Merge 96e5287a72 into 15a4494a4e

* Standardize DEPRECATED label in node display name

* Promote category image/video to root level video/

* Update images and masks names and categories

comfy/cli_args.py

@@ -91,6 +91,7 @@ parser.add_argument("--directml", type=int, nargs="?", metavar="DIRECTML_DEVICE"
 
 parser.add_argument("--oneapi-device-selector", type=str, default=None, metavar="SELECTOR_STRING", help="Sets the oneAPI device(s) this instance will use.")
 parser.add_argument("--supports-fp8-compute", action="store_true", help="ComfyUI will act like if the device supports fp8 compute.")
+parser.add_argument("--enable-triton-backend", action="store_true", help="ComfyUI will enable the use of Triton backend in comfy-kitchen. Is disabled at launch by default.")
 
 class LatentPreviewMethod(enum.Enum):
     NoPreviews = "none"

comfy/model_management.py

@@ -721,13 +721,15 @@ def load_models_gpu(models, memory_required=0, force_patch_weights=False, minimu
     else:
         minimum_memory_required = max(inference_memory, minimum_memory_required + extra_reserved_memory())
 
-    models_temp = set()
+    # Order-preserving dedup. A plain set() would randomize iteration order across runs
+    models_temp = {}
     for m in models:
-        models_temp.add(m)
+        models_temp[m] = None
         for mm in m.model_patches_models():
-            models_temp.add(mm)
+            models_temp[mm] = None
 
-    models = models_temp
+    models = list(models_temp)
+    models.reverse()
 
     models_to_load = []
 

comfy/model_prefetch.py

@@ -37,7 +37,8 @@ def prefetch_queue_pop(queue, device, module):
     consumed = queue.pop(0)
     if consumed is not None:
         offload_stream, prefetch_state = consumed
-        offload_stream.wait_stream(comfy.model_management.current_stream(device))
+        if offload_stream is not None:
+            offload_stream.wait_stream(comfy.model_management.current_stream(device))
         _, comfy_modules = prefetch_state
         if comfy_modules is not None:
             cleanup_prefetched_modules(comfy_modules)

comfy/ops.py

@@ -253,6 +253,9 @@ def resolve_cast_module_with_vbar(s, dtype, device, bias_dtype, compute_dtype, w
     if bias is not None:
         bias = post_cast(s, "bias", bias, bias_dtype, prefetch["resident"], update_weight)
 
+    if prefetch["signature"] is not None:
+        prefetch["resident"] = True
+
     return weight, bias
 
 
@@ -1084,6 +1087,51 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                             orig_dtype=MixedPrecisionOps._compute_dtype,
                             orig_shape=(self.out_features, self.in_features),
                         )
+                    elif self.quant_format == "svdquant_w4a4":
+                        # SVDQuant W4A4: per-group weight scales + low-rank correction
+                        # (proj_down, proj_up) + activation smoothing (smooth_factor)
+                        wscales = self._load_scale_param(state_dict, prefix, "weight_scale", device, manually_loaded_keys)
+                        proj_down = self._load_scale_param(state_dict, prefix, "proj_down", device, manually_loaded_keys)
+                        proj_up = self._load_scale_param(state_dict, prefix, "proj_up", device, manually_loaded_keys)
+                        smooth_factor = self._load_scale_param(state_dict, prefix, "smooth_factor", device, manually_loaded_keys)
+                        act_unsigned = bool(layer_conf.get("act_unsigned", False))
+
+                        # Early Qwen-Image conversion artifacts did not persist the
+                        # fused GELU -> fc2 unsigned-activation flag. Those layers
+                        # are the second linear in the feed-forward block.
+                        if not act_unsigned and (
+                            layer_name.endswith(".img_mlp.net.2") or layer_name.endswith(".txt_mlp.net.2")
+                        ):
+                            act_unsigned = True
+
+                        if any(t is None for t in (wscales, proj_down, proj_up, smooth_factor)):
+                            raise ValueError(f"Missing SVDQuant W4A4 parameters for layer {layer_name}")
+
+                        params = layout_cls.Params(
+                            scale=wscales,
+                            orig_dtype=MixedPrecisionOps._compute_dtype,
+                            orig_shape=(self.out_features, self.in_features),
+                            proj_down=proj_down,
+                            proj_up=proj_up,
+                            smooth_factor=smooth_factor,
+                            act_unsigned=act_unsigned,
+                        )
+                    elif self.quant_format == "awq_w4a16":
+                        # AWQ W4A16: int4 weight, fp16/bf16 activation. Used for
+                        # the modulation linears (img_mod.1 / txt_mod.1) so they
+                        # stay int4 in checkpoint + VRAM rather than getting
+                        # dequantized to bf16 at conversion time (~10 GB saving).
+                        wscales = self._load_scale_param(state_dict, prefix, "weight_scale", device, manually_loaded_keys)
+                        wzeros = self._load_scale_param(state_dict, prefix, "weight_zero", device, manually_loaded_keys)
+                        if wscales is None or wzeros is None:
+                            raise ValueError(f"Missing AWQ W4A16 parameters for layer {layer_name}")
+                        params = layout_cls.Params(
+                            scale=wscales,
+                            zeros=wzeros,
+                            group_size=int(layer_conf.get("group_size", qconfig.get("group_size", 64))),
+                            orig_dtype=MixedPrecisionOps._compute_dtype,
+                            orig_shape=(self.out_features, self.in_features),
+                        )
                     else:
                         raise ValueError(f"Unsupported quantization format: {self.quant_format}")
 
