(f"Malformed task response: missing {e.args[0]!r}") from e
+
+
 def parse_sfx_task(body: Dict[str, Any]) -> SfxTask:
     """Map a submission ack to SfxTask."""
     try:

src/sonilo/resources/text_to_music.py

@@ -2,9 +2,15 @@
 
 from typing import TYPE_CHECKING, AsyncIterator, Iterator, List, Optional
 
-from sonilo._requests import build_t2m_data
+from sonilo._requests import build_t2m_async_data, build_t2m_data
 from sonilo._streaming import acollect_track, collect_track
-from sonilo.types import Segment, StreamEvent, Track
+from sonilo.resources.tasks import (
+    DEFAULT_POLL_INTERVAL,
+    DEFAULT_WAIT_TIMEOUT,
+    parse_music_result,
+    parse_sfx_task,
+)
+from sonilo.types import MusicResult, Segment, SfxTask, StreamEvent, Track
 
 if TYPE_CHECKING:
     from sonilo._async_client import AsyncSonilo
@@ -36,6 +42,44 @@ def generate(
     ) -> Track:
         return collect_track(self.stream(prompt=prompt, duration=duration, segments=segments))
 
+    def submit(
+        self,
+        *,
+        prompt: str,
+        duration: int,
+        segments: Optional[List[Segment]] = None,
+        mode: Optional[str] = None,
+        output_format: Optional[str] = None,
+    ) -> SfxTask:
+        """Submit an async text-to-music task; poll with
+        `client.tasks.wait(task_id, parser=sonilo.resources.tasks.parse_music_result)`.
+        Required for output_format="wav". `stream()`/`generate()` remain the
+        streaming path.
+        """
+        data = build_t2m_async_data(prompt, duration, segments, mode, output_format)
+        return parse_sfx_task(self._client._post_json(PATH, data=data))
+
+    def generate_async(
+        self,
+        *,
+        prompt: str,
+        duration: int,
+        segments: Optional[List[Segment]] = None,
+        mode: Optional[str] = None,
+        output_format: Optional[str] = None,
+        poll_interval: float = DEFAULT_POLL_INTERVAL,
+        timeout: float = DEFAULT_WAIT_TIMEOUT,
+    ) -> MusicResult:
+        """submit() + tasks.wait(), returning the parsed MusicResult."""
+        task = self.submit(
+            prompt=prompt, duration=duration, segments=segments,
+            mode=mode, output_format=output_format,
+        )
+        return self._client.tasks.wait(
+            task.task_id, poll_interval=poll_interval, timeout=timeout,
+            parser=parse_music_result,
+        )
+
 
 class AsyncTextToMusic:
     def __init__(self, client: "AsyncSonilo") -> None:
@@ -61,3 +105,41 @@ async def generate(
         return await acollect_track(
             self.stream(prompt=prompt, duration=duration, segments=segments)
         )
+
+    async def submit(
+        self,
+        *,
+        prompt: str,
+        duration: int,
+        segments: Optional[List[Segment]] = None,
+        mode: Optional[str] = None,
+        output_format: Optional[str] = None,
+    ) -> SfxTask:
+        """Submit an async text-to-music task; poll with
+        `client.tasks.wait(task_id, parser=sonilo.resources.tasks.parse_music_result)`.
+        Required for output_format="wav". `stream()`/`generate()` remain the
+        streaming path.
+        """
+        data = build_t2m_async_data(prompt, duration, segments, mode, output_format)
+        return parse_sfx_task(await self._client._post_json(PATH, data=data))
+
+    async def generate_async(
+        self,
+        *,
+        prompt: str,
+        duration: int,
+        segments: Optional[List[Segment]] = None,
+        mode: Optional[str] = None,
+        output_format: Optional[str] = None,
+        poll_interval: float = DEFAULT_POLL_INTERVAL,
+        timeout: float = DEFAULT_WAIT_TIMEOUT,
+    ) -> MusicResult:
+        """submit() + tasks.wait(), returning the parsed MusicResult."""
+        task = await self.submit(
+            prompt=prompt, duration=duration, segments=segments,
+            mode=mode, output_format=output_format,
+        )
+        return await self._client.tasks.wait(
+            task.task_id, poll_interval=poll_interval, timeout=timeout,
+            parser=parse_music_result,
+        )