'update'

__init__.py

@@ -114,14 +114,13 @@ class MZ_LLamaCPPModelConfig_DownloaderSelect:
             "optional": {
             },
         }
-    
+
     RETURN_TYPES = ("LLamaCPPModelConfig",)
     RETURN_NAMES = ("llama_cpp_model_config",)
 
     FUNCTION = "create"
     CATEGORY = f"{CATEGORY_NAME}/others"
 
-
     def create(self, **kwargs):
         kwargs = kwargs.copy()
 
@@ -133,7 +132,9 @@ class MZ_LLamaCPPModelConfig_DownloaderSelect:
 
 
 NODE_CLASS_MAPPINGS["MZ_LLamaCPPModelConfig_DownloaderSelect"] = MZ_LLamaCPPModelConfig_DownloaderSelect
-NODE_DISPLAY_NAME_MAPPINGS["MZ_LLamaCPPModelConfig_DownloaderSelect"] = f"{AUTHOR_NAME} - ModelConfigDownloaderSelect(LLamaCPP)"
+NODE_DISPLAY_NAME_MAPPINGS[
+    "MZ_LLamaCPPModelConfig_DownloaderSelect"] = f"{AUTHOR_NAME} - ModelConfigDownloaderSelect(LLamaCPP)"
+
 
 class MZ_LLamaCPPCLIPTextEncode:
     @classmethod
@@ -232,7 +233,7 @@ class MZ_CustomizeInstruct:
         return {
             "required": {
                 "system": ("STRING", {"multiline": True, "default": mz_prompts.Long_prompt}),
-                "instruct": ("STRING", {"multiline": True, "default": "Short: %text%"}),
+                "instruct": ("STRING", {"multiline": True, "default": ""}),
             },
         }
 
@@ -334,7 +335,7 @@ class MZ_OpenAIApiCLIPTextEncode:
     def encode(self, **kwargs):
         kwargs = kwargs.copy()
 
-        import mz_openaiapi
+        from . import mz_openaiapi
         importlib.reload(mz_openaiapi)
 
         if kwargs.get("api_key", "").endswith("******"):
@@ -451,6 +452,50 @@ NODE_CLASS_MAPPINGS["MZ_ImageInterrogatorModelConfig_ManualSelect"] = MZ_ImageIn
 NODE_DISPLAY_NAME_MAPPINGS[
     "MZ_ImageInterrogatorModelConfig_ManualSelect"] = f"{AUTHOR_NAME} - ModelConfigManualSelect(ImageInterrogator)"
 
+
+class MZ_ImageInterrogatorModelConfig_DownloaderSelect:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "model_name": ([
+                    "ggml_llava1_5-7b-q4_k_m",
+                    "ggml_bakllava-1-q4_k_m",
+                    "llava_v1_6_mistral_7b_q5_k_m",
+                ],),
+                "mmproj_model_name": ([
+                    "auto",
+                    "ggml_llava1_5-7b-mmproj-f16",
+                    "ggml_bakllava-1-mmproj-f16",
+                    "llava_v1_6_mistral_7b_mmproj_f16",
+                ],),
+            },
+            "optional": {
+            },
+        }
+
+    RETURN_TYPES = ("ImageInterrogatorModelConfig",)
+    RETURN_NAMES = ("image_interrogator_model",)
+
+    FUNCTION = "create"
+    CATEGORY = f"{CATEGORY_NAME}/others"
+
+    def create(self, **kwargs):
+        kwargs = kwargs.copy()
+        model_name = kwargs.get("model_name")
+        mmproj_model_name = kwargs.get("mmproj_model_name", "auto")
+        return ({
+            "type": "DownloaderSelect",
+            "model_name": model_name,
+            "mmproj_model_name": mmproj_model_name,
+        },)
+
+
+NODE_CLASS_MAPPINGS["MZ_ImageInterrogatorModelConfig_DownloaderSelect"] = MZ_ImageInterrogatorModelConfig_DownloaderSelect
+NODE_DISPLAY_NAME_MAPPINGS[
+    "MZ_ImageInterrogatorModelConfig_DownloaderSelect"] = f"{AUTHOR_NAME} - ModelConfigDownloaderSelect(ImageInterrogator)"
+
+
 from . import mz_gen_translate
 mz_gen_translate.gen_translate(NODE_DISPLAY_NAME_MAPPINGS, NODE_CLASS_MAPPINGS)
 