@@ -1133,6 +1181,8 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                     quant_conf = {"format": self.quant_format}
                     if self._full_precision_mm_config:
                         quant_conf["full_precision_matrix_mult"] = True
+                    if bool(getattr(getattr(self.weight, "_params", None), "act_unsigned", False)):
+                        quant_conf["act_unsigned"] = True
                     sd["{}comfy_quant".format(prefix)] = torch.tensor(list(json.dumps(quant_conf).encode('utf-8')), dtype=torch.uint8)
 
                     input_scale = getattr(self, 'input_scale', None)
@@ -1190,18 +1240,24 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
 
                 # Inference path (unchanged)
                 if _use_quantized:
+                    # Some layouts (e.g. SVDQuant W4A4) do activation quantization
+                    # inside their fused kernel and cannot pre-quantize a float
+                    # tensor up-front. Skip the input wrapping for those.
+                    layout_cls = get_layout_class(self.layout_type)
+                    layout_quantizes_input = getattr(layout_cls, "QUANTIZES_INPUT", True)
 
-                    # Reshape 3D tensors to 2D for quantization (needed for NVFP4 and others)
-                    input_reshaped = input.reshape(-1, input_shape[2]) if input.ndim == 3 else input
+                    if layout_quantizes_input:
+                        # Reshape 3D tensors to 2D for quantization (needed for NVFP4 and others)
+                        input_reshaped = input.reshape(-1, input_shape[2]) if input.ndim == 3 else input
 
-                    # Fall back to non-quantized for non-2D tensors
-                    if input_reshaped.ndim == 2:
-                        reshaped_3d = input.ndim == 3
-                        # dtype is now implicit in the layout class
-                        scale = getattr(self, 'input_scale', None)
-                        if scale is not None:
-                            scale = comfy.model_management.cast_to_device(scale, input.device, None)
-                        input = QuantizedTensor.from_float(input_reshaped, self.layout_type, scale=scale)
+                        # Fall back to non-quantized for non-2D tensors
+                        if input_reshaped.ndim == 2:
+                            reshaped_3d = input.ndim == 3
+                            # dtype is now implicit in the layout class
+                            scale = getattr(self, 'input_scale', None)
+                            if scale is not None:
+                                scale = comfy.model_management.cast_to_device(scale, input.device, None)
+                            input = QuantizedTensor.from_float(input_reshaped, self.layout_type, scale=scale)
 
                 output = self.forward_comfy_cast_weights(input, compute_dtype, want_requant=isinstance(input, QuantizedTensor))
 

comfy/quant_ops.py

@@ -1,6 +1,8 @@
 import torch
 import logging
 
+from comfy.cli_args import args
+
 try:
     import comfy_kitchen as ck
     from comfy_kitchen.tensor import (
@@ -21,7 +23,15 @@ try:
             ck.registry.disable("cuda")
             logging.warning("WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.")
 
