diff --git a/image_filter_nodes.py b/image_filter_nodes.py
index 476077a..a5432b4 100644
--- a/image_filter_nodes.py
+++ b/image_filter_nodes.py
@@ -207,22 +207,33 @@ def mask_from_data(data) -> torch.Tensor:
     return mask.unsqueeze(0)
 
 class InOutStore:
-    previous_inputs:list[Any] = []
-    last_output:tuple[torch.Tensor, torch.Tensor|None, str, str, str]|None = None
-
+    stores:dict[str, "InOutStore"] = {}
     @classmethod
-    def check_input_unchanged(cls, *args) -> bool:
+    def get_store(cls, graph_id:str) -> "InOutStore":
+        if graph_id not in cls.stores:
+            cls.stores[graph_id] = InOutStore()
+        return cls.stores[graph_id]
+
+    def __init__(self): 
+        self.previous_inputs:list[Any] = []
+        self.last_output:tuple[torch.Tensor, torch.Tensor|None, str, str, str]|None = None
+
+    def get_last(self) -> tuple[torch.Tensor, torch.Tensor|None, str, str, str]:
+        assert self.last_output is not None, "No last output stored"
+        return self.last_output
+
+    def check_input_unchanged(self, *args) -> bool:
         def make_copy(x): return x.clone() if isinstance(x, torch.Tensor) else x
         try:
-            if len(cls.previous_inputs)!=len(args): return False
-            for prev, new in zip(cls.previous_inputs, args):
+            if len(self.previous_inputs)!=len(args): return False
+            for prev, new in zip(self.previous_inputs, args):
                 if isinstance(prev, torch.Tensor) and isinstance(new, torch.Tensor):
                     if not torch.equal(prev, new): return False
                 else:
                     if prev != new: return False
             return True
         finally:
-            cls.previous_inputs = [ make_copy(x) for x in args ]
+            self.previous_inputs = [ make_copy(x) for x in args ]
     
 class MaskImageFilter(io.ComfyNode, FilterNodeBase):
     @classmethod
@@ -240,7 +251,10 @@ class MaskImageFilter(io.ComfyNode, FilterNodeBase):
                 io.String.Input("extra1", default="", optional=True),
                 io.String.Input("extra2", default="", optional=True),
                 io.String.Input("extra3", default="", optional=True),
-                io.String.Input("graph_id", default="")
+                io.String.Input("graph_id", default=""),
+            ],
+            hidden=[
+                io.Hidden.unique_id,
             ],
             outputs = [
                 io.Image.Output("image", display_name="image"),
@@ -254,13 +268,15 @@ class MaskImageFilter(io.ComfyNode, FilterNodeBase):
 
     @classmethod
     def execute(cls, image, timeout, if_no_mask, graph_id, if_inputs_unchanged="Run normally", mask=None, extra1="", extra2="", extra3="", tip="", **kwargs): # type: ignore
+        iostore = InOutStore.get_store(f"{graph_id}_{cls.hidden.unique_id}")
+
         # check if everything is unchanged (and store these inputs for next check)
-        if InOutStore.check_input_unchanged(image, timeout, if_no_mask, graph_id, mask, extra1, extra2, extra3, tip) and InOutStore.last_output is not None:
+        if iostore.check_input_unchanged(image, timeout, if_no_mask, graph_id, mask, extra1, extra2, extra3, tip) and iostore.last_output is not None:
             if if_inputs_unchanged == "Start with last output":
-                image, mask, extra1, extra2, extra3 = InOutStore.last_output
+                image, mask, extra1, extra2, extra3 = iostore.get_last()
                 mask = 1.0 - mask if mask is not None else None  # The mask editor works in inverse
             elif if_inputs_unchanged == "Resend last output":
-                return io.NodeOutput( *InOutStore.last_output )
+                return io.NodeOutput( *iostore.get_last() )
             
         if mask is not None and mask.shape[:3] == image.shape[:3] and not torch.all(mask==0):
             input_to_send = torch.cat((image, mask.unsqueeze(-1)), dim=-1)
@@ -284,10 +300,10 @@ class MaskImageFilter(io.ComfyNode, FilterNodeBase):
         if mask is None: mask = torch.zeros_like(image[...,0]) 
         if if_no_mask == 'cancel' and torch.all(mask==0): raise InterruptProcessingException() 
 
-        InOutStore.last_output = ( image.clone(), mask.clone(), *response.get_extras((extra1, extra2, extra3)) )
+        iostore.last_output = ( image.clone(), mask.clone(), *response.get_extras((extra1, extra2, extra3)) )
         if (image.shape[0:3] != mask.shape[0:3]):
             print(f"Mask shape {mask.shape} does not match image shape {image.shape}")
-        return io.NodeOutput( *InOutStore.last_output )
+        return io.NodeOutput( *iostore.get_last() )
 
     @classmethod
     def fingerprint_inputs(cls, **kwargs) -> Any: