aid.patch

diff --git a/comfy/lora.py b/comfy/lora.py
index 8760a21f..e6456509 100644
--- a/comfy/lora.py
+++ b/comfy/lora.py
@@ -44,6 +44,12 @@ def load_lora(lora, to_load, log_missing=True):
             alpha = lora[alpha_name].item()
             loaded_keys.add(alpha_name)
 
+        aid_p_name = "{}.aid_p".format(x)
+        aid_p = None
+        if aid_p_name in lora.keys():
+            aid_p = lora[aid_p_name].item()
+            loaded_keys.add(aid_p_name)
+
         dora_scale_name = "{}.dora_scale".format(x)
         dora_scale = None
         if dora_scale_name in lora.keys():
@@ -53,6 +59,8 @@ def load_lora(lora, to_load, log_missing=True):
         for adapter_cls in weight_adapter.adapters:
             adapter = adapter_cls.load(x, lora, alpha, dora_scale, loaded_keys)
             if adapter is not None:
+                if hasattr(adapter, "add_aid") and aid_p is not None:
+                    adapter.add_aid(aid_p)
                 patch_dict[to_load[x]] = adapter
                 loaded_keys.update(adapter.loaded_keys)
                 continue
diff --git a/comfy/weight_adapter/lora.py b/comfy/weight_adapter/lora.py
index b2e62392..5f66a16a 100644
--- a/comfy/weight_adapter/lora.py
+++ b/comfy/weight_adapter/lora.py
@@ -5,6 +5,20 @@ import torch
 import comfy.model_management
 from .base import WeightAdapterBase, weight_decompose, pad_tensor_to_shape
 
+class AID(torch.nn.Module):
+    def __init__(self, p=0.9):
+        super().__init__()
+        self.p = p
+
+    def forward(self, x: torch.Tensor):
+        if self.training:
+            pos_mask = (x >= 0) * torch.bernoulli(torch.ones_like(x) * self.p)
+            neg_mask = (x < 0) * torch.bernoulli(torch.ones_like(x) * (1 - self.p))
+            return x * (pos_mask + neg_mask)
+        else:
+            pos_part = (x >= 0) * x * self.p
+            neg_part = (x < 0) * x * (1 - self.p)
+            return pos_part + neg_part
 
 class LoRAAdapter(WeightAdapterBase):
     name = "lora"
@@ -13,6 +27,8 @@ class LoRAAdapter(WeightAdapterBase):
         self.loaded_keys = loaded_keys
         self.weights = weights
 
+        self.aid: AID | None = None
+
     @classmethod
     def load(
         cls,
@@ -78,6 +94,9 @@ class LoRAAdapter(WeightAdapterBase):
         else:
             return None
 
+    def add_aid(self, aid_p: float):
+        self.aid = AID(p=aid_p)
+
     def calculate_weight(
         self,
         weight,
@@ -125,6 +144,8 @@ class LoRAAdapter(WeightAdapterBase):
             lora_diff = torch.mm(
                 mat1.flatten(start_dim=1), mat2.flatten(start_dim=1)
             ).reshape(weight.shape)
+            if self.aid is not None:
+                lora_diff = self.aid(lora_diff)
             if dora_scale is not None:
                 weight = weight_decompose(
                     dora_scale,