-    ck.registry.disable("triton")
+    if args.enable_triton_backend:
+        try:
+            import triton
+            logging.info("Found triton %s. Enabling comfy-kitchen triton backend.", triton.__version__)
+        except ImportError as e:
+            logging.error(f"Failed to import triton, Error: {e}, the comfy-kitchen triton backend will not be available.")
+            ck.registry.disable("triton")
+    else:
+        ck.registry.disable("triton")
     for k, v in ck.list_backends().items():
         logging.info(f"Found comfy_kitchen backend {k}: {v}")
 except ImportError as e:
@@ -37,6 +47,12 @@ except ImportError as e:
     class _CKNvfp4Layout:
         pass
 
+    class _CKSVDQuantW4A4Layout:
+        pass
+
+    class _CKAWQW4A16Layout:
+        pass
+
     def register_layout_class(name, cls):
         pass
 
@@ -55,6 +71,26 @@ if not _CK_MXFP8_AVAILABLE:
     class _CKMxfp8Layout:
         pass
 
+_CK_SVDQUANT_W4A4_AVAILABLE = False
+if _CK_AVAILABLE:
+    try:
+        from comfy_kitchen.tensor import TensorCoreSVDQuantW4A4Layout as _CKSVDQuantW4A4Layout
+        _CK_SVDQUANT_W4A4_AVAILABLE = True
+    except ImportError:
+        logging.info("comfy_kitchen does not expose SVDQuant W4A4 layout; int4 SVDQuant checkpoints will not be supported.")
+        class _CKSVDQuantW4A4Layout:
+            pass
+
+_CK_AWQ_W4A16_AVAILABLE = False
+if _CK_AVAILABLE:
+    try:
+        from comfy_kitchen.tensor import TensorCoreAWQW4A16Layout as _CKAWQW4A16Layout
+        _CK_AWQ_W4A16_AVAILABLE = True
+    except ImportError:
+        logging.info("comfy_kitchen does not expose AWQ W4A16 layout; int4 AWQ modulation checkpoints will fall back to bf16-dequantized layers.")
+        class _CKAWQW4A16Layout:
+            pass
+
 import comfy.float
 
 # ==============================================================================
@@ -162,6 +198,21 @@ class TensorCoreFP8E5M2Layout(_TensorCoreFP8LayoutBase):
     FP8_DTYPE = torch.float8_e5m2
 
 
+# SVDQuant W4A4 — pre-quantized offline (no runtime quantize), pass through the
+# kitchen-registered layout class unchanged. Comfy-side extension reserved in
+# case per-layer input scales or other Comfy-specific metadata are added later.
+class TensorCoreSVDQuantW4A4Layout(_CKSVDQuantW4A4Layout):
+    pass
+
+
+# AWQ W4A16 — pre-quantized offline (no runtime quantize) via the kitchen
+# eager `gemv_awq_w4a16` op. Used for modulation linears (img_mod.1 /
+# txt_mod.1) on Qwen-Image-Edit and similar topologies where keeping the
+# weight at int4 saves ~10 GB of VRAM vs the bf16-dequantized fallback.
+class TensorCoreAWQW4A16Layout(_CKAWQW4A16Layout):
+    pass
+
+
 # Backward compatibility alias - default to E4M3
 TensorCoreFP8Layout = TensorCoreFP8E4M3Layout
 
@@ -176,6 +227,10 @@ register_layout_class("TensorCoreFP8E5M2Layout", TensorCoreFP8E5M2Layout)
 register_layout_class("TensorCoreNVFP4Layout", TensorCoreNVFP4Layout)
 if _CK_MXFP8_AVAILABLE:
     register_layout_class("TensorCoreMXFP8Layout", TensorCoreMXFP8Layout)
+if _CK_SVDQUANT_W4A4_AVAILABLE:
+    register_layout_class("TensorCoreSVDQuantW4A4Layout", TensorCoreSVDQuantW4A4Layout)
+if _CK_AWQ_W4A16_AVAILABLE:
+    register_layout_class("TensorCoreAWQW4A16Layout", TensorCoreAWQW4A16Layout)
 
 QUANT_ALGOS = {
     "float8_e4m3fn": {
@@ -204,6 +259,22 @@ if _CK_MXFP8_AVAILABLE:
         "group_size": 32,
     }
 
+if _CK_SVDQUANT_W4A4_AVAILABLE:
+    QUANT_ALGOS["svdquant_w4a4"] = {
+        "storage_t": torch.int8,
+        "parameters": {"weight_scale", "proj_down", "proj_up", "smooth_factor"},
+        "comfy_tensor_layout": "TensorCoreSVDQuantW4A4Layout",
+        "group_size": 64,
+    }
+
+if _CK_AWQ_W4A16_AVAILABLE:
+    QUANT_ALGOS["awq_w4a16"] = {
+        "storage_t": torch.int8,
+        "parameters": {"weight_scale", "weight_zero"},
+        "comfy_tensor_layout": "TensorCoreAWQW4A16Layout",
+        "group_size": 64,
+    }
+
 
