diff --git a/__init__.py b/__init__.py
index 7487ad8..fa90f03 100644
--- a/__init__.py
+++ b/__init__.py
@@ -50,6 +50,43 @@ llama3_models = [
     "Meta-Llama-3-8B-Instruct.Q8_0.gguf",
 ]
 
+class MZ_LLamaCPPOptions:
+    @classmethod
+    def INPUT_TYPES(s):
+        opt = mz_llama_cpp.LlamaCppOptions()
+        value = opt.value
+        result = {}
+
+        for key in value:
+            if type(value[key]) == bool:
+                result[key] = ([True, False], {"default": value[key]})
+            elif type(value[key]) == int:
+                result[key] = ("INT", {"default": value[key]})
+            elif type(value[key]) == float:
+                result[key] = ("FLOAT", {"default": value[key]})
+            elif type(value[key]) == str:
+                result[key] = ("STRING", {"default": value[key]})
+            else:
+                raise Exception(f"Unknown type: {type(value[key])}")
+        
+        return {
+            "required": result,
+        }
+    
+
+    RETURN_TYPES = ("LLamaCPPOptions",)
+    FUNCTION = "create"
+    CATEGORY = CATEGORY_NAME
+    def create(self, **kwargs):
+        importlib.reload(mz_llama_cpp)
+        opt = mz_llama_cpp.LlamaCppOptions()
+        for key in kwargs:
+            opt.value[key] = kwargs[key]
+        return (opt,)
+
+NODE_CLASS_MAPPINGS["MZ_LLamaCPPOptions"] = MZ_LLamaCPPOptions
+NODE_DISPLAY_NAME_MAPPINGS["MZ_LLamaCPPOptions"] = f"{AUTHOR_NAME} - LLamaCPPOptions"
+
 class MZ_LLama3CLIPTextEncode:
     @classmethod
     def INPUT_TYPES(s):
@@ -64,7 +101,7 @@ class MZ_LLama3CLIPTextEncode:
 
         return {
             "required": {
-                "llama3_model": (m_llama3_models, {"default": m_llama3_models[0]}),
+                "llama_cpp_model": (m_llama3_models, {"default": m_llama3_models[0]}),
                 "download_source": (
                     ["none", "modelscope", "hf-mirror.com",],
                     {"default": "none"}
@@ -78,17 +115,18 @@ class MZ_LLama3CLIPTextEncode:
             },
             "optional": {
                 "clip": ("CLIP", ),
+                "options": ("LLamaCPPOptions", ),
             },
         }
     RETURN_TYPES = ("STRING", "CONDITIONING",)
     FUNCTION = "encode"
     CATEGORY = CATEGORY_NAME
-    def encode(self, text, llama3_model, style_presets, clip=None, download_source=None, seed=0, n_gpu_layers=40):
+    def encode(self, text, llama_cpp_model, style_presets, clip=None, download_source=None, seed=0, n_gpu_layers=40):
         importlib.reload(mz_llama3)
 
-        llama3_model = llama3_model.replace("[downloaded]", "")
+        llama_cpp_model = llama_cpp_model.replace("[downloaded]", "")
 
-        text = mz_llama3.query_beautify_prompt_text(llama3_model, n_gpu_layers, text, style_presets, download_source,) 
+        text = mz_llama3.query_beautify_prompt_text(llama_cpp_model, n_gpu_layers, text, style_presets, download_source,) 
         conditionings = None
         if clip is not None:
             tokens = clip.tokenize(text)
@@ -132,7 +170,7 @@ class MZ_LLavaImageInterrogator:
 
