Merge branch 'master' into feat/load3d-model-info-output

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

@@ -777,6 +777,17 @@ class Load3DCamera(ComfyTypeIO):
     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_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
@@ -22,7 +22,7 @@ alembic
 SQLAlchemy>=2.0.0
 filelock
 av>=16.0.0
-comfy-kitchen>=0.2.8
+comfy-kitchen==0.2.9
 comfy-aimdo==0.4.5
 requests
 simpleeval>=1.0.0