diff --git a/__init__.py b/__init__.py
index c6ab4ef..7bfa56d 100644
--- a/__init__.py
+++ b/__init__.py
@@ -50,3 +50,100 @@ class MZ_Flux1PartialLoad_Patch:
 
 NODE_CLASS_MAPPINGS["MZ_Flux1PartialLoad_Patch"] = MZ_Flux1PartialLoad_Patch
 NODE_DISPLAY_NAME_MAPPINGS["MZ_Flux1PartialLoad_Patch"] = f"{AUTHOR_NAME} - Flux1PartialLoad_Patch"
+
+import nodes
+
+
+class MZ_Flux1CheckpointLoaderNF4_cpuDynOffload:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "ckpt_name": (folder_paths.get_filename_list("checkpoints"), ),
+            "double_blocks_cuda_size": ("INT", {"min": 0, "max": 16, "default": 7}),
+            "single_blocks_cuda_size": ("INT", {"min": 0, "max": 37, "default": 7}),
+        }}
+
+    RETURN_TYPES = ("MODEL", "CLIP", "VAE")
+    FUNCTION = "load_checkpoint"
+
+    CATEGORY = f"{CATEGORY_NAME}"
+
+    def load_checkpoint(self, ckpt_name, **kwargs):
+        CheckpointLoaderNF4 = nodes.NODE_CLASS_MAPPINGS.get(
+            "CheckpointLoaderNF4", None)
+        if CheckpointLoaderNF4 is None:
+            # 必须安装 https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4
+            raise Exception(
+                "Please install comfyanonymous/ComfyUI_bitsandbytes_NF4 to use this node.")
+
+        model, clip, vae = CheckpointLoaderNF4().load_checkpoint(ckpt_name)
+        return mz_fluxext_core.Flux1PartialLoad_Patch({
+            "model": model,
+            "double_blocks_cuda_size": kwargs.get("double_blocks_cuda_size", 7),
+            "single_blocks_cuda_size": kwargs.get("single_blocks_cuda_size", 7),
+        })[0], clip, vae
+
+
+NODE_CLASS_MAPPINGS["MZ_Flux1CheckpointLoaderNF4_cpuDynOffload"] = MZ_Flux1CheckpointLoaderNF4_cpuDynOffload
+NODE_DISPLAY_NAME_MAPPINGS[
+    "MZ_Flux1CheckpointLoaderNF4_cpuDynOffload"] = f"{AUTHOR_NAME} - Flux1CheckpointLoaderNF4_cpuDynOffload"
+
+
+class MZ_Flux1CheckpointLoader_cpuDynOffload:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "ckpt_name": (folder_paths.get_filename_list("checkpoints"), ),
+            "double_blocks_cuda_size": ("INT", {"min": 0, "max": 16, "default": 7}),
+            "single_blocks_cuda_size": ("INT", {"min": 0, "max": 37, "default": 7}),
+        }}
+
+    RETURN_TYPES = ("MODEL", "CLIP", "VAE")
+    FUNCTION = "load_checkpoint"
+
+    CATEGORY = f"{CATEGORY_NAME}"
+
+    def load_checkpoint(self, ckpt_name, **kwargs):
+        model, clip, vae = nodes.CheckpointLoaderSimple().load_checkpoint(
+            ckpt_name=ckpt_name)
+        return mz_fluxext_core.Flux1PartialLoad_Patch({
+            "model": model,
+            "double_blocks_cuda_size": kwargs.get("double_blocks_cuda_size", 7),
+            "single_blocks_cuda_size": kwargs.get("single_blocks_cuda_size", 7),
+        })[0], clip, vae
+
+
+NODE_CLASS_MAPPINGS["MZ_Flux1CheckpointLoader_cpuDynOffload"] = MZ_Flux1CheckpointLoader_cpuDynOffload
+NODE_DISPLAY_NAME_MAPPINGS[
+    "MZ_Flux1CheckpointLoader_cpuDynOffload"] = f"{AUTHOR_NAME} - Flux1CheckpointLoader_cpuDynOffload"
+
+
+class MZ_Flux1UnetLoader_cpuDynOffload:
+    @classmethod
+    def INPUT_TYPES(s):
+        args = nodes.UNETLoader().INPUT_TYPES()
+        args["required"]["double_blocks_cuda_size"] = (
+            "INT", {"min": 0, "max": 16, "default": 7})
+        args["required"]["single_blocks_cuda_size"] = (
+            "INT", {"min": 0, "max": 37, "default": 7})
+        return args
+
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "load_unet"
+
+    CATEGORY = f"{CATEGORY_NAME}"
+
+    def load_unet(self, **kwargs):
+        model = nodes.UNETLoader().load_unet(
+            **{k: v for k, v in kwargs.items() if k != "double_blocks_cuda_size" and k != "single_blocks_cuda_size"})[0]
+
+        return mz_fluxext_core.Flux1PartialLoad_Patch({
+            "model": model,
+            "double_blocks_cuda_size": kwargs.get("double_blocks_cuda_size", 7),
+            "single_blocks_cuda_size": kwargs.get("single_blocks_cuda_size", 7),
+        })
+
+
+NODE_CLASS_MAPPINGS["MZ_Flux1UnetLoader_cpuDynOffload"] = MZ_Flux1UnetLoader_cpuDynOffload
+NODE_DISPLAY_NAME_MAPPINGS[
+    "MZ_Flux1UnetLoader_cpuDynOffload"] = f"{AUTHOR_NAME} - Flux1UnetLoader_cpuDynOffload"
diff --git a/mz_fluxext_core.py b/mz_fluxext_core.py
index 7ab171a..fe510b8 100644
--- a/mz_fluxext_core.py
+++ b/mz_fluxext_core.py
@@ -99,11 +99,11 @@ def Flux1PartialLoad_Patch(args={}):
         return pre_only_single_blocks_forward_hook
 
     def pre_only_model_forward_hook(module, inp):
-        print("double_blocks to cpu")
+        # print("double_blocks to cpu")
         double_blocks_to_cpu()
-        print("single_blocks to cpu")
+        # print("single_blocks to cpu")
         single_blocks_to_cpu()
-        print("other to cuda")
+        # print("other to cuda")
         other_to_cuda()
         return inp