[Partner Nodes] feat: add Runway Aleph2 node (#14306)

Signed-off-by: bigcat88 <bigcat88@icloud.com>

comfy_api_nodes/apis/runway.py

@@ -67,15 +67,6 @@ class RunwayImageToVideoResponse(BaseModel):
     id: Optional[str] = Field(None, description='Task ID')
 
 
-class RunwayTaskStatusEnum(str, Enum):
-    SUCCEEDED = 'SUCCEEDED'
-    RUNNING = 'RUNNING'
-    FAILED = 'FAILED'
-    PENDING = 'PENDING'
-    CANCELLED = 'CANCELLED'
-    THROTTLED = 'THROTTLED'
-
-
 class RunwayTaskStatusResponse(BaseModel):
     createdAt: datetime = Field(..., description='Task creation timestamp')
     id: str = Field(..., description='Task ID')
@@ -86,7 +77,7 @@ class RunwayTaskStatusResponse(BaseModel):
         ge=0.0,
         le=1.0,
     )
-    status: RunwayTaskStatusEnum
+    status: str = Field(..., description="SUCCEEDED, RUNNING, FAILED, PENDING, CANCELLED or THROTTLED")
 
 
 class Model4(str, Enum):
@@ -125,3 +116,144 @@ class RunwayTextToImageRequest(BaseModel):
 
 class RunwayTextToImageResponse(BaseModel):
     id: Optional[str] = Field(None, description='Task ID')
+
+
+class RunwayAleph2IO:
+    """Custom socket types for chaining Aleph2 guidance images."""
+
+    KEYFRAME = "RUNWAY_ALEPH2_KEYFRAME"
+    PROMPT_IMAGE = "RUNWAY_ALEPH2_PROMPT_IMAGE"
+
+
+# Keyframe timing modes (anchored to the INPUT video). Stored on the chain item and used to
+# choose the request model below. The values match the Aleph2 keyframe union field names.
+KEYFRAME_MODE_SECONDS = "seconds"  # absolute time, in seconds, from the start of the input video
+KEYFRAME_MODE_AT = "at"  # fraction [0.0, 1.0] of the input video duration
+
+# Prompt-image position modes (anchored to the OUTPUT video). Values match the Aleph2 position `type`.
+PROMPT_IMAGE_MODE_TIMESTAMP = "timestamp"  # absolute time, in seconds, from the start of the output video
+PROMPT_IMAGE_MODE_POSITION = "position"  # fraction [0.0, 1.0] of the output video duration
+
+
+class RunwayAleph2KeyframeItem:
+    """A guidance image anchored to a point of the INPUT video (one Aleph2 ``keyframe``)."""
+
+    def __init__(self, image, mode: str, value: float):
+        self.image = image
+        self.mode = mode  # KEYFRAME_MODE_SECONDS | KEYFRAME_MODE_AT
+        self.value = value
+
+
+class RunwayAleph2KeyframeChain:
+    """An ordered collection of keyframes, built by chaining Runway Aleph2 Keyframe nodes."""
+
+    def __init__(self):
+        self.items: list[RunwayAleph2KeyframeItem] = []
+
+    def add(self, item: RunwayAleph2KeyframeItem) -> None:
+        self.items.append(item)
+
+    def clone(self) -> "RunwayAleph2KeyframeChain":
+        c = RunwayAleph2KeyframeChain()
+        c.items = list(self.items)
+        return c
+
+
+class RunwayAleph2PromptImageItem:
+    """A guidance image anchored to a point of the OUTPUT video (one Aleph2 ``promptImage``)."""
+
+    def __init__(self, image, mode: str, value: float):
+        self.image = image
+        self.mode = mode  # PROMPT_IMAGE_MODE_TIMESTAMP | PROMPT_IMAGE_MODE_POSITION
+        self.value = value
+
+
+class RunwayAleph2PromptImageChain:
+    """An ordered collection of prompt images, built by chaining Runway Aleph2 Prompt Image nodes."""
+
+    def __init__(self):
+        self.items: list[RunwayAleph2PromptImageItem] = []
+
+    def add(self, item: RunwayAleph2PromptImageItem) -> None:
+        self.items.append(item)
+
+    def clone(self) -> "RunwayAleph2PromptImageChain":
+        c = RunwayAleph2PromptImageChain()
+        c.items = list(self.items)
+        return c
+
+
+class RunwayAleph2KeyframeSeconds(BaseModel):
+    seconds: float = Field(
+        ...,
+        description="Absolute timestamp in seconds from the start of the input video when this guidance image should apply.",
+        ge=0.0,
+    )
+    uri: str = Field(...)
+
+
+class RunwayAleph2KeyframeAt(BaseModel):
+    at: float = Field(
+        ...,
+        description="Position as a fraction [0.0, 1.0] of the input video duration.",
+        ge=0.0,
+        le=1.0,
+    )
+    uri: str = Field(...)
+
+
+class RunwayAleph2TimestampPosition(BaseModel):
+    type: str = Field(default="timestamp")
+    timestampSeconds: float = Field(
+        ...,
+        description="Absolute timestamp in seconds from the start of the output video.",
+        ge=0.0,
+    )
+
+
+class RunwayAleph2RelativePosition(BaseModel):
+    type: str = Field(default="position")
+    positionPercentage: float = Field(
+        ...,
+        description="Position as a fraction [0.0, 1.0] of the total output video duration.",
+        ge=0.0,
+        le=1.0,
+    )
+
+
+class RunwayAleph2PromptImage(BaseModel):
+    position: RunwayAleph2TimestampPosition | RunwayAleph2RelativePosition
+    uri: str = Field(...)
+
+
+class RunwayAleph2ContentModeration(BaseModel):
+    publicFigureThreshold: str = Field(
+        ...,
+        description='When set to "low", the content moderation system is less strict about '
+        'recognizable public figures. One of "auto" or "low".',
+    )
+
+
+class RunwayAleph2Request(BaseModel):
+    model: str = Field(default="aleph2")
+    promptText: str = Field(
+        ...,
+        description="A non-empty string describing what should appear in the output.",
+        min_length=1,
+        max_length=1000,
+    )
+    videoUri: str = Field(...)
+    seed: int = Field(..., description="Random seed for generation", ge=0, le=4294967295)
+    contentModeration: RunwayAleph2ContentModeration = Field(...)
+    keyframes: list[RunwayAleph2KeyframeSeconds | RunwayAleph2KeyframeAt] | None = Field(
+        None,
+        description="Timed guidance images placed at specific points in the input video. Up to 5.",
+    )
+    promptImage: list[RunwayAleph2PromptImage] | None = Field(
+        None,
+        description="Up to 5 image keyframes for guiding the edit at specific points in the output video.",
+    )
+
+
+class RunwayAleph2Response(BaseModel):
+    id: str | None = Field(None, description="Task ID")