 
         return {"required": {
-            "llava_model": (m_llava_models, {"default": m_llava_models[0]}),
+            "llama_cpp_model": (m_llava_models, {"default": m_llava_models[0]}),
             "mmproj_model": (m_llava_mmproj_models, {"default": m_llava_mmproj_models[0]}),
             "download_source": (
                 [
@@ -149,15 +187,16 @@ class MZ_LLavaImageInterrogator:
             "n_gpu_layers": ("INT", {"default": 40, "min": -1, "max": 0xffffffffffffffff}),
         },
         "optional": {
-            "clip": ("CLIP",),
+            "clip": ("CLIP", ),
+            "options": ("LLamaCPPOptions", ),
         }}
     RETURN_TYPES = ("STRING", "CONDITIONING",)
     FUNCTION = "interrogate"
     CATEGORY = CATEGORY_NAME
-    def interrogate(self, llava_model, mmproj_model, image, resolution, prefix, download_source=None, seed=0, clip=None, n_gpu_layers=40):
+    def interrogate(self, llama_cpp_model, mmproj_model, image, resolution, prefix, download_source=None, seed=0, clip=None, n_gpu_layers=40, options={}):
         importlib.reload(mz_llava)
 
-        llava_model = llava_model.replace("[downloaded]", "")
+        llama_cpp_model = llama_cpp_model.replace("[downloaded]", "")
         mmproj_model = mmproj_model.replace("[downloaded]", "")
 
         prefix = Utils.prompt_zh_to_en(prefix)
@@ -166,12 +205,13 @@ class MZ_LLavaImageInterrogator:
 
         image_pil = Utils.tensor2pil(image)
         response = mz_llava.image_interrogator(
-            llava_model,
+            llama_cpp_model,
             mmproj_model,
             n_gpu_layers,
             image_pil, 
             resolution, 
             download_source,
+            options,
         ) 
         conditionings = None
         if clip is not None:
diff --git a/mz_llama3.py b/mz_llama3.py
index fe52308..f2a61d6 100644
--- a/mz_llama3.py
+++ b/mz_llama3.py
@@ -36,7 +36,9 @@ def get_style_presets():
         "illustration",  
     ]
 
-def query_beautify_prompt_text(model_name, n_gpu_layers, text, style_presets, download_source=None):     
+def query_beautify_prompt_text(model_name, n_gpu_layers, text, style_presets, download_source=None, options={}):     
+    if options is None:
+        options = {}
     import mz_prompts
     importlib.reload(mz_prompts)
     importlib.reload(mz_llama_cpp)
@@ -81,14 +83,17 @@ def query_beautify_prompt_text(model_name, n_gpu_layers, text, style_presets, do
             ]
         )
 
+        options["max_tokens"] = options.get("max_tokens", 2048)
+        options["temperature"] = options.get("temperature", 1.6)
+
+
         response_json = mz_llama_cpp.llama_cpp_simple_interrogator_to_json(
             model_file=model_file,
             n_gpu_layers=n_gpu_layers,
             system=mz_prompts.Beautify_Prompt,
             question=f"IDEA: {style_presets},{text}",
             schema=schema,
-            temperature=1.6,
-            max_tokens=2048,
+            options=options,
         ) 
         mz_utils.Utils.print_log(f"response_json: {response_json}")
         response = json.loads(response_json)
diff --git a/mz_llama_cpp.py b/mz_llama_cpp.py
index ae6e841..b94c878 100644
--- a/mz_llama_cpp.py
+++ b/mz_llama_cpp.py
@@ -8,6 +8,13 @@ try:
 except ImportError:
     pass
   
+
+class LlamaCppOptions:
+    value = {
+        "max_tokens": 2048,
+        "temperature": 0.8,
+    }
+
 def check_llama_cpp_requirements(): 
     last_version = "0.2.63"
     try:
@@ -84,6 +91,8 @@ def freed_gpu_memory(model_file):
     mz_utils.Utils.cache_set(f"llama_cpp_model_and_opt_{model_file}", None)
 
 def llama_cpp_messages(model_file, n_gpu_layers, chat_handler=None, messages=[], options={}):
+    if options is None:
+        options = {}
     print(f"Find local model file: {model_file}")
     init_opts = ["n_ctx", "logits_all", "chat_format",]
 
