Merge b8688389e5 into b138133ffa

comfy_extras/frame_interpolation_models/film_net.py

@@ -199,9 +199,6 @@ 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

@@ -74,9 +74,6 @@ 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.CoreModelPatcher(
+        patcher = comfy.model_patcher.ModelPatcher(
             model,
             load_device=model_management.get_torch_device(),
             offload_device=model_management.unet_offload_device(),
@@ -98,13 +98,16 @@ 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
-        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)
+
+        # 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)
+        align = getattr(inference_model, "pad_align", 1)
 
         # Prepare a single padded frame on device for determining output dimensions
         def prepare_frame(idx):

comfy_extras/nodes_post_processing.py

@@ -666,13 +666,12 @@ 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", tooltip="Reference image(s) to match colors to."),
+                io.Image.Input("image_ref", optional=True, tooltip="Reference image(s) to match colors to. If not provided, processing is skipped"),
                 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=io.ControlAfterGenerate.fixed),
+                io.Int.Input("value", min=-sys.maxsize, max=sys.maxsize, control_after_generate=True),
             ],
             outputs=[io.Int.Output()],
         )

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
 
-#a1111:
+#a111:
 #     base_path: path/to/stable-diffusion-webui/
 #     checkpoints: models/Stable-diffusion
 #     configs: models/Stable-diffusion

nodes.py

@@ -1754,49 +1754,57 @@ class LoadImage:
 
         return True
 
-
-class LoadImageMask(LoadImage):
+class LoadImageMask:
     ESSENTIALS_CATEGORY = "Image Tools"
     SEARCH_ALIASES = ["import mask", "alpha mask", "channel mask"]
 
     _color_channels = ["alpha", "red", "green", "blue"]
-
     @classmethod
     def INPUT_TYPES(s):
-        types = super().INPUT_TYPES()
-        return {
-            "required": {
-                **types["required"],
-                "channel": (s._color_channels, )
-            }
-        }
+        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, ), }
+                }
 
     CATEGORY = "mask"
+
     RETURN_TYPES = ("MASK",)
-    FUNCTION = "load_image_mask"
-
-    def load_image_mask(self, image, channel):
-        image_tensor, mask_tensor = super().load_image(image)
+    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
         c = channel[0].upper()
-
-        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(),)
+        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
         else:
-            empty_mask = torch.zeros(
-                image_tensor.shape[:-1],
-                dtype=image_tensor.dtype,
-                device=image_tensor.device
-            )
-            return (empty_mask,)
+            mask = torch.zeros((64,64), dtype=torch.float32, device="cpu")
+        return (mask.unsqueeze(0),)
 
     @classmethod
     def IS_CHANGED(s, image, channel):
-        return super().IS_CHANGED(image)
+        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
 
 
 class LoadImageOutput(LoadImage):