mz_gen_translate.py

@@ -8,6 +8,8 @@ COMFY_PATH = Path(folder_paths.__file__).parent
 
 
 ZH_Replace_Map = {
+    "mmproj_model_name": "mmproj模型名称",
+    "model_name": "模型名称",
     "llama_cpp_model": "llama.cpp模型",
     "mmproj_model": "mmproj模型",
     "resolution": "分辨率",
@@ -22,6 +24,7 @@ ZH_Replace_Map = {
     "keep_device": "模型常驻显存",
     "llama_cpp_options": "llama.cpp可选配置",
     "Options": "可选配置",
+    "LLamaCPPModelConfigManualSelect": "llama.cpp模型配置手动选择",
     "LLamaCPP": "llama.cpp",
     "CLIPTextEncode": "CLIP文本编码器",
     "clip": "CLIP",
@@ -29,7 +32,6 @@ ZH_Replace_Map = {
     "customize_instruct": "自定义指令",
     "CustomizeInstruct": "自定义指令",
     "deprecated": "已废弃",
-    "LLamaCPPModelConfigManualSelect": "llama.cpp模型配置手动选择",
     "ModelConfigManualSelect": "手动模型选择器",
     "ModelConfigDownloaderSelect": "预设模型下载器",
     "captioner_config": "打标器配置",

mz_llama_core_nodes.py

@@ -129,54 +129,35 @@ def llama_cpp_node_encode(args_dict):
             mz_prompt_utils.Utils.print_log(f"question: {question}")
 
         if schema is not None:
-            response_json = mz_llama_cpp.llama_cpp_simple_interrogator_to_json(
+            response_text = mz_llama_cpp.llama_cpp_simple_interrogator_to_json(
                 model_file=model_file,
                 system=system_prompt,
                 question=question,
                 schema=schema,
                 options=options,
             )
-            mz_prompt_utils.Utils.print_log(f"response_json: {response_json}")
 
-            response = json.loads(response_json)
-            full_responses = []
+            response_json = json.loads(response_text)
+            mz_prompt_utils.Utils.print_log(
+                f"response_json: {json.dumps(response_json, indent=2)}")
 
-            if response["description"] != "":
-                full_responses.append(f"({response['description']})")
-            if response["long_prompt"] != "":
-                full_responses.append(f"({response['long_prompt']})")
-            if response["main_color_word"] != "":
-                full_responses.append(f"({response['main_color_word']})")
-            if response["camera_angle_word"] != "":
-                full_responses.append(f"({response['camera_angle_word']})")
+            responses = []
+            for key, value in response_json.items():
+                if type(value) == list:
+                    # 去除开头.和空格
+                    value = [v.strip().lstrip(".") for v in value]
+                    # 去除空字符串
+                    value = [v for v in value if v != ""]
+                    if len(value) > 0:
+                        responses.append(f"({', '.join(value)})")
 
-            response["style_words"] = [
-                x for x in response["style_words"] if x != ""]
-            if len(response["style_words"]) > 0:
-                full_responses.append(
-                    f"({', '.join(response['style_words'])})")
+                else:
+                    if value != "":
+                        responses.append(f"({value})")
 
-            response["subject_words"] = [
-                x for x in response["subject_words"] if x != ""]
-            if len(response["subject_words"]) > 0:
-                full_responses.append(
-                    f"({', '.join(response['subject_words'])})")
-
-            response["light_words"] = [
-                x for x in response["light_words"] if x != ""]
-            if len(response["light_words"]) > 0:
-                full_responses.append(
-                    f"({', '.join(response['light_words'])})")
-
-            response["environment_words"] = [
-                x for x in response["environment_words"] if x != ""]
-            if len(response["environment_words"]) > 0:
-                full_responses.append(
-                    f"({', '.join(response['environment_words'])})")
-
-            full_response = ", ".join(full_responses)
+            response = ", ".join(responses)
         else:
-            full_response = mz_llama_cpp.llama_cpp_simple_interrogator(
+            response = mz_llama_cpp.llama_cpp_simple_interrogator(
                 model_file=model_file,
                 system=system_prompt,
                 question=question,
@@ -184,42 +165,42 @@ def llama_cpp_node_encode(args_dict):
             )
 
             start_str = customize_instruct.get("start_str", "")
-            if start_str != "" and full_response.find(start_str) != -1:
-                full_response_list = full_response.split(start_str)
+            if start_str != "" and response.find(start_str) != -1:
+                full_response_list = response.split(start_str)
                 # 删除第一个元素
                 full_response_list.pop(0)
-                full_response = start_str.join(full_response_list)
+                response = start_str.join(full_response_list)
 
             end_str = customize_instruct.get("end_str", "")
-            if end_str != "" and full_response.find(end_str) != -1:
-                full_response_list = full_response.split(end_str)
+            if end_str != "" and response.find(end_str) != -1:
+                full_response_list = response.split(end_str)
                 # 删除最后一个元素
                 full_response_list.pop()
-                full_response = end_str.join(full_response_list)
+                response = end_str.join(full_response_list)
 
         if keep_device is False:
             mz_llama_cpp.freed_gpu_memory(model_file=model_file)
 
