diff --git a/animatediff/utils/util.py b/animatediff/utils/util.py
index b0dabe9..b28ba9c 100644
--- a/animatediff/utils/util.py
+++ b/animatediff/utils/util.py
@@ -94,7 +94,7 @@ def ddim_inversion(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt
 def load_weights(
     animation_pipeline,
     # motion module
-    motion_module_path         = "",
+    motion_model         = "",
     motion_module_lora_configs = [],
     # domain adapter
     adapter_lora_path          = "",
@@ -106,20 +106,24 @@ def load_weights(
 ):
     # motion module
     unet_state_dict = {}
-    if isinstance(motion_module_path, str) and motion_module_path != "":
-        print(f"load motion module from {motion_module_path}")
-        if motion_module_path.endswith(".safetensors"):
+    if isinstance(motion_model, str) and motion_model != "":
+        print(f"load motion module from {motion_model}")
+        if motion_model.endswith(".safetensors"):
             motion_module_state_dict = {}
-            with safe_open(motion_module_path, framework="pt", device="cpu") as f:
+            with safe_open(motion_model, framework="pt", device="cpu") as f:
                 for key in f.keys():
                     motion_module_state_dict[key] = f.get_tensor(key)
-        elif motion_module_path.endswith(".ckpt"):
-            motion_module_state_dict = torch.load(motion_module_path, map_location="cpu")  
+        elif motion_model.endswith(".ckpt"):
+            motion_module_state_dict = torch.load(motion_model, map_location="cpu")  
         motion_module_state_dict = motion_module_state_dict["state_dict"] if "state_dict" in motion_module_state_dict else motion_module_state_dict
         unet_state_dict.update({name: param for name, param in motion_module_state_dict.items() if "motion_modules." in name})
         unet_state_dict.pop("animatediff_config", "")
     else:
-        motion_module_state_dict = motion_module_path.model.state_dict()
+        unet_state_dict = {}
+        motion_module_state_dict = motion_model.model.state_dict()
+        if motion_model.model.mm_info.mm_format == "AnimateLCM":
+            motion_module_state_dict = {k: v for k, v in motion_module_state_dict.items() if "pos_encoder" not in k}
+        
         unet_state_dict.update({name: param for name, param in motion_module_state_dict.items() if "motion_modules." in name})
         unet_state_dict.pop("animatediff_config", "")
 
diff --git a/nodes.py b/nodes.py
index 8eea39f..936e303 100644
--- a/nodes.py
+++ b/nodes.py
@@ -558,11 +558,11 @@ class ADMD_DiffusersLoader:
                 unet=unet, vae=vae, tokenizer=tokenizer, text_encoder=text_encoder, scheduler=noise_scheduler,
             ).to(device)
 
-            motion_module_path, domain_adapter_path = additional_models 
+            motion_model, domain_adapter_path = additional_models 
 
             validation_pipeline = load_weights(
                 validation_pipeline, 
-                motion_module_path=motion_module_path,
+                motion_model=motion_model,
                 adapter_lora_path=domain_adapter_path, 
                 dreambooth_model_path=""
             )
@@ -687,11 +687,11 @@ class ADMD_CheckpointLoader:
             # Enable gradient checkpointing
             unet.enable_gradient_checkpointing()
 
-            motion_module_path, domain_adapter_path = additional_models 
+            motion_model, domain_adapter_path = additional_models 
 
             validation_pipeline = load_weights(
                 validation_pipeline, 
-                motion_module_path=motion_module_path,
+                motion_model=motion_model,
                 adapter_lora_path=domain_adapter_path, 
                 dreambooth_model_path=""
             )
@@ -741,7 +741,7 @@ class ADMD_ComfyModelLoader:
 
     def load_checkpoint(self, model, clip, vae, scheduler, use_xformers, motion_model):
         with torch.inference_mode(False):
-            print(motion_model)
+        
             pbar = comfy.utils.ProgressBar(4)
             original_config = OmegaConf.load(os.path.join(script_directory, f"configs/v1-inference.yaml"))
             ad_unet_config = OmegaConf.load(os.path.join(script_directory, f"configs/ad_unet_config.yaml"))
@@ -823,7 +823,7 @@ class ADMD_ComfyModelLoader:
 
             validation_pipeline = load_weights(
                 validation_pipeline, 
-                motion_module_path=motion_model,
+                motion_model=motion_model,
                 adapter_lora_path=domain_adapter_path, 
                 dreambooth_model_path=""
             )
@@ -862,9 +862,9 @@ class ADMD_AdditionalModelSelect:
 
     def select_models(self, motion_module, use_adapter_lora, optional_adapter_lora=""):
         additional_models = []
-        motion_module_path = folder_paths.get_full_path("animatediff_models", motion_module)
-        if not Path(motion_module_path).is_file():
-            raise ValueError(f"Motion model {motion_module_path} does not exist")
+        motion_model = folder_paths.get_full_path("animatediff_models", motion_module)
+        if not Path(motion_model).is_file():
+            raise ValueError(f"Motion model {motion_model} does not exist")
         if use_adapter_lora:
             adapter_lora_path = folder_paths.get_full_path("loras", optional_adapter_lora)
             if not Path(adapter_lora_path).is_file():
@@ -872,7 +872,7 @@ class ADMD_AdditionalModelSelect:
         else:
             adapter_lora_path = ""        
 
-        additional_models.append(motion_module_path)
+        additional_models.append(motion_model)
         additional_models.append(adapter_lora_path)  
         return (additional_models,)