[pull] master from Comfy-Org:master

comfy/model_patcher.py

@@ -19,7 +19,6 @@
 from __future__ import annotations
 
 import collections
-import copy
 import inspect
 import logging
 import math
@@ -317,7 +316,7 @@ def clone(self):
 
         n.object_patches = self.object_patches.copy()
         n.weight_wrapper_patches = self.weight_wrapper_patches.copy()
-        n.model_options = copy.deepcopy(self.model_options)
+        n.model_options = comfy.utils.deepcopy_list_dict(self.model_options)
         n.backup = self.backup
         n.object_patches_backup = self.object_patches_backup
         n.parent = self

comfy/ops.py

@@ -169,8 +169,8 @@ def to_dequant(tensor, dtype):
                 if orig.dtype == dtype and len(fns) == 0:
                     #The layer actually wants our freshly saved QT
                     x = y
-            else:
-                y = x
+            elif update_weight:
+                y = comfy.float.stochastic_rounding(x, orig.dtype, seed = comfy.utils.string_to_seed(s.seed_key))
             if update_weight:
                 orig.copy_(y)
         for f in fns:

comfy/sd.py

@@ -793,8 +793,6 @@ def estimate_memory(shape, dtype, num_layers = 16, kv_cache_multiplier = 2):
             self.first_stage_model = AutoencoderKL(**(config['params']))
         self.first_stage_model = self.first_stage_model.eval()
 
-        model_management.archive_model_dtypes(self.first_stage_model)
-
         if device is None:
             device = model_management.vae_device()
         self.device = device
@@ -803,6 +801,7 @@ def estimate_memory(shape, dtype, num_layers = 16, kv_cache_multiplier = 2):
             dtype = model_management.vae_dtype(self.device, self.working_dtypes)
         self.vae_dtype = dtype
         self.first_stage_model.to(self.vae_dtype)
+        model_management.archive_model_dtypes(self.first_stage_model)
         self.output_device = model_management.intermediate_device()
 
         mp = comfy.model_patcher.CoreModelPatcher

comfy/utils.py

@@ -1376,3 +1376,21 @@ def string_to_seed(data):
             else:
                 crc >>= 1
     return crc ^ 0xFFFFFFFF
+
+def deepcopy_list_dict(obj, memo=None):
+    if memo is None:
+        memo = {}
+
+    obj_id = id(obj)
+    if obj_id in memo:
+        return memo[obj_id]
+
+    if isinstance(obj, dict):
+        res = {deepcopy_list_dict(k, memo): deepcopy_list_dict(v, memo) for k, v in obj.items()}
+    elif isinstance(obj, list):
+        res = [deepcopy_list_dict(i, memo) for i in obj]
+    else:
+        res = obj
+
+    memo[obj_id] = res
+    return res

comfy_api_nodes/apis/kling.py

@@ -1,12 +1,22 @@
 from pydantic import BaseModel, Field
 
 
+class MultiPromptEntry(BaseModel):
+    index: int = Field(...)
+    prompt: str = Field(...)
+    duration: str = Field(...)
+
+
 class OmniProText2VideoRequest(BaseModel):
     model_name: str = Field(..., description="kling-video-o1")
     aspect_ratio: str = Field(..., description="'16:9', '9:16' or '1:1'")
     duration: str = Field(..., description="'5' or '10'")
     prompt: str = Field(...)
     mode: str = Field("pro")
+    multi_shot: bool | None = Field(None)
+    multi_prompt: list[MultiPromptEntry] | None = Field(None)
+    shot_type: str | None = Field(None)
+    sound: str = Field(..., description="'on' or 'off'")
 
 
 class OmniParamImage(BaseModel):
@@ -26,6 +36,10 @@ class OmniProFirstLastFrameRequest(BaseModel):
     duration: str = Field(..., description="'5' or '10'")
     prompt: str = Field(...)
     mode: str = Field("pro")
+    sound: str | None = Field(None, description="'on' or 'off'")
+    multi_shot: bool | None = Field(None)
+    multi_prompt: list[MultiPromptEntry] | None = Field(None)
+    shot_type: str | None = Field(None)
 
 
 class OmniProReferences2VideoRequest(BaseModel):
@@ -38,6 +52,10 @@ class OmniProReferences2VideoRequest(BaseModel):
     duration: str | None = Field(..., description="From 3 to 10.")
     prompt: str = Field(...)
     mode: str = Field("pro")
+    sound: str | None = Field(None, description="'on' or 'off'")
+    multi_shot: bool | None = Field(None)
+    multi_prompt: list[MultiPromptEntry] | None = Field(None)
+    shot_type: str | None = Field(None)
 
 
 class TaskStatusVideoResult(BaseModel):
@@ -54,6 +72,7 @@ class TaskStatusImageResult(BaseModel):
 class TaskStatusResults(BaseModel):
     videos: list[TaskStatusVideoResult] | None = Field(None)
     images: list[TaskStatusImageResult] | None = Field(None)
+    series_images: list[TaskStatusImageResult] | None = Field(None)
 
 
 class TaskStatusResponseData(BaseModel):
@@ -77,31 +96,42 @@ class OmniImageParamImage(BaseModel):
 
 
 class OmniProImageRequest(BaseModel):
-    model_name: str = Field(..., description="kling-image-o1")
-    resolution: str = Field(..., description="'1k' or '2k'")
+    model_name: str = Field(...)
+    resolution: str = Field(...)
     aspect_ratio: str | None = Field(...)
     prompt: str = Field(...)
     mode: str = Field("pro")
     n: int | None = Field(1, le=9)
     image_list: list[OmniImageParamImage] | None = Field(..., max_length=10)
+    result_type: str | None = Field(None, description="Set to 'series' for series generation")
+    series_amount: int | None = Field(None, ge=2, le=9, description="Number of images in a series")
 
 
 class TextToVideoWithAudioRequest(BaseModel):
-    model_name: str = Field(..., description="kling-v2-6")
+    model_name: str = Field(...)
     aspect_ratio: str = Field(..., description="'16:9', '9:16' or '1:1'")
-    duration: str = Field(..., description="'5' or '10'")
-    prompt: str = Field(...)
+    duration: str = Field(...)
+    prompt: str | None = Field(...)
+    negative_prompt: str | None = Field(None)
     mode: str = Field("pro")
     sound: str = Field(..., description="'on' or 'off'")
+    multi_shot: bool | None = Field(None)
+    multi_prompt: list[MultiPromptEntry] | None = Field(None)
+    shot_type: str | None = Field(None)
 
 
 class ImageToVideoWithAudioRequest(BaseModel):
-    model_name: str = Field(..., description="kling-v2-6")
+    model_name: str = Field(...)
     image: str = Field(...)
-    duration: str = Field(..., description="'5' or '10'")
-    prompt: str = Field(...)
+    image_tail: str | None = Field(None)
+    duration: str = Field(...)
+    prompt: str | None = Field(...)
+    negative_prompt: str | None = Field(None)
     mode: str = Field("pro")
     sound: str = Field(..., description="'on' or 'off'")
+    multi_shot: bool | None = Field(None)
+    multi_prompt: list[MultiPromptEntry] | None = Field(None)
+    shot_type: str | None = Field(None)
 
 
 class MotionControlRequest(BaseModel):