 # ==============================================================================
 # Re-exports for backward compatibility
@@ -212,10 +283,12 @@ if _CK_MXFP8_AVAILABLE:
 __all__ = [
     "QuantizedTensor",
     "QuantizedLayout",
+    "TensorCoreAWQW4A16Layout",
     "TensorCoreFP8Layout",
     "TensorCoreFP8E4M3Layout",
     "TensorCoreFP8E5M2Layout",
     "TensorCoreNVFP4Layout",
+    "TensorCoreSVDQuantW4A4Layout",
     "QUANT_ALGOS",
     "register_layout_op",
 ]

comfy/sampler_helpers.py

@@ -89,7 +89,8 @@ def get_additional_models(conds, dtype):
         gligen += get_models_from_cond(conds[k], "gligen")
         add_models += get_models_from_cond(conds[k], "additional_models")
 
-    control_nets = set(cnets)
+    # Order-preserving dedup. A plain set() would randomize iteration order across runs
+    control_nets = list(dict.fromkeys(cnets))
 
     inference_memory = 0
     control_models = []

comfy_api_nodes/nodes_sora.py

@@ -33,7 +33,7 @@ class OpenAIVideoSora2(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="OpenAIVideoSora2",
-            display_name="OpenAI Sora - Video (Deprecated)",
+            display_name="OpenAI Sora - Video (DEPRECATED)",
             category="api node/video/Sora",
             description=(
                 "OpenAI video and audio generation.\n\n"

comfy_extras/frame_interpolation_models/film_net.py

@@ -199,6 +199,9 @@ class FILMNet(nn.Module):
     def get_dtype(self):
         return self.extract.extract_sublevels.convs[0][0].conv.weight.dtype
 
+    def memory_used_forward(self, shape, dtype):
+        return 1700 * shape[1] * shape[2] * dtype.itemsize
+
     def _build_warp_grids(self, H, W, device):
         """Pre-compute warp grids for all pyramid levels."""
         if (H, W) in self._warp_grids:

comfy_extras/frame_interpolation_models/ifnet.py

@@ -74,6 +74,9 @@ class IFNet(nn.Module):
     def get_dtype(self):
         return self.encode.cnn0.weight.dtype
 
+    def memory_used_forward(self, shape, dtype):
+        return 300 * shape[1] * shape[2] * dtype.itemsize
+
     def _build_warp_grids(self, H, W, device):
         if (H, W) in self._warp_grids:
             return

comfy_extras/nodes_frame_interpolation.py

@@ -37,7 +37,7 @@ class FrameInterpolationModelLoader(io.ComfyNode):
         model = cls._detect_and_load(sd)
         dtype = torch.float16 if model_management.should_use_fp16(model_management.get_torch_device()) else torch.float32
         model.eval().to(dtype)
-        patcher = comfy.model_patcher.ModelPatcher(
+        patcher = comfy.model_patcher.CoreModelPatcher(
             model,
             load_device=model_management.get_torch_device(),
             offload_device=model_management.unet_offload_device(),
@@ -78,7 +78,7 @@ class FrameInterpolate(io.ComfyNode):
         return io.Schema(
             node_id="FrameInterpolate",
             display_name="Frame Interpolate",
-            category="image/video",
+            category="video",
             search_aliases=["rife", "film", "frame interpolation", "slow motion", "interpolate frames", "vfi"],
             inputs=[
                 FrameInterpolationModel.Input("interp_model"),
@@ -98,16 +98,13 @@ class FrameInterpolate(io.ComfyNode):
         if num_frames < 2 or multiplier < 2:
             return io.NodeOutput(images)
 
-        model_management.load_model_gpu(interp_model)
         device = interp_model.load_device
         dtype = interp_model.model_dtype()
         inference_model = interp_model.model
-
-        # Free VRAM for inference activations (model weights + ~20x a single frame's worth)
-        H, W = images.shape[1], images.shape[2]
-        activation_mem = H * W * 3 * images.element_size() * 20
-        model_management.free_memory(activation_mem, device)
+        activation_mem = inference_model.memory_used_forward(images.shape, dtype)
+        model_management.load_models_gpu([interp_model], memory_required=activation_mem)
         align = getattr(inference_model, "pad_align", 1)
+        H, W = images.shape[1], images.shape[2]
 
         # Prepare a single padded frame on device for determining output dimensions
         def prepare_frame(idx):

comfy_extras/nodes_image_compare.py

@@ -11,7 +11,7 @@ class ImageCompare(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ImageCompare",
-            display_name="Image Compare",
+            display_name="Compare Images",
             description="Compares two images side by side with a slider.",
             category="image",
             essentials_category="Image Tools",

comfy_extras/nodes_images.py

@@ -24,7 +24,7 @@ class ImageCrop(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageCrop",
             search_aliases=["trim"],
-            display_name="Image Crop (Deprecated)",
+            display_name="Crop Image (DEPRECATED)",
             category="image/transform",
             is_deprecated=True,
             essentials_category="Image Tools",
@@ -56,7 +56,7 @@ class ImageCropV2(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageCropV2",
             search_aliases=["trim"],
-            display_name="Image Crop",
+            display_name="Crop Image",
             category="image/transform",
             essentials_category="Image Tools",
             has_intermediate_output=True,
@@ -109,6 +109,7 @@ class RepeatImageBatch(IO.ComfyNode):
         return IO.Schema(
             node_id="RepeatImageBatch",
             search_aliases=["duplicate image", "clone image"],
+            display_name="Repeat Image Batch",
             category="image/batch",
             inputs=[
                 IO.Image.Input("image"),
@@ -131,6 +132,7 @@ class ImageFromBatch(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageFromBatch",
             search_aliases=["select image", "pick from batch", "extract image"],
+            display_name="Get Image from Batch",
             category="image/batch",
             inputs=[
                 IO.Image.Input("image"),
@@ -157,7 +159,8 @@ class ImageAddNoise(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageAddNoise",
             search_aliases=["film grain"],
-            category="image",
+            display_name="Add Noise to Image",
+            category="image/postprocessing",
             inputs=[
                 IO.Image.Input("image"),
                 IO.Int.Input(
@@ -259,7 +262,7 @@ class ImageStitch(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageStitch",
             search_aliases=["combine images", "join images", "concatenate images", "side by side"],
-            display_name="Image Stitch",
+            display_name="Stitch Images",
             description="Stitches image2 to image1 in the specified direction.\n"
             "If image2 is not provided, returns image1 unchanged.\n"
             "Optional spacing can be added between images.",
@@ -434,6 +437,7 @@ class ResizeAndPadImage(IO.ComfyNode):
         return IO.Schema(
             node_id="ResizeAndPadImage",
             search_aliases=["fit to size"],
+            display_name="Resize And Pad Image",
             category="image/transform",
             inputs=[
                 IO.Image.Input("image"),
@@ -485,6 +489,7 @@ class SaveSVGNode(IO.ComfyNode):
         return IO.Schema(
             node_id="SaveSVGNode",
             search_aliases=["export vector", "save vector graphics"],
+            display_name="Save SVG",
             description="Save SVG files on disk.",
             category="image/save",
             inputs=[
@@ -591,7 +596,7 @@ class ImageRotate(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ImageRotate",
-            display_name="Image Rotate",
+            display_name="Rotate Image",
             search_aliases=["turn", "flip orientation"],
             category="image/transform",
             essentials_category="Image Tools",
@@ -624,6 +629,7 @@ class ImageFlip(IO.ComfyNode):
         return IO.Schema(
             node_id="ImageFlip",
             search_aliases=["mirror", "reflect"],
+            display_name="Flip Image",
             category="image/transform",
             inputs=[
                 IO.Image.Input("image"),
@@ -650,6 +656,7 @@ class ImageScaleToMaxDimension(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ImageScaleToMaxDimension",
+            display_name="Scale Image to Max Dimension",
             category="image/upscaling",
             inputs=[
                 IO.Image.Input("image"),

