Merge branch 'master' into feat/api-nodes/Tripo3D-P1

.github/workflows/detect-unreviewed-merge.yml

@@ -0,0 +1,24 @@
+name: Detect Unreviewed Merge
+
+# SOC 2 compliance — reusable workflow lives in Comfy-Org/github-workflows,
+# tracking issues are filed in Comfy-Org/unreviewed-merges.
+
+on:
+  push:
+    branches: [master]
+
+concurrency:
+  group: detect-unreviewed-merge-${{ github.sha }}
+  cancel-in-progress: false
+
+permissions:
+  contents: read
+  pull-requests: read
+
+jobs:
+  detect:
+    uses: Comfy-Org/github-workflows/.github/workflows/detect-unreviewed-merge.yml@4d9cb6b87f953bb7cd69954280e1465fb9bd2040 # v1
+    with:
+      approval-mode: latest-per-reviewer
+    secrets:
+      UNREVIEWED_MERGES_TOKEN: ${{ secrets.UNREVIEWED_MERGES_TOKEN }}

comfy/bg_removal_model.py

@@ -55,12 +55,7 @@ class BackgroundRemovalModel():
         out = torch.nn.functional.interpolate(out, size=(H, W), mode="bicubic", antialias=False)
 
         mask = out.sigmoid().to(device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype())
-        if mask.ndim == 3:
+        return mask.squeeze(1)  # (B, 1, H, W) -> (B, H, W)
-            mask = mask.unsqueeze(0)
-        if mask.shape[1] != 1:
-            mask = mask.movedim(-1, 1)
-
-        return mask
 
 
 def load_background_removal_model(sd):

comfy/float.py

@@ -1,5 +1,20 @@
+import logging
+
 import torch
 
+_CK_STOCHASTIC_ROUNDING_AVAILABLE = False
+try:
+    import comfy_kitchen as ck
+    _ck_stochastic_rounding_fp8 = ck.stochastic_rounding_fp8
+    _CK_STOCHASTIC_ROUNDING_AVAILABLE = True
+except (AttributeError, ImportError):
+    logging.warning("comfy_kitchen does not support stochastic FP8 rounding, please update comfy_kitchen.")
+
+if not _CK_STOCHASTIC_ROUNDING_AVAILABLE:
+    def _ck_stochastic_rounding_fp8(value, rng, dtype):
+        raise NotImplementedError("comfy_kitchen does not support stochastic FP8 rounding")
+
+
 def calc_mantissa(abs_x, exponent, normal_mask, MANTISSA_BITS, EXPONENT_BIAS, generator=None):
     mantissa_scaled = torch.where(
         normal_mask,
@@ -57,6 +72,10 @@ def stochastic_rounding(value, dtype, seed=0):
     if dtype == torch.float8_e4m3fn or dtype == torch.float8_e5m2:
         generator = torch.Generator(device=value.device)
         generator.manual_seed(seed)
+        if _CK_STOCHASTIC_ROUNDING_AVAILABLE:
+            rng = torch.randint(0, 256, value.size(), dtype=torch.uint8, layout=value.layout, device=value.device, generator=generator)
+            return _ck_stochastic_rounding_fp8(value, rng, dtype)
+
         output = torch.empty_like(value, dtype=dtype)
         num_slices = max(1, (value.numel() / (4096 * 4096)))
         slice_size = max(1, round(value.shape[0] / num_slices))

comfy_api/latest/_io.py

@@ -762,21 +762,32 @@ class Accumulation(ComfyTypeIO):
 @comfytype(io_type="LOAD3D_CAMERA")
 class Load3DCamera(ComfyTypeIO):
     class CameraInfo(TypedDict):
-        position: dict[str, float | int]
+        # Coordinate system: right-handed, Y-up, camera looks down -Z
-        target: dict[str, float | int]
+        position: dict[str, float | int]  # scene units
-        zoom: int
+        target: dict[str, float | int]  # scene units; OrbitControls focus point
-        cameraType: str
+        zoom: float | int  # dimensionless, 1 = 100%
-        quaternion: NotRequired[dict[str, float | int]]
+        cameraType: str  # 'perspective' | 'orthographic'
-        rotation: NotRequired[dict[str, float | int | str]]
+        quaternion: NotRequired[dict[str, float | int]]  # normalized, dimensionless; camera world rotation
-        fov: NotRequired[float | int]
+        fov: NotRequired[float | int]  # degrees, vertical FOV (perspective only)
-        aspect: NotRequired[float | int]
+        aspect: NotRequired[float | int]  # width / height (perspective only)
-        near: NotRequired[float | int]
+        near: NotRequired[float | int]  # scene units
-        far: NotRequired[float | int]
+        far: NotRequired[float | int]  # scene units
-        frustum: NotRequired[dict[str, float | int]]
+        frustum: NotRequired[dict[str, float | int]]  # orthographic only: {left, right, top, bottom} in scene units
 