         # 去除换行
-        while full_response.find("\n") != -1:
-            full_response = full_response.replace("\n", " ")
+        while response.find("\n") != -1:
+            response = response.replace("\n", " ")
 
         # 句号换成逗号
-        while full_response.find(".") != -1:
-            full_response = full_response.replace(".", ",")
+        while response.find(".") != -1:
+            response = response.replace(".", ",")
 
         # 去除多余逗号
-        while full_response.find(",,") != -1:
-            full_response = full_response.replace(",,", ",")
-        while full_response.find(", ,") != -1:
-            full_response = full_response.replace(", ,", ",")
+        while response.find(",,") != -1:
+            response = response.replace(",,", ",")
+        while response.find(", ,") != -1:
+            response = response.replace(", ,", ",")
 
-        full_response = mz_prompt_utils.Utils.prompt_zh_to_en(full_response)
+        response = mz_prompt_utils.Utils.prompt_zh_to_en(response)
 
         style_presets_prompt_text = style_presets_prompt.get(style_presets, "")
 
         if style_presets_prompt_text != "":
-            full_response = f"{style_presets_prompt_text}, {full_response}"
+            response = f"{style_presets_prompt_text}, {response}"
 
     except Exception as e:
         mz_llama_cpp.freed_gpu_memory(model_file=model_file)
@@ -229,9 +210,9 @@ def llama_cpp_node_encode(args_dict):
     clip = args_dict.get("clip", None)
     if clip is not None:
         conditionings = mz_prompt_utils.Utils.a1111_clip_text_encode(
-            clip, full_response, )
+            clip, response, )
 
-    return {"ui": {"string": [full_response,]}, "result": (full_response, conditionings)}
+    return {"ui": {"string": [mz_prompt_utils.Utils.to_debug_prompt(response),]}, "result": (response, conditionings)}
 
 
 def image_interrogator_captioner(args_dict):
@@ -331,16 +312,15 @@ def image_interrogator_captioner(args_dict):
     return result
 
 
-
 image_interrogator_auto_mmproj_model = {
-    
+    "7ac9c2f7b8d76cc7f3118cdf0953ebab7a7a9b12bad5dbe237219d2ab61765ea": "ggml_llava1_5-7b-mmproj-f16",
+    "c93de1376be9b6977cc94d252a3d165d6059e07b528de0fa762534d9599b27d6": "ggml_bakllava-1-mmproj-f16",
+    "b1d37fc65ecb80aa8f1ce185bf4d7605bc3c5cc5bcc77a160c3a1b0377631112": "llava_v1_6_mistral_7b_q5_k_m",
 }
 
-def image_interrogator_node_encode(args_dict):
-    image_interrogator_model = args_dict.get("image_interrogator_model", {})
 
-    llama_cpp_model = image_interrogator_model.get("llama_cpp_model", "auto")
-    mmproj_model = image_interrogator_model.get("mmproj_model", "auto")
+def image_interrogator_node_encode(args_dict):
+    importlib.reload(mz_prompts)
 
     captioner_config = args_dict.get("captioner_config", None)
     if captioner_config is not None:
@@ -348,7 +328,56 @@ def image_interrogator_node_encode(args_dict):
         raise Exception(
             "图片批量反推任务已完成 ; Image batch reverse push task completed")
 