comfy_extras/nodes_mask.py

@@ -80,7 +80,8 @@ class ImageCompositeMasked(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ImageCompositeMasked",
-            search_aliases=["paste image", "overlay", "layer"],
+            search_aliases=["overlay", "layer", "paste image", "images composition"],
+            display_name="Image Composite Masked",
             category="image",
             inputs=[
                 IO.Image.Input("destination"),
@@ -201,6 +202,7 @@ class InvertMask(IO.ComfyNode):
         return IO.Schema(
             node_id="InvertMask",
             search_aliases=["reverse mask", "flip mask"],
+            display_name="Invert Mask",
             category="mask",
             inputs=[
                 IO.Mask.Input("mask"),
@@ -222,6 +224,7 @@ class CropMask(IO.ComfyNode):
         return IO.Schema(
             node_id="CropMask",
             search_aliases=["cut mask", "extract mask region", "mask slice"],
+            display_name="Crop Mask",
             category="mask",
             inputs=[
                 IO.Mask.Input("mask"),
@@ -247,7 +250,8 @@ class MaskComposite(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="MaskComposite",
-            search_aliases=["combine masks", "blend masks", "layer masks"],
+            search_aliases=["combine masks", "blend masks", "layer masks", "masks composition"],
+            display_name="Combine Masks",
             category="mask",
             inputs=[
                 IO.Mask.Input("destination"),
@@ -298,6 +302,7 @@ class FeatherMask(IO.ComfyNode):
         return IO.Schema(
             node_id="FeatherMask",
             search_aliases=["soft edge mask", "blur mask edges", "gradient mask edge"],
+            display_name="Feather Mask",
             category="mask",
             inputs=[
                 IO.Mask.Input("mask"),

comfy_extras/nodes_morphology.py

@@ -59,7 +59,8 @@ class ImageRGBToYUV(io.ComfyNode):
         return io.Schema(
             node_id="ImageRGBToYUV",
             search_aliases=["color space conversion"],
-            category="image/batch",
+            display_name="Image RGB to YUV",
+            category="image/color",
             inputs=[
                 io.Image.Input("image"),
             ],
@@ -81,7 +82,8 @@ class ImageYUVToRGB(io.ComfyNode):
         return io.Schema(
             node_id="ImageYUVToRGB",
             search_aliases=["color space conversion"],
-            category="image/batch",
+            display_name="Image YUV to RGB",
+            category="image/color",
             inputs=[
                 io.Image.Input("Y"),
                 io.Image.Input("U"),

comfy_extras/nodes_post_processing.py

@@ -20,7 +20,8 @@ class Blend(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageBlend",
-            display_name="Image Blend",
+            search_aliases=["mix images"],
+            display_name="Blend Images",
             category="image/postprocessing",
             essentials_category="Image Tools",
             inputs=[
@@ -224,6 +225,7 @@ class ImageScaleToTotalPixels(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageScaleToTotalPixels",
+            display_name="Scale Image to Total Pixels",
             category="image/upscaling",
             inputs=[
                 io.Image.Input("image"),
@@ -568,7 +570,7 @@ class BatchImagesNode(io.ComfyNode):
         return io.Schema(
             node_id="BatchImagesNode",
             display_name="Batch Images",
-            category="image",
+            category="image/batch",
             essentials_category="Image Tools",
             search_aliases=["batch", "image batch", "batch images", "combine images", "merge images", "stack images"],
             inputs=[
@@ -666,12 +668,13 @@ class ColorTransfer(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ColorTransfer",
+            display_name="Color Transfer",
             category="image/postprocessing",
             description="Match the colors of one image to another using various algorithms.",
             search_aliases=["color match", "color grading", "color correction", "match colors", "color transform", "mkl", "reinhard", "histogram"],
             inputs=[
                 io.Image.Input("image_target", tooltip="Image(s) to apply the color transform to."),
-                io.Image.Input("image_ref", optional=True, tooltip="Reference image(s) to match colors to. If not provided, processing is skipped"),
+                io.Image.Input("image_ref", tooltip="Reference image(s) to match colors to."),
                 io.Combo.Input("method", options=['reinhard_lab', 'mkl_lab', 'histogram'],),
                 io.DynamicCombo.Input("source_stats",
                     tooltip="per_frame: each frame matched to image_ref individually. uniform: pool stats across all source frames as baseline, match to image_ref. target_frame: use one chosen frame as the baseline for the transform to image_ref, applied uniformly to all frames (preserves relative differences)",