     Type = CameraInfo
 
 
+@comfytype(io_type="LOAD3D_MODEL_INFO")
+class Load3DModelInfo(ComfyTypeIO):
+    class Model3DTransform(TypedDict):
+        # Coordinate system: right-handed, Y-up, world space
+        position: dict[str, float | int]  # scene units
+        quaternion: dict[str, float | int]  # normalized, dimensionless; world rotation
+        scale: dict[str, float | int]  # dimensionless multiplier
+
+    Type = list[Model3DTransform]
+
+
 @comfytype(io_type="LOAD_3D")
 class Load3D(ComfyTypeIO):
     """3D models are stored as a dictionary."""
@@ -786,6 +797,7 @@ class Load3D(ComfyTypeIO):
         normal: str
         camera_info: Load3DCamera.CameraInfo
         recording: NotRequired[str]
+        model_3d_info: NotRequired[list[Load3DModelInfo.Model3DTransform]]
 
     Type = Model3DDict
 
@@ -2298,6 +2310,7 @@ __all__ = [
     "FlowControl",
     "Accumulation",
     "Load3DCamera",
+    "Load3DModelInfo",
     "Load3D",
     "Load3DAnimation",
     "Photomaker",

comfy_api_nodes/nodes_beeble.py

@@ -206,7 +206,7 @@ class BeebleSwitchXVideoEdit(IO.ComfyNode):
         return IO.Schema(
             node_id="BeebleSwitchXVideoEdit",
             display_name="Beeble SwitchX Video Edit",
-            category="api node/video/Beeble",
+            category="video/partner/Beeble",
             description=(
                 "Edit a video with Beeble SwitchX. Switches anything in the scene (background, "
                 "lighting, costume) while preserving the original subject's pixels and motion. "
@@ -302,7 +302,7 @@ class BeebleSwitchXImageEdit(IO.ComfyNode):
         return IO.Schema(
             node_id="BeebleSwitchXImageEdit",
             display_name="Beeble SwitchX Image Edit",
-            category="api node/image/Beeble",
+            category="image/partner/Beeble",
             description=(
                 "Edit a single image with Beeble SwitchX. Switches anything in the scene "
                 "(background, lighting, costume) while preserving the original subject's pixels. "

comfy_api_nodes/nodes_grok.py

@@ -58,7 +58,6 @@ class GrokImageNode(IO.ComfyNode):
                         "grok-imagine-image-quality",
                         "grok-imagine-image-pro",
                         "grok-imagine-image",
-                        "grok-imagine-image-beta",
                     ],
                 ),
                 IO.String.Input(
@@ -233,7 +232,6 @@ class GrokImageEditNode(IO.ComfyNode):
                         "grok-imagine-image-quality",
                         "grok-imagine-image-pro",
                         "grok-imagine-image",
-                        "grok-imagine-image-beta",
                     ],
                 ),
                 IO.Image.Input("image", display_name="images"),
@@ -506,7 +504,7 @@ class GrokVideoNode(IO.ComfyNode):
             category="video/partner/Grok",
             description="Generate video from a prompt or an image",
             inputs=[
-                IO.Combo.Input("model", options=["grok-imagine-video", "grok-imagine-video-beta"]),
+                IO.Combo.Input("model", options=["grok-imagine-video"]),
                 IO.String.Input(
                     "prompt",
                     multiline=True,
@@ -576,8 +574,6 @@ class GrokVideoNode(IO.ComfyNode):
         seed: int,
         image: Input.Image | None = None,
     ) -> IO.NodeOutput:
-        if model == "grok-imagine-video-beta":
-            model = "grok-imagine-video"
         image_url = None
         if image is not None:
             if get_number_of_images(image) != 1:
@@ -618,7 +614,7 @@ class GrokVideoEditNode(IO.ComfyNode):
             category="video/partner/Grok",
             description="Edit an existing video based on a text prompt.",
             inputs=[
-                IO.Combo.Input("model", options=["grok-imagine-video", "grok-imagine-video-beta"]),
+                IO.Combo.Input("model", options=["grok-imagine-video"]),
                 IO.String.Input(
                     "prompt",
                     multiline=True,

comfy_extras/nodes_dataset.py

@@ -157,7 +157,7 @@ class LoadImageTextDataSetFromFolderNode(io.ComfyNode):
         return io.NodeOutput(output_tensor, captions)
 
 
-def save_images_to_folder(image_list, output_dir, prefix="image"):
+def save_images_to_folder(image_list, output_dir, prefix="image", overwrite=True):
     """Utility function to save a list of image tensors to disk.
 