comfy_api_nodes/nodes_runway.py

@@ -30,13 +30,33 @@ from comfy_api_nodes.apis.runway import (
     Model4,
     ReferenceImage,
     RunwayTextToImageAspectRatioEnum,
+    RunwayAleph2IO,
+    RunwayAleph2KeyframeChain,
+    RunwayAleph2KeyframeItem,
+    RunwayAleph2PromptImageChain,
+    RunwayAleph2PromptImageItem,
+    RunwayAleph2Request,
+    RunwayAleph2Response,
+    RunwayAleph2KeyframeSeconds,
+    RunwayAleph2KeyframeAt,
+    RunwayAleph2PromptImage,
+    RunwayAleph2TimestampPosition,
+    RunwayAleph2RelativePosition,
+    RunwayAleph2ContentModeration,
+    KEYFRAME_MODE_SECONDS,
+    KEYFRAME_MODE_AT,
+    PROMPT_IMAGE_MODE_TIMESTAMP,
+    PROMPT_IMAGE_MODE_POSITION,
 )
 from comfy_api_nodes.util import (
     image_tensor_pair_to_batch,
     validate_string,
     validate_image_dimensions,
     validate_image_aspect_ratio,
+    validate_video_duration,
     upload_images_to_comfyapi,
+    upload_image_to_comfyapi,
+    upload_video_to_comfyapi,
     download_url_to_video_output,
     download_url_to_image_tensor,
     ApiEndpoint,
@@ -45,6 +65,7 @@ from comfy_api_nodes.util import (
 )
 
 PATH_IMAGE_TO_VIDEO = "/proxy/runway/image_to_video"
+PATH_VIDEO_TO_VIDEO = "/proxy/runway/video_to_video"
 PATH_TEXT_TO_IMAGE = "/proxy/runway/text_to_image"
 PATH_GET_TASK_STATUS = "/proxy/runway/tasks"
 
@@ -53,12 +74,6 @@ AVERAGE_DURATION_FLF_SECONDS = 256
 AVERAGE_DURATION_T2I_SECONDS = 41
 
 
-class RunwayApiError(Exception):
-    """Base exception for Runway API errors."""
-
-    pass
-
-
 class RunwayGen4TurboAspectRatio(str, Enum):
     """Aspect ratios supported for Image to Video API when using gen4_turbo model."""
 
@@ -84,14 +99,6 @@ def get_video_url_from_task_status(response: TaskStatusResponse) -> str | None:
     return None
 
 
-def extract_progress_from_task_status(
-    response: TaskStatusResponse,
-) -> float | None:
-    if hasattr(response, "progress") and response.progress is not None:
-        return response.progress * 100
-    return None
-
-
 def get_image_url_from_task_status(response: TaskStatusResponse) -> str | None:
     """Returns the image URL from the task status response if it exists."""
     if hasattr(response, "output") and len(response.output) > 0:
@@ -102,14 +109,13 @@ def get_image_url_from_task_status(response: TaskStatusResponse) -> str | None:
 async def get_response(
     cls: type[IO.ComfyNode], task_id: str, estimated_duration: int | None = None
 ) -> TaskStatusResponse:
-    """Poll the task status until it is finished then get the response."""
     return await poll_op(
         cls,
         ApiEndpoint(path=f"{PATH_GET_TASK_STATUS}/{task_id}"),
         response_model=TaskStatusResponse,
-        status_extractor=lambda r: r.status.value,
+        status_extractor=lambda r: r.status,
         estimated_duration=estimated_duration,
-        progress_extractor=extract_progress_from_task_status,
+        progress_extractor=lambda r: r.progress * 100 if r.progress is not None else None,
     )
 
 
@@ -127,7 +133,7 @@ async def generate_video(
 
     final_response = await get_response(cls, initial_response.id, estimated_duration)
     if not final_response.output:
-        raise RunwayApiError("Runway task succeeded but no video data found in response.")
+        raise ValueError("Runway task succeeded but no video data found in response.")
 
     video_url = get_video_url_from_task_status(final_response)
     return await download_url_to_video_output(video_url)
@@ -410,7 +416,7 @@ class RunwayFirstLastFrameNode(IO.ComfyNode):
             mime_type="image/png",
         )
         if len(download_urls) != 2:
-            raise RunwayApiError("Failed to upload one or more images to comfy api.")
+            raise ValueError("Failed to upload one or more images to comfy api.")
 
         return IO.NodeOutput(
             await generate_video(
@@ -514,11 +520,321 @@ class RunwayTextToImageNode(IO.ComfyNode):
             estimated_duration=AVERAGE_DURATION_T2I_SECONDS,
         )
         if not final_response.output:
-            raise RunwayApiError("Runway task succeeded but no image data found in response.")
+            raise ValueError("Runway task succeeded but no image data found in response.")
 
         return IO.NodeOutput(await download_url_to_image_tensor(get_image_url_from_task_status(final_response)))
 
 
+_TIMING_ABSOLUTE = "Absolute time (seconds)"
+_TIMING_FRACTION = "Fraction of duration (0.0-1.0)"
+
+
+class RunwayAleph2KeyframeNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RunwayAleph2KeyframeNode",
+            display_name="Runway Aleph2 Keyframe",
+            category="partner/video/Runway",
+            description="Anchor a guidance image to a moment of the input (source) video, so Aleph2 "
+            "steers the edit at that point of your footage. Connect this to the 'keyframes' input of "
+            "the Runway Aleph2 Video to Video node; chain several together (up to 5) via the optional "
+            "'keyframes' input below.",
+            inputs=[
+                IO.Image.Input(
+                    "image",
+                    tooltip="The guidance image to apply at the chosen moment of the input video.",
+                ),
+                IO.DynamicCombo.Input(
+                    "timing",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            _TIMING_ABSOLUTE,
+                            [
+                                IO.Float.Input(
+                                    "seconds",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=30.0,
+                                    step=0.1,
+                                    display_mode=IO.NumberDisplay.number,
+                                    tooltip="Time in seconds from start of the input video where this image applies.",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            _TIMING_FRACTION,
+                            [
+                                IO.Float.Input(
+                                    "fraction",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=1.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.number,
+                                    tooltip="Where in the input video this image applies, "
+                                    "as a fraction of its duration (0.0 = start, 1.0 = end).",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="How to place this image on the input video's timeline.",
+                ),
+                IO.Custom(RunwayAleph2IO.KEYFRAME).Input(
+                    "keyframes",
+                    optional=True,
+                    tooltip="Optional earlier keyframes to chain with this one.",
+                ),
+            ],
+            outputs=[IO.Custom(RunwayAleph2IO.KEYFRAME).Output(display_name="keyframes")],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        image: Input.Image,
+        timing: dict,
+        keyframes: RunwayAleph2KeyframeChain | None = None,
+    ) -> IO.NodeOutput:
+        chain = keyframes.clone() if keyframes is not None else RunwayAleph2KeyframeChain()
+        if timing["timing"] == _TIMING_ABSOLUTE:
+            mode, value = KEYFRAME_MODE_SECONDS, float(timing["seconds"])
+        else:
+            mode, value = KEYFRAME_MODE_AT, float(timing["fraction"])
+        chain.add(RunwayAleph2KeyframeItem(image=image, mode=mode, value=value))
+        return IO.NodeOutput(chain)
+
+
+class RunwayAleph2PromptImageNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RunwayAleph2PromptImageNode",
+            display_name="Runway Aleph2 Prompt Image",
+            category="partner/video/Runway",
+            description="Anchor a guidance image to a moment of the output (result) video, to guide what "
+            "the edited video looks like at that point. Connect this to the 'prompt_images' input of the "
+            "Runway Aleph2 Video to Video node; chain several together (up to 5) via the optional "
+            "'prompt_images' input below.",
+            inputs=[
+                IO.Image.Input(
+                    "image",
+                    tooltip="The guidance image to place at the chosen moment of the output video.",
+                ),
+                IO.DynamicCombo.Input(
+                    "position",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            _TIMING_ABSOLUTE,
+                            [
+                                IO.Float.Input(
+                                    "seconds",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=30.0,
+                                    step=0.1,
+                                    display_mode=IO.NumberDisplay.number,
+                                    tooltip="Time in seconds from start of the output video where this image applies.",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            _TIMING_FRACTION,
+                            [
+                                IO.Float.Input(
+                                    "fraction",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=1.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.number,
+                                    tooltip="Where in the output video this image applies, "
+                                    "as a fraction of its duration (0.0 = start, 1.0 = end).",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="How to place this image on the output video's timeline.",
+                ),
+                IO.Custom(RunwayAleph2IO.PROMPT_IMAGE).Input(
+                    "prompt_images",
+                    optional=True,
+                    tooltip="Optional earlier prompt images to chain with this one.",
+                ),
+            ],
+            outputs=[IO.Custom(RunwayAleph2IO.PROMPT_IMAGE).Output(display_name="prompt_images")],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        image: Input.Image,
+        position: dict,
+        prompt_images: RunwayAleph2PromptImageChain | None = None,
+    ) -> IO.NodeOutput:
+        chain = prompt_images.clone() if prompt_images is not None else RunwayAleph2PromptImageChain()
+        if position["position"] == _TIMING_ABSOLUTE:
+            mode, value = PROMPT_IMAGE_MODE_TIMESTAMP, float(position["seconds"])
+        else:
+            mode, value = PROMPT_IMAGE_MODE_POSITION, float(position["fraction"])
+        chain.add(RunwayAleph2PromptImageItem(image=image, mode=mode, value=value))
+        return IO.NodeOutput(chain)
+
+
+class RunwayAleph2VideoToVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RunwayAleph2VideoToVideoNode",
+            display_name="Runway Aleph2 Video to Video",
+            category="partner/video/Runway",
+            description="Edit a video with a text prompt using Runway's Aleph2 model. Aleph2 transforms "
+            "your footage (restyle, relight, add or remove elements, change the viewpoint) while keeping "
+            "the original motion and timing; the output resolution matches the input video, which must be "
+            "2-30 seconds at 30 fps or lower. Optionally steer the edit with either keyframes (anchored to "
+            "the input video) or prompt images (anchored to the output video) - use one or the other, not both.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Describes what should appear in the output (1-1000 characters).",
+                ),
+                IO.Video.Input(
+                    "video",
+                    tooltip="Input video to edit. Must be 2-30 seconds at 30 fps or lower.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=4294967295,
+                    step=1,
+                    control_after_generate=True,
+                    display_mode=IO.NumberDisplay.number,
+                    tooltip="Random seed for generation",
+                ),
+                IO.Combo.Input(
+                    "public_figure_threshold",
+                    options=["auto", "low"],
+                    default="low",
+                    tooltip="Content moderation for recognizable public figures.",
+                ),
+                IO.Custom(RunwayAleph2IO.KEYFRAME).Input(
+                    "keyframes",
+                    optional=True,
+                    tooltip="Guidance images anchored to the input video, from Aleph2 Keyframe nodes (up to 5). "
+                    "Use keyframes or prompt images, not both.",
+                ),
+                IO.Custom(RunwayAleph2IO.PROMPT_IMAGE).Input(
+                    "prompt_images",
+                    optional=True,
+                    tooltip="Guidance images anchored to the output video, from Aleph2 Prompt Image nodes (up to 5). "
+                    "Use keyframes or prompt images, not both.",
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+            price_badge=IO.PriceBadge(
+                expr="""{"type":"usd","usd": 0.4004, "format":{"suffix":"/second"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        video: Input.Video,
+        seed: int,
+        public_figure_threshold: str = "low",
+        keyframes: RunwayAleph2KeyframeChain | None = None,
+        prompt_images: RunwayAleph2PromptImageChain | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=1000)
+        validate_video_duration(
+            video,
+            min_duration=2.0,
+            max_duration=30.0,
+        )
+        try:
+            fps = float(video.get_frame_rate())
+        except Exception:
+            fps = None
+        if fps is not None and fps > 30.0 + 0.01:
+            raise ValueError(f"Input video frame rate ({fps:.2f} fps) exceeds Aleph2's maximum of 30 fps.")
+
+        if (keyframes and keyframes.items) and (prompt_images and prompt_images.items):
+            raise ValueError("Aleph2 accepts either keyframes or prompt images, not both.")
+
+        video_duration: float | None = None
+        try:
+            video_duration = video.get_duration()
+        except Exception:
+            video_duration = None
+
+        def _check_seconds(value: float, label: str) -> None:
+            if video_duration is not None and value > video_duration + 0.0001:
+                raise ValueError(f"{label} {value:.2f}s exceeds the input video duration ({video_duration:.2f}s).")
+
+        video_url = await upload_video_to_comfyapi(cls, video)
+
+        keyframe_models: list[RunwayAleph2KeyframeSeconds | RunwayAleph2KeyframeAt] = []
+        if keyframes is not None:
+            if len(keyframes.items) > 5:
+                raise ValueError("Aleph2 supports at most 5 keyframes.")
+            for item in keyframes.items:
+                image_url = await upload_image_to_comfyapi(cls, item.image, mime_type="image/png")
+                if item.mode == KEYFRAME_MODE_SECONDS:
+                    _check_seconds(item.value, "Keyframe timestamp")
+                    keyframe_models.append(RunwayAleph2KeyframeSeconds(seconds=item.value, uri=image_url))
+                else:
+                    keyframe_models.append(RunwayAleph2KeyframeAt(at=item.value, uri=image_url))
+
+        prompt_image_models: list[RunwayAleph2PromptImage] = []
+        if prompt_images is not None:
+            if len(prompt_images.items) > 5:
+                raise ValueError("Aleph2 supports at most 5 prompt images.")
+            for item in prompt_images.items:
+                image_url = await upload_image_to_comfyapi(cls, item.image, mime_type="image/png")
+                position: RunwayAleph2TimestampPosition | RunwayAleph2RelativePosition
+                if item.mode == PROMPT_IMAGE_MODE_TIMESTAMP:
+                    _check_seconds(item.value, "Prompt image timestamp")
+                    position = RunwayAleph2TimestampPosition(timestampSeconds=item.value)
+                else:
+                    position = RunwayAleph2RelativePosition(positionPercentage=item.value)
+                prompt_image_models.append(RunwayAleph2PromptImage(position=position, uri=image_url))
+
+        initial_response = await sync_op(
+            cls,
+            endpoint=ApiEndpoint(path=PATH_VIDEO_TO_VIDEO, method="POST"),
+            response_model=RunwayAleph2Response,
+            data=RunwayAleph2Request(
+                promptText=prompt,
+                videoUri=video_url,
+                seed=seed,
+                contentModeration=RunwayAleph2ContentModeration(publicFigureThreshold=public_figure_threshold),
+                keyframes=keyframe_models or None,
+                promptImage=prompt_image_models or None,
+            ),
+        )
+
+        final_response = await get_response(cls, initial_response.id)
+        if not final_response.output:
+            raise ValueError("Runway task succeeded but no video data found in response.")
+
+        return IO.NodeOutput(await download_url_to_video_output(get_video_url_from_task_status(final_response)))
+
+
 class RunwayExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -527,6 +843,9 @@ class RunwayExtension(ComfyExtension):
             RunwayImageToVideoNodeGen3a,
             RunwayImageToVideoNodeGen4,
             RunwayTextToImageNode,
+            RunwayAleph2VideoToVideoNode,
+            RunwayAleph2KeyframeNode,
+            RunwayAleph2PromptImageNode,
         ]