+    model_config = args_dict.get("image_interrogator_model", {})
+    llama_cpp_model = model_config.get("llama_cpp_model", "auto")
+    mmproj_model = model_config.get("mmproj_model", "auto")
+
+    select_model_type = model_config.get("type", "ManualSelect")
+    if select_model_type == "ManualSelect":
+        llama_cpp_model = model_config.get("model_path", "auto")
+        if llama_cpp_model == "auto":
+            llama_cpp_model = mz_prompt_utils.Utils.get_auto_model_fullpath(
+                "ggml_llava1_5-7b-q4_k_m")
+
+        if mmproj_model == "auto":
+            llama_cpp_model_sha256 = mz_prompt_utils.Utils.file_sha256(
+                llama_cpp_model)
+            mmproj_model_name = image_interrogator_auto_mmproj_model.get(
+                llama_cpp_model_sha256, None)
+            if mmproj_model_name is None:
+                mz_prompt_utils.Utils.print_log(
+                    "llama_cpp_model_sha256: ", llama_cpp_model_sha256)
+                raise Exception(
+                    "未能自动找到对应的mmproj文件 ; Failed to automatically find the corresponding mmproj file.")
+            mmproj_model = mz_prompt_utils.Utils.get_auto_model_fullpath(
+                mmproj_model_name)
+
+    elif select_model_type == "DownloaderSelect":
+        model_name = model_config.get("model_name")
+        llama_cpp_model = mz_prompt_utils.Utils.get_auto_model_fullpath(
+            model_name)
+
+        mmproj_model = model_config.get("mmproj_model_name", "auto")
+        if mmproj_model == "auto":
+            llama_cpp_model_sha256 = mz_prompt_utils.Utils.file_sha256(
+                llama_cpp_model)
+
+            mz_prompt_utils.Utils.print_log(
+                "llama_cpp_model_sha256: ", llama_cpp_model_sha256)
+
+            mmproj_model_name = image_interrogator_auto_mmproj_model.get(
+                llama_cpp_model_sha256, None)
+            if mmproj_model_name is None:
+                raise Exception(
+                    "未能自动找到对应的mmproj文件 ; Failed to automatically find the corresponding mmproj file")
+            mmproj_model = mz_prompt_utils.Utils.get_auto_model_fullpath(
+                mmproj_model_name)
+    else:
+        raise Exception("Unknown select_model_type")
+
     image = args_dict.get("image", None)
+    image = mz_prompt_utils.Utils.tensor2pil(image)
+
     resolution = args_dict.get("resolution", 512)
     keep_device = args_dict.get("keep_device", False)
     seed = args_dict.get("seed", -1)
@@ -359,23 +388,27 @@ def image_interrogator_node_encode(args_dict):
 
     customize_instruct = args_dict.get("customize_instruct", None)
     if customize_instruct is None:
-        system_text = mz_prompts.GPT4VImageCaptioner_System
-        question_text = mz_prompts.GPT4VImageCaptioner_Prompt
+        system_prompt = mz_prompts.M_ImageCaptioner_System
+        question = mz_prompts.M_ImageCaptioner_Prompt
     else:
-
         system_prompt = customize_instruct.get("system", "")
-        question = customize_instruct.get("instruct", "%text%")
-        system_prompt = system_prompt.replace("%text%", "")
-        question = question.replace("%text%", "")
+        question = customize_instruct.get("instruct", "")
 
     response = mz_llama_cpp.llava_cpp_simple_interrogator(
         model_file=llama_cpp_model,
         mmproj_file=mmproj_model,
         image=image,
         options=options,
-        system=system_text,
-        question=question_text,
+        system=system_prompt,
+        question=question,
     )
+    if response is not None:
+        response = response.strip()
+    if customize_instruct is None:
+        # 去除开头的In the image,
+        start_text = "In the image, "
+        if response.startswith(start_text):
+            response = response[len(start_text):]
 
     if keep_device is False:
         mz_llama_cpp.freed_gpu_memory(model_file=llama_cpp_model)
@@ -388,4 +421,4 @@ def image_interrogator_node_encode(args_dict):
         conditionings = mz_prompt_utils.Utils.a1111_clip_text_encode(
             clip, response, )
 
-    return {"ui": {"string": [response,]}, "result": (response, conditionings)}
+    return {"ui": {"string": [mz_prompt_utils.Utils.to_debug_prompt(response),]}, "result": (response, conditionings)}

mz_llama_cpp.py

@@ -13,9 +13,13 @@ except ImportError:
 
 
 def check_llama_cpp_requirements():
-    last_version = "0.2.63"
+    min_version = "0.2.63"
+    last_version = "0.2.76"
     try:
         from llama_cpp import Llama
+        import llama_cpp
+        if llama_cpp.__version__ < min_version:
+            raise ImportError("llama_cpp version is too low. (llama_cpp版本过低)")
     except ImportError:
         py_version = ""
         if sys.version_info.major == 3:
@@ -65,9 +69,39 @@ def check_llama_cpp_requirements():
             wheel_url = f"https://github.com/abetlen/llama-cpp-python/releases/download/v{last_version}-{cuda_version}/{wheel_name}"
 
         print(f"pip install {wheel_url}")
-        ret