     Args:
@@ -197,7 +197,11 @@ def save_images_to_folder(image_list, output_dir, prefix="image"):
             raise ValueError(f"Expected torch.Tensor, got {type(img_tensor)}")
 
         # Save image
-        filename = f"{prefix}_{idx:05d}.png"
+        if overwrite:
+            filename = f"{prefix}_{idx:05d}.png"
+        else:
+            _, _, counter, _, resolved_prefix = folder_paths.get_save_image_path(prefix, output_dir)
+            filename = f"{resolved_prefix}_{counter:05}_{idx:05d}.png"
         filepath = os.path.join(output_dir, filename)
         img.save(filepath)
         saved_files.append(filename)
@@ -230,19 +234,26 @@ class SaveImageDataSetToFolderNode(io.ComfyNode):
                     tooltip="Prefix for saved image filenames.",
                     advanced=True,
                 ),
+                io.Combo.Input(
+                    "mode",
+                    default="overwrite",
+                    options=["overwrite", "increment"],
+                    tooltip="Whether to overwrite existing files or increment filenames to avoid overwriting."
+                ),
             ],
             outputs=[],
             is_deprecated=True,  # This node is redundant and superseded by existing Save Image nodes where the target folder can be specified in the filename_prefix
         )
 
     @classmethod
-    def execute(cls, images, folder_name, filename_prefix):
+    def execute(cls, images, folder_name, filename_prefix, mode):
         # Extract scalar values
         folder_name = folder_name[0]
         filename_prefix = filename_prefix[0]
+        mode = mode[0]
 
         output_dir = os.path.join(folder_paths.get_output_directory(), folder_name)
-        saved_files = save_images_to_folder(images, output_dir, filename_prefix)
+        saved_files = save_images_to_folder(images, output_dir, filename_prefix, mode=='overwrite')
 
         logging.info(f"Saved {len(saved_files)} images to {output_dir}.")
         return io.NodeOutput()
@@ -278,18 +289,25 @@ class SaveImageTextDataSetToFolderNode(io.ComfyNode):
                     tooltip="Prefix for saved image filenames.",
                     advanced=True,
                 ),
+                io.Combo.Input(
+                    "mode",
+                    default="overwrite",
+                    options=["overwrite", "increment"],
+                    tooltip="Whether to overwrite existing files or increment filenames to avoid overwriting."
+                ),
             ],
             outputs=[],
         )
 
     @classmethod
-    def execute(cls, images, folder_name, filename_prefix, texts=None):
+    def execute(cls, images, folder_name, filename_prefix, mode, texts=None):
         # Extract scalar values
         folder_name = folder_name[0]
         filename_prefix = filename_prefix[0]
+        mode = mode[0]
 
         output_dir = os.path.join(folder_paths.get_output_directory(), folder_name)
-        saved_files = save_images_to_folder(images, output_dir, filename_prefix)
+        saved_files = save_images_to_folder(images, output_dir, filename_prefix, mode=='overwrite')
 
         # Save captions
         if texts:

comfy_extras/nodes_load_3d.py

@@ -47,6 +47,7 @@ class Load3D(IO.ComfyNode):
                 IO.Load3DCamera.Output(display_name="camera_info"),
                 IO.Video.Output(display_name="recording_video"),
                 IO.File3DAny.Output(display_name="model_3d"),
+                IO.Load3DModelInfo.Output(display_name="model_3d_info"),
             ],
         )
 
@@ -73,7 +74,8 @@ class Load3D(IO.ComfyNode):
         if model_file and model_file != "none":
             file_3d = Types.File3D(folder_paths.get_annotated_filepath(model_file))
             mesh_path = model_file
-        return IO.NodeOutput(output_image, output_mask, mesh_path, normal_image, image['camera_info'], video, file_3d)
+        model_3d_info = image.get('model_3d_info', [])
+        return IO.NodeOutput(output_image, output_mask, mesh_path, normal_image, image['camera_info'], video, file_3d, model_3d_info)
 
     process = execute  # TODO: remove
 

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.44.19
-comfyui-workflow-templates==0.9.85
+comfyui-workflow-templates==0.9.91
 comfyui-embedded-docs==0.5.1
 torch
 torchsde
@@ -21,8 +21,8 @@ psutil
 alembic
 SQLAlchemy>=2.0.0
 filelock
-av>=14.2.0
+av>=16.0.0
-comfy-kitchen>=0.2.8
+comfy-kitchen==0.2.9
 comfy-aimdo==0.4.5
 requests
 simpleeval>=1.0.0