comfy_extras/nodes_primitive.py

@@ -49,7 +49,7 @@ class Int(io.ComfyNode):
             display_name="Int",
             category="utils/primitive",
             inputs=[
-                io.Int.Input("value", min=-sys.maxsize, max=sys.maxsize, control_after_generate=True),
+                io.Int.Input("value", min=-sys.maxsize, max=sys.maxsize, control_after_generate=io.ControlAfterGenerate.fixed),
             ],
             outputs=[io.Int.Output()],
         )

comfy_extras/nodes_video.py

@@ -17,7 +17,8 @@ class SaveWEBM(io.ComfyNode):
         return io.Schema(
             node_id="SaveWEBM",
             search_aliases=["export webm"],
-            category="image/video",
+            display_name="Save WEBM",
+            category="video",
             is_experimental=True,
             inputs=[
                 io.Image.Input("images"),
@@ -72,7 +73,7 @@ class SaveVideo(io.ComfyNode):
             node_id="SaveVideo",
             search_aliases=["export video"],
             display_name="Save Video",
-            category="image/video",
+            category="video",
             essentials_category="Basics",
             description="Saves the input images to your ComfyUI output directory.",
             inputs=[
@@ -121,7 +122,7 @@ class CreateVideo(io.ComfyNode):
             node_id="CreateVideo",
             search_aliases=["images to video"],
             display_name="Create Video",
-            category="image/video",
+            category="video",
             description="Create a video from images.",
             inputs=[
                 io.Image.Input("images", tooltip="The images to create a video from."),
@@ -146,7 +147,7 @@ class GetVideoComponents(io.ComfyNode):
             node_id="GetVideoComponents",
             search_aliases=["extract frames", "split video", "video to images", "demux"],
             display_name="Get Video Components",
-            category="image/video",
+            category="video",
             description="Extracts all components from a video: frames, audio, and framerate.",
             inputs=[
                 io.Video.Input("video", tooltip="The video to extract components from."),
@@ -174,7 +175,7 @@ class LoadVideo(io.ComfyNode):
             node_id="LoadVideo",
             search_aliases=["import video", "open video", "video file"],
             display_name="Load Video",
-            category="image/video",
+            category="video",
             essentials_category="Basics",
             inputs=[
                 io.Combo.Input("file", options=sorted(files), upload=io.UploadType.video),
@@ -216,7 +217,7 @@ class VideoSlice(io.ComfyNode):
                 "frame load cap",
                 "start time",
             ],
-            category="image/video",
+            category="video",
             essentials_category="Video Tools",
             inputs=[
                 io.Video.Input("video"),

extra_model_paths.yaml.example

@@ -28,7 +28,7 @@
 #config for a1111 ui
 #all you have to do is uncomment this (remove the #) and change the base_path to where yours is installed
 
-#a111:
+#a1111:
 #     base_path: path/to/stable-diffusion-webui/
 #     checkpoints: models/Stable-diffusion
 #     configs: models/Stable-diffusion

nodes.py

@@ -1754,57 +1754,49 @@ class LoadImage:
 
         return True
 
-class LoadImageMask:
+
+class LoadImageMask(LoadImage):
     ESSENTIALS_CATEGORY = "Image Tools"
     SEARCH_ALIASES = ["import mask", "alpha mask", "channel mask"]
 
     _color_channels = ["alpha", "red", "green", "blue"]
+
     @classmethod
     def INPUT_TYPES(s):
-        input_dir = folder_paths.get_input_directory()
-        files = [f for f in os.listdir(input_dir) if os.path.isfile(os.path.join(input_dir, f))]
-        return {"required":
-                    {"image": (sorted(files), {"image_upload": True}),
-                     "channel": (s._color_channels, ), }
-                }
+        types = super().INPUT_TYPES()
+        return {
+            "required": {
+                **types["required"],
+                "channel": (s._color_channels, )
+            }
+        }
 
