From 52e778fff3c1d6f32c8d14cba9864faddba8475d Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Fri, 28 Nov 2025 12:52:59 +0200 Subject: [PATCH 1/5] feat(Kling-API-Nodes): add v2-5-turbo model to FirstLastFrame node (#10938) --- comfy_api_nodes/nodes_kling.py | 60 +++++++++++++++------------------- 1 file changed, 26 insertions(+), 34 deletions(-) diff --git a/comfy_api_nodes/nodes_kling.py b/comfy_api_nodes/nodes_kling.py index 36852038b..23a7f55f1 100644 --- a/comfy_api_nodes/nodes_kling.py +++ b/comfy_api_nodes/nodes_kling.py @@ -4,8 +4,6 @@ For source of truth on the allowed permutations of request fields, please refere - [Compatibility Table](https://app.klingai.com/global/dev/document-api/apiReference/model/skillsMap) """ -from __future__ import annotations -from typing import Optional, TypeVar import math import logging @@ -66,9 +64,7 @@ from comfy_api_nodes.util import ( poll_op, ) from comfy_api.input_impl import VideoFromFile -from comfy_api.input.basic_types import AudioInput -from comfy_api.input.video_types import VideoInput -from comfy_api.latest import ComfyExtension, IO +from comfy_api.latest import ComfyExtension, IO, Input KLING_API_VERSION = "v1" PATH_TEXT_TO_VIDEO = f"/proxy/kling/{KLING_API_VERSION}/videos/text2video" @@ -94,8 +90,6 @@ AVERAGE_DURATION_IMAGE_GEN = 32 AVERAGE_DURATION_VIDEO_EFFECTS = 320 AVERAGE_DURATION_VIDEO_EXTEND = 320 -R = TypeVar("R") - MODE_TEXT2VIDEO = { "standard mode / 5s duration / kling-v1": ("std", "5", "kling-v1"), @@ -130,6 +124,8 @@ MODE_START_END_FRAME = { "pro mode / 10s duration / kling-v1-6": ("pro", "10", "kling-v1-6"), "pro mode / 5s duration / kling-v2-1": ("pro", "5", "kling-v2-1"), "pro mode / 10s duration / kling-v2-1": ("pro", "10", "kling-v2-1"), + "pro mode / 5s duration / kling-v2-5-turbo": ("pro", "5", "kling-v2-5-turbo"), + "pro mode / 10s duration / kling-v2-5-turbo": ("pro", "10", "kling-v2-5-turbo"), } """ Returns a mapping of mode strings to their corresponding (mode, duration, model_name) tuples. @@ -296,7 +292,7 @@ def get_video_from_response(response) -> KlingVideoResult: return video -def get_video_url_from_response(response) -> Optional[str]: +def get_video_url_from_response(response) -> str | None: """Returns the first video url from the Kling video generation task result. Will not raise an error if the response is not valid. """ @@ -315,7 +311,7 @@ def get_images_from_response(response) -> list[KlingImageResult]: return images -def get_images_urls_from_response(response) -> Optional[str]: +def get_images_urls_from_response(response) -> str | None: """Returns the list of image urls from the Kling image generation task result. Will not raise an error if the response is not valid. If there is only one image, returns the url as a string. If there are multiple images, returns a list of urls. """ @@ -349,7 +345,7 @@ async def execute_text2video( model_mode: str, duration: str, aspect_ratio: str, - camera_control: Optional[KlingCameraControl] = None, + camera_control: KlingCameraControl | None = None, ) -> IO.NodeOutput: validate_prompts(prompt, negative_prompt, MAX_PROMPT_LENGTH_T2V) task_creation_response = await sync_op( @@ -394,8 +390,8 @@ async def execute_image2video( model_mode: str, aspect_ratio: str, duration: str, - camera_control: Optional[KlingCameraControl] = None, - end_frame: Optional[torch.Tensor] = None, + camera_control: KlingCameraControl | None = None, + end_frame: torch.Tensor | None = None, ) -> IO.NodeOutput: validate_prompts(prompt, negative_prompt, MAX_PROMPT_LENGTH_I2V) validate_input_image(start_frame) @@ -451,8 +447,8 @@ async def execute_video_effect( model_name: str, duration: KlingVideoGenDuration, image_1: torch.Tensor, - image_2: Optional[torch.Tensor] = None, - model_mode: Optional[KlingVideoGenMode] = None, + image_2: torch.Tensor | None = None, + model_mode: KlingVideoGenMode | None = None, ) -> tuple[VideoFromFile, str, str]: if dual_character: request_input_field = KlingDualCharacterEffectInput( @@ -499,13 +495,13 @@ async def execute_video_effect( async def execute_lipsync( cls: type[IO.ComfyNode], - video: VideoInput, - audio: Optional[AudioInput] = None, - voice_language: Optional[str] = None, - model_mode: Optional[str] = None, - text: Optional[str] = None, - voice_speed: Optional[float] = None, - voice_id: Optional[str] = None, + video: Input.Video, + audio: Input.Audio | None = None, + voice_language: str | None = None, + model_mode: str | None = None, + text: str | None = None, + voice_speed: float | None = None, + voice_id: str | None = None, ) -> IO.NodeOutput: if text: validate_string(text, field_name="Text", max_length=MAX_PROMPT_LENGTH_LIP_SYNC) @@ -787,7 +783,7 @@ class KlingCameraControlT2VNode(IO.ComfyNode): negative_prompt: str, cfg_scale: float, aspect_ratio: str, - camera_control: Optional[KlingCameraControl] = None, + camera_control: KlingCameraControl | None = None, ) -> IO.NodeOutput: return await execute_text2video( cls, @@ -854,8 +850,8 @@ class KlingImage2VideoNode(IO.ComfyNode): mode: str, aspect_ratio: str, duration: str, - camera_control: Optional[KlingCameraControl] = None, - end_frame: Optional[torch.Tensor] = None, + camera_control: KlingCameraControl | None = None, + end_frame: torch.Tensor | None = None, ) -> IO.NodeOutput: return await execute_image2video( cls, @@ -965,15 +961,11 @@ class KlingStartEndFrameNode(IO.ComfyNode): IO.String.Input("prompt", multiline=True, tooltip="Positive text prompt"), IO.String.Input("negative_prompt", multiline=True, tooltip="Negative text prompt"), IO.Float.Input("cfg_scale", default=0.5, min=0.0, max=1.0), - IO.Combo.Input( - "aspect_ratio", - options=[i.value for i in KlingVideoGenAspectRatio], - default="16:9", - ), + IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]), IO.Combo.Input( "mode", options=modes, - default=modes[2], + default=modes[8], tooltip="The configuration to use for the video generation following the format: mode / duration / model_name.", ), ], @@ -1254,8 +1246,8 @@ class KlingLipSyncAudioToVideoNode(IO.ComfyNode): @classmethod async def execute( cls, - video: VideoInput, - audio: AudioInput, + video: Input.Video, + audio: Input.Audio, voice_language: str, ) -> IO.NodeOutput: return await execute_lipsync( @@ -1314,7 +1306,7 @@ class KlingLipSyncTextToVideoNode(IO.ComfyNode): @classmethod async def execute( cls, - video: VideoInput, + video: Input.Video, text: str, voice: str, voice_speed: float, @@ -1471,7 +1463,7 @@ class KlingImageGenerationNode(IO.ComfyNode): human_fidelity: float, n: int, aspect_ratio: KlingImageGenAspectRatio, - image: Optional[torch.Tensor] = None, + image: torch.Tensor | None = None, ) -> IO.NodeOutput: validate_string(prompt, field_name="prompt", min_length=1, max_length=MAX_PROMPT_LENGTH_IMAGE_GEN) validate_string(negative_prompt, field_name="negative_prompt", max_length=MAX_PROMPT_LENGTH_IMAGE_GEN) From ca7808f240d4d53e594d3b95753240313864c992 Mon Sep 17 00:00:00 2001 From: "Dr.Lt.Data" <128333288+ltdrdata@users.noreply.github.com> Date: Sat, 29 Nov 2025 05:43:17 +0900 Subject: [PATCH 2/5] fix(user_manager): fix typo in move_userdata dest validation (#10967) Check `dest` instead of `source` when validating destination path in move_userdata endpoint. --- app/user_manager.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/app/user_manager.py b/app/user_manager.py index a2d376c0c..675f6c0c6 100644 --- a/app/user_manager.py +++ b/app/user_manager.py @@ -424,7 +424,7 @@ class UserManager(): return source dest = get_user_data_path(request, check_exists=False, param="dest") - if not isinstance(source, str): + if not isinstance(dest, str): return dest overwrite = request.query.get("overwrite", 'true') != "false" From f55c98a89f76fc06c435a728bc2e76b6b4051463 Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Fri, 28 Nov 2025 13:16:46 -0800 Subject: [PATCH 3/5] Disable offload stream when torch compile. (#10961) --- comfy/model_management.