Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

[Partner Nodes] feat(Kling): add support for Kling V3-Turbo model (#14528)

Alexander Piskun191a75a2cdc0
diff --git a/comfy_api_nodes/apis/kling.py b/comfy_api_nodes/apis/kling.py
index fe0f97cb3..2c98c23b7 100644
--- a/comfy_api_nodes/apis/kling.py
+++ b/comfy_api_nodes/apis/kling.py
@@ -149,3 +149,59 @@ class MotionControlRequest(BaseModel):
character_orientation: str = Field(...)
mode: str = Field(..., description="'pro' or 'std'")
model_name: str = Field(...)
+
+
+class Kling3TurboSettings(BaseModel):
+ resolution: str = Field("720p", description="'720p' or '1080p'")
+ aspect_ratio: str | None = Field(None, description="'16:9'/'9:16'/'1:1'; text-to-video only")
+ duration: int = Field(5, description="3-15 second")
+
+
+class Kling3TurboText2VideoRequest(BaseModel):
+ prompt: str = Field(..., description="<=3072 chars; may use multi-shot 'shot n, m, words; ...'")
+ settings: Kling3TurboSettings | None = Field(None)
+
+
+class Kling3TurboContent(BaseModel):
+ type: str = Field(..., description="'prompt' or 'first_frame'")
+ text: str | None = Field(None, description="for type=prompt; <=2500 chars")
+ url: str | None = Field(None, description="for type=first_frame")
+
+
+class Kling3TurboImage2VideoRequest(BaseModel):
+ contents: list[Kling3TurboContent] = Field(..., description="prompt + first_frame materials")
+ settings: Kling3TurboSettings | None = Field(None)
+
+
+class Kling3TurboCreateData(BaseModel):
+ id: str | None = Field(None, description="Task ID")
+ status: str | None = Field(None)
+ message: str | None = Field(None)
+
+
+class Kling3TurboCreateResponse(BaseModel):
+ code: int | None = Field(None)
+ message: str | None = Field(None)
+ request_id: str | None = Field(None)
+ data: Kling3TurboCreateData | None = Field(None)
+
+
+class Kling3TurboOutput(BaseModel):
+ type: str | None = Field(None, description="'video', 'image', 'audio', ...")
+ id: str | None = Field(None)
+ url: str | None = Field(None)
+ duration: str | None = Field(None)
+
+
+class Kling3TurboTaskData(BaseModel):
+ id: str | None = Field(None)
+ status: str | None = Field(None, description="submitted | processing | succeeded | failed")
+ message: str | None = Field(None)
+ outputs: list[Kling3TurboOutput] | None = Field(None)
+
+
+class Kling3TurboQueryResponse(BaseModel):
+ code: int | None = Field(None)
+ message: str | None = Field(None)
+ request_id: str | None = Field(None)
+ data: list[Kling3TurboTaskData] | None = Field(None)
diff --git a/comfy_api_nodes/nodes_kling.py b/comfy_api_nodes/nodes_kling.py
index c81d3503d..b27de2549 100644
--- a/comfy_api_nodes/nodes_kling.py
+++ b/comfy_api_nodes/nodes_kling.py
@@ -60,6 +60,12 @@ from comfy_api_nodes.apis.kling import (
OmniProImageRequest,
OmniProReferences2VideoRequest,
OmniProText2VideoRequest,
+ Kling3TurboSettings,
+ Kling3TurboText2VideoRequest,
+ Kling3TurboContent,
+ Kling3TurboImage2VideoRequest,
+ Kling3TurboCreateResponse,
+ Kling3TurboQueryResponse,
TaskStatusResponse,
TextToVideoWithAudioRequest,
)
@@ -2847,6 +2853,67 @@ class MotionControl(IO.ComfyNode):
return IO.NodeOutput(await download_url_to_video_output(final_response.data.task_result.videos[0].url))
+def build_turbo_shot_prompt(multi_prompt: list[MultiPromptEntry]) -> str:
+ """Render storyboard entries into the Turbo multi-shot prompt 'shot n, m, words; ...'."""
+ return "; ".join(f"shot {i}, {int(e.duration)}, {e.prompt}" for i, e in enumerate(multi_prompt, 1)) + ";"
+
+
+def _turbo_video_url(response: Kling3TurboQueryResponse) -> str:
+ """Extract the result video URL from a /tasks response (data[].outputs[] where type == 'video')."""
+ task = response.data[0] if response.data else None
+ if task and task.outputs:
+ for output in task.outputs:
+ if output.type == "video" and output.url:
+ return output.url
+ raise RuntimeError(f"Kling 3.0 Turbo task finished without a video output: {response.model_dump()}")
+
+
+async def execute_kling_turbo(
+ cls: type[IO.ComfyNode],
+ *,
+ prompt: str,
+ resolution: str,
+ aspect_ratio: str,
+ duration: int,
+ start_frame: torch.Tensor | None,
+) -> IO.NodeOutput:
+ """Create + poll a Kling 3.0 Turbo task. Image-to-video when start_frame is given, else text-to-video."""