     CATEGORY = "mask"
-
     RETURN_TYPES = ("MASK",)
-    FUNCTION = "load_image"
-    def load_image(self, image, channel):
-        image_path = folder_paths.get_annotated_filepath(image)
-        i = node_helpers.pillow(Image.open, image_path)
-        i = node_helpers.pillow(ImageOps.exif_transpose, i)
-        if i.getbands() != ("R", "G", "B", "A"):
-            if i.mode == 'I':
-                i = i.point(lambda i: i * (1 / 255))
-            i = i.convert("RGBA")
-        mask = None
+    FUNCTION = "load_image_mask"
+
+    def load_image_mask(self, image, channel):
+        image_tensor, mask_tensor = super().load_image(image)
         c = channel[0].upper()
-        if c in i.getbands():
-            mask = np.array(i.getchannel(c)).astype(np.float32) / 255.0
-            mask = torch.from_numpy(mask)
-            if c == 'A':
-                mask = 1. - mask
+
+        if c == 'A':
+            return (mask_tensor,)
+
+        channel_idx = {'R': 0, 'G': 1, 'B': 2}.get(c, 0)
+
+        if channel_idx < image_tensor.shape[-1]:
+            return (image_tensor[..., channel_idx].clone(),)
         else:
-            mask = torch.zeros((64,64), dtype=torch.float32, device="cpu")
-        return (mask.unsqueeze(0),)
+            empty_mask = torch.zeros(
+                image_tensor.shape[:-1],
+                dtype=image_tensor.dtype,
+                device=image_tensor.device
+            )
+            return (empty_mask,)
 
     @classmethod
     def IS_CHANGED(s, image, channel):
-        image_path = folder_paths.get_annotated_filepath(image)
-        m = hashlib.sha256()
-        with open(image_path, 'rb') as f:
-            m.update(f.read())
-        return m.digest().hex()
-
-    @classmethod
-    def VALIDATE_INPUTS(s, image):
-        if not folder_paths.exists_annotated_filepath(image):
-            return "Invalid image file: {}".format(image)
-
-        return True
+        return super().IS_CHANGED(image)
 
 
 class LoadImageOutput(LoadImage):
@@ -1895,7 +1887,7 @@ class ImageInvert:
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "invert"
 
-    CATEGORY = "image"
+    CATEGORY = "image/color"
 
     def invert(self, image):
         s = 1.0 - image
@@ -1911,7 +1903,7 @@ class ImageBatch:
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "batch"
 
-    CATEGORY = "image"
+    CATEGORY = "image/batch"
     DEPRECATED = True
 
     def batch(self, image1, image2):
@@ -1968,7 +1960,7 @@ class ImagePadForOutpaint:
     RETURN_TYPES = ("IMAGE", "MASK")
     FUNCTION = "expand_image"
 
-    CATEGORY = "image"
+    CATEGORY = "image/transform"
 
     def expand_image(self, image, left, top, right, bottom, feathering):
         d1, d2, d3, d4 = image.size()
@@ -2111,7 +2103,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "ConditioningSetArea": "Conditioning (Set Area)",
     "ConditioningSetAreaPercentage": "Conditioning (Set Area with Percentage)",
     "ConditioningSetMask": "Conditioning (Set Mask)",
-    "ControlNetApply": "Apply ControlNet (OLD)",
+    "ControlNetApply": "Apply ControlNet (DEPRECATED)",
     "ControlNetApplyAdvanced": "Apply ControlNet",
     # Latent
     "VAEEncodeForInpaint": "VAE Encode (for Inpainting)",
@@ -2129,6 +2121,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "LatentFromBatch" : "Latent From Batch",
     "RepeatLatentBatch": "Repeat Latent Batch",
     # Image
+    "EmptyImage": "Empty Image",
     "SaveImage": "Save Image",
     "PreviewImage": "Preview Image",
     "LoadImage": "Load Image",
@@ -2136,15 +2129,15 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "LoadImageOutput": "Load Image (from Outputs)",
     "ImageScale": "Upscale Image",
     "ImageScaleBy": "Upscale Image By",
-    "ImageInvert": "Invert Image",
+    "ImageInvert": "Invert Image Colors",
     "ImagePadForOutpaint": "Pad Image for Outpainting",
-    "ImageBatch": "Batch Images",
-    "ImageCrop": "Image Crop",
-    "ImageStitch": "Image Stitch",
-    "ImageBlend": "Image Blend",
-    "ImageBlur": "Image Blur",
-    "ImageQuantize": "Image Quantize",
-    "ImageSharpen": "Image Sharpen",
+    "ImageBatch": "Batch Images (DEPRECATED)",
+    "ImageCrop": "Crop Image",
+    "ImageStitch": "Stitch Images",
+    "ImageBlend": "Blend Images",
+    "ImageBlur": "Blur Image",
+    "ImageQuantize": "Quantize Image",
+    "ImageSharpen": "Sharpen Image",
     "ImageScaleToTotalPixels": "Scale Image to Total Pixels",
     "GetImageSize": "Get Image Size",
     # _for_testing