@@ -176,7 +185,7 @@ def get_schema_array(item_type="string"):
 
 
 
-def llama_cpp_simple_interrogator_to_json(model_file, n_gpu_layers, use_system=True, system=None, question="", schema={}, max_tokens=1024, temperature=0.8):
+def llama_cpp_simple_interrogator_to_json(model_file, n_gpu_layers, use_system=True, system=None, question="", schema={}, options={}):
     if system is None:
         system = ""
         messages = [
@@ -218,14 +227,18 @@ def llama_cpp_simple_interrogator_to_json(model_file, n_gpu_layers, use_system=T
     } 
 
 
-    return llama_cpp_messages(model_file, n_gpu_layers, None, messages, options={
-        "response_format": response_format,
-        "chat_format":"chatml",
-        "temperature": temperature,
-        "max_tokens": max_tokens,  
-    })
+    options["response_format"] = response_format
+    options["chat_format"] = "chatml"
+    if options.get("temperature", None) is None:
+        options["temperature"] = 0.8
+    if options.get("max_tokens", None) is None:
+        options["max_tokens"] = 2048
+
+    return llama_cpp_messages(model_file, n_gpu_layers, None, messages, options=options)
     
-def llama_cpp_simple_interrogator(model_file, n_gpu_layers, use_system=True, system=None, question=""):
+def llama_cpp_simple_interrogator(model_file, n_gpu_layers, use_system=True, system=None, question="", options={}):
+    if options is None:
+        options = {}
     if system is None:
         system = ""
         messages = [
@@ -260,17 +273,25 @@ def llama_cpp_simple_interrogator(model_file, n_gpu_layers, use_system=True, sys
                 "content": question
             },
         ]
-    return llama_cpp_messages(model_file, n_gpu_layers, None, messages)
+    return llama_cpp_messages(model_file, n_gpu_layers, None, messages, options={})
 
 
 def llava_cpp_messages(model_file, n_gpu_layers, chat_handler, messages, options={}):
+    if options is None:
+        options = {}
     options["logits_all"] = True
     options["n_ctx"] = max(2048, options.get("n_ctx", 2048)) 
     return llama_cpp_messages(model_file, n_gpu_layers, chat_handler, messages, options)
 
  
 
-def llava_cpp_simple_interrogator(model_file, mmproj_file, n_gpu_layers, system="You are an assistant who perfectly describes images.", question="Describe this image in detail please.", image=None):
+def llava_cpp_simple_interrogator(
+        model_file, mmproj_file, n_gpu_layers, system="You are an assistant who perfectly describes images.", question="Describe this image in detail please.", 
+        image=None, options={}):
+    if options is None:
+        options = {}
+
+
     check_llama_cpp_requirements()
 
     content = []
@@ -294,7 +315,7 @@ def llava_cpp_simple_interrogator(model_file, mmproj_file, n_gpu_layers, system=
             "role": "user",
             "content": content,
         }, 
-    ])
+    ], options=options)
 
 
 
diff --git a/mz_llava.py b/mz_llava.py
index 64071af..8487728 100644
--- a/mz_llava.py
+++ b/mz_llava.py
@@ -45,7 +45,9 @@ def get_exist_model(model_name):
     return None
 
 
-def image_interrogator(model_name, mmproj_name, n_gpu_layers, image, resolution, download_source=None):  
+def image_interrogator(model_name, mmproj_name, n_gpu_layers, image, resolution, download_source=None, options={}):  
+    if options is None:
+        options = {}
     image = mz_utils.Utils.resize_max(image, resolution, resolution)
     model_file = get_exist_model(model_name)
     mmproj_file = get_exist_model(mmproj_name)
@@ -95,6 +97,7 @@ def image_interrogator(model_name, mmproj_name, n_gpu_layers, image, resolution,
         mmproj_file=mmproj_file,
         n_gpu_layers=n_gpu_layers,
         image=image,
+        options=options,
     )