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/comfy/model_management.py b/comfy/model_management.py index 38c506df5..d8ce80010 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -1042,6 +1042,9 @@ def get_offload_stream(device): if NUM_STREAMS == 0: return None + if torch.compiler.is_compiling(): + return None + if device in STREAMS: ss = STREAMS[device] #Sync the oldest stream in the queue with the current From 6484ac89dc683b178d9ef3f7406800f7132147ba Mon Sep 17 00:00:00 2001 From: Urle Sistiana <55231606+urlesistiana@users.noreply.github.com> Date: Sat, 29 Nov 2025 05:33:07 +0800 Subject: [PATCH 4/5] fix QuantizedTensor.is_contiguous (#10956) (#10959) --- comfy/quant_ops.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/comfy/quant_ops.py b/comfy/quant_ops.py index 9b924560b..bb1fb860c 100644 --- a/comfy/quant_ops.py +++ b/comfy/quant_ops.py @@ -235,8 +235,8 @@ class QuantizedTensor(torch.Tensor): def is_pinned(self): return self._qdata.is_pinned() - def is_contiguous(self): - return self._qdata.is_contiguous() + def is_contiguous(self, *arg, **kwargs): + return self._qdata.is_contiguous(*arg, **kwargs) # ============================================================================== # Generic Utilities (Layout-Agnostic Operations) From 0ff0457892467ef8a6ea235dcd0618c10ca44ee3 Mon Sep 17 00:00:00 2001 From: rattus <46076784+rattus128@users.noreply.github.com> Date: Sat, 29 Nov 2025 07:38:12 +1000 Subject: [PATCH 5/5] mm: wrap the raw stream in context manager (#10958) The documentation of torch.foo.Stream being usable with with: suggests it starts at version 2.7. Use the old API for backwards compatibility. --- comfy/model_management.py | 19 +++++++++++++++---- comfy/ops.py | 2 ++ 2 files changed, 17 insertions(+), 4 deletions(-) diff --git a/comfy/model_management.py b/comfy/model_management.py index d8ce80010..aeddbaefe 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -1055,7 +1055,9 @@ def get_offload_stream(device): elif is_device_cuda(device): ss = [] for k in range(NUM_STREAMS): - ss.append(torch.cuda.Stream(device=device, priority=0)) + s1 = torch.cuda.Stream(device=device, priority=0) + s1.as_context = torch.cuda.stream + ss.append(s1) STREAMS[device] = ss s = ss[stream_counter] stream_counters[device] = stream_counter @@ -1063,7 +1065,9 @@ def get_offload_stream(device): elif is_device_xpu(device): ss = [] for k in range(NUM_STREAMS): - ss.append(torch.xpu.Stream(device=device, priority=0)) + s1 = torch.xpu.Stream(device=device, priority=0) + s1.as_context = torch.xpu.stream + ss.append(s1) STREAMS[device] = ss s = ss[stream_counter] stream_counters[device] = stream_counter @@ -1081,12 +1085,19 @@ def cast_to(weight, dtype=None, device=None, non_blocking=False, copy=False, str if dtype is None or weight.dtype == dtype: return weight if stream is not None: - with stream: + wf_context = stream + if hasattr(wf_context, "as_context"): + wf_context = wf_context.as_context(stream) + with wf_context: return weight.to(dtype=dtype, copy=copy) return weight.to(dtype=dtype, copy=copy) + if stream is not None: - with stream: + wf_context = stream + if hasattr(wf_context, "as_context"): + wf_context = wf_context.as_context(stream) + with wf_context: r = torch.empty_like(weight, dtype=dtype, device=device) r.copy_(weight, non_blocking=non_blocking) else: diff --git a/comfy/ops.py b/comfy/ops.py index a0ff4e8f1..61a2f0754 100644 --- a/comfy/ops.py +++ b/comfy/ops.py @@ -95,6 +95,8 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of if offload_stream is not None: wf_context = offload_stream + if hasattr(wf_context, "as_context"): + wf_context = wf_context.as_context(offload_stream) else: wf_context = contextlib.nullcontext()