Merge upstream/master, keep local README.md

comfy_api_nodes/nodes_kling.py

@@ -858,7 +858,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode):
                     tooltip="A text prompt describing the video content. "
                     "This can include both positive and negative descriptions.",
                 ),
-                IO.Combo.Input("duration", options=["5", "10"]),
+                IO.Int.Input("duration", default=5, min=3, max=10, display_mode=IO.NumberDisplay.slider),
                 IO.Image.Input("first_frame"),
                 IO.Image.Input(
                     "end_frame",
@@ -897,6 +897,10 @@ class OmniProFirstLastFrameNode(IO.ComfyNode):
         validate_string(prompt, min_length=1, max_length=2500)
         if end_frame is not None and reference_images is not None:
             raise ValueError("The 'end_frame' input cannot be used simultaneously with 'reference_images'.")
+        if duration not in (5, 10) and end_frame is None and reference_images is None:
+            raise ValueError(
+                "Duration is only supported for 5 or 10 seconds if there is no end frame or reference images."
+            )
         validate_image_dimensions(first_frame, min_width=300, min_height=300)
         validate_image_aspect_ratio(first_frame, (1, 2.5), (2.5, 1))
         image_list: list[OmniParamImage] = [

comfy_extras/nodes_custom_sampler.py

@@ -760,8 +760,12 @@ class SamplerCustom(io.ComfyNode):
         out = latent.copy()
         out["samples"] = samples
         if "x0" in x0_output:
+            x0_out = model.model.process_latent_out(x0_output["x0"].cpu())
+            if samples.is_nested:
+                latent_shapes = [x.shape for x in samples.unbind()]
+                x0_out = comfy.nested_tensor.NestedTensor(comfy.utils.unpack_latents(x0_out, latent_shapes))
             out_denoised = latent.copy()
-            out_denoised["samples"] = model.model.process_latent_out(x0_output["x0"].cpu())
+            out_denoised["samples"] = x0_out
         else:
             out_denoised = out
         return io.NodeOutput(out, out_denoised)
@@ -948,8 +952,12 @@ class SamplerCustomAdvanced(io.ComfyNode):
         out = latent.copy()
         out["samples"] = samples
         if "x0" in x0_output:
+            x0_out = guider.model_patcher.model.process_latent_out(x0_output["x0"].cpu())
+            if samples.is_nested:
+                latent_shapes = [x.shape for x in samples.unbind()]
+                x0_out = comfy.nested_tensor.NestedTensor(comfy.utils.unpack_latents(x0_out, latent_shapes))
             out_denoised = latent.copy()
-            out_denoised["samples"] = guider.model_patcher.model.process_latent_out(x0_output["x0"].cpu())
+            out_denoised["samples"] = x0_out
         else:
             out_denoised = out
         return io.NodeOutput(out, out_denoised)

comfy_extras/nodes_qwen.py

@@ -3,7 +3,9 @@ import comfy.utils
 import math
 from typing_extensions import override
 from comfy_api.latest import ComfyExtension, io
-
+import comfy.model_management
+import torch
+import nodes
 
 class TextEncodeQwenImageEdit(io.ComfyNode):
     @classmethod
@@ -104,12 +106,37 @@ class TextEncodeQwenImageEditPlus(io.ComfyNode):
         return io.NodeOutput(conditioning)
 
 
+class EmptyQwenImageLayeredLatentImage(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="EmptyQwenImageLayeredLatentImage",
+            display_name="Empty Qwen Image Layered Latent",
+            category="latent/qwen",
+            inputs=[
+                io.Int.Input("width", default=640, min=16, max=nodes.MAX_RESOLUTION, step=16),
+                io.Int.Input("height", default=640, min=16, max=nodes.MAX_RESOLUTION, step=16),
+                io.Int.Input("layers", default=3, min=0, max=nodes.MAX_RESOLUTION, step=1),
+                io.Int.Input("batch_size", default=1, min=1, max=4096),
+            ],
+            outputs=[
+                io.Latent.Output(),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, width, height, layers, batch_size=1) -> io.NodeOutput:
+        latent = torch.zeros([batch_size, 16, layers + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
+        return io.NodeOutput({"samples": latent})
+
+
 class QwenExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[io.ComfyNode]]:
         return [
             TextEncodeQwenImageEdit,
             TextEncodeQwenImageEditPlus,
+            EmptyQwenImageLayeredLatentImage,
         ]
 
 

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.34.9
-comfyui-workflow-templates==0.7.60
+comfyui-workflow-templates==0.7.62
 comfyui-embedded-docs==0.3.1
 torch
 torchsde