+ if start_frame is not None:
+ validate_image_dimensions(start_frame, min_width=300, min_height=300)
+ validate_image_aspect_ratio(start_frame, (1, 2.5), (2.5, 1))
+ contents = [Kling3TurboContent(type="first_frame", url=tensor_to_base64_string(start_frame))]
+ if prompt:
+ contents.insert(0, Kling3TurboContent(type="prompt", text=prompt))
+ create = await sync_op(
+ cls,
+ ApiEndpoint(path="/proxy/kling/image-to-video/kling-3.0-turbo", method="POST"),
+ response_model=Kling3TurboCreateResponse,
+ data=Kling3TurboImage2VideoRequest(
+ contents=contents,
+ settings=Kling3TurboSettings(resolution=resolution, duration=duration), # i2v: no aspect_ratio
+ ),
+ )
+ else:
+ create = await sync_op(
+ cls,
+ ApiEndpoint(path="/proxy/kling/text-to-video/kling-3.0-turbo", method="POST"),
+ response_model=Kling3TurboCreateResponse,
+ data=Kling3TurboText2VideoRequest(
+ prompt=prompt,
+ settings=Kling3TurboSettings(resolution=resolution, aspect_ratio=aspect_ratio, duration=duration),
+ ),
+ )
+ if not (create.data and create.data.id):
+ raise RuntimeError(f"Kling 3.0 Turbo create failed. Code: {create.code}, Message: {create.message}")
+ final_response = await poll_op(
+ cls,
+ ApiEndpoint(path="/proxy/kling/tasks", query_params={"task_ids": create.data.id}),
+ response_model=Kling3TurboQueryResponse,
+ status_extractor=lambda r: (r.data[0].status if r.data else None),
+ )
+ return IO.NodeOutput(await download_url_to_video_output(_turbo_video_url(final_response)))
+
+
class KlingVideoNode(IO.ComfyNode):
@classmethod
@@ -2884,7 +2951,11 @@ class KlingVideoNode(IO.ComfyNode):
],
tooltip="Generate a series of video segments with individual prompts and durations.",
),
- IO.Boolean.Input("generate_audio", default=True),
+ IO.Boolean.Input(
+ "generate_audio",
+ default=True,
+ tooltip="'kling-3.0-turbo' always generates native audio, so the audio toggle is ignored.",
+ ),
IO.DynamicCombo.Input(
"model",
options=[
@@ -2899,6 +2970,17 @@ class KlingVideoNode(IO.ComfyNode):
),
],
),
+ IO.DynamicCombo.Option(
+ "kling-3.0-turbo",
+ [
+ IO.Combo.Input("resolution", options=["1080p", "720p"], default="720p"),
+ IO.Combo.Input(
+ "aspect_ratio",
+ options=["16:9", "9:16", "1:1"],
+ tooltip="Ignored in image-to-video mode.",
+ ),
+ ],
+ ),
],
tooltip="Model and generation settings.",
),
@@ -2930,6 +3012,7 @@ class KlingVideoNode(IO.ComfyNode):
price_badge=IO.PriceBadge(
depends_on=IO.PriceBadgeDepends(
widgets=[
+ "model",
"model.resolution",
"generate_audio",
"multi_shot",
@@ -2944,14 +3027,7 @@ class KlingVideoNode(IO.ComfyNode):
),
expr="""
(
- $rates := {
- "4k": {"off": 0.42, "on": 0.42},
- "1080p": {"off": 0.112, "on": 0.168},
- "720p": {"off": 0.084, "on": 0.126}
- };
$res := $lookup(widgets, "model.resolution");
- $audio := widgets.generate_audio ? "on" : "off";
- $rate := $lookup($lookup($rates, $res), $audio);
$ms := widgets.multi_shot;
$isSb := $ms != "disabled";
$n := $isSb ? $number($substring($ms, 0, 1)) : 0;
@@ -2962,7 +3038,18 @@ class KlingVideoNode(IO.ComfyNode):
$d5 := $n >= 5 ? $lookup(widgets, "multi_shot.storyboard_5_duration") : 0;
$d6 := $n >= 6 ? $lookup(widgets, "multi_shot.storyboard_6_duration") : 0;
$dur := $isSb ? $d1 + $d2 + $d3 + $d4 + $d5 + $d6 : $lookup(widgets, "multi_shot.duration");
- {"type":"usd","usd": $rate * $dur}
+ widgets.model = "kling-3.0-turbo"
+ ? {"type":"usd","usd": ($res = "1080p" ? 0.14 : 0.112) * $dur}
+ : (
+ $rates := {
+ "4k": {"off": 0.42, "on": 0.42},
+ "1080p": {"off": 0.112, "on": 0.168},
+ "720p": {"off": 0.084, "on": 0.126}
+ };
+ $audio := widgets.generate_audio ? "on" : "off";
+ $rate := $lookup($lookup($rates, $res), $audio);
+ {"type":"usd","usd": $rate * $dur}
+ )
)
""",
),
@@ -3015,6 +3102,17 @@ class KlingVideoNode(IO.ComfyNode):
duration = multi_shot["duration"]
validate_string(multi_shot["prompt"], min_length=1, max_length=2500)
+ if model["model"] == "kling-3.0-turbo":
+ turbo_prompt = build_turbo_shot_prompt(multi_prompt_list) if custom_multi_shot else multi_shot["prompt"]
+ return await execute_kling_turbo(
+ cls,
+ prompt=turbo_prompt,
+ resolution=model["resolution"],
+ aspect_ratio=model["aspect_ratio"],
+ duration=duration,
+ start_frame=start_frame,
+ )
+
if start_frame is not None:
validate_image_dimensions(start_frame, min_width=300, min_height=300)
validate_image_aspect_ratio(start_frame, (1, 2.5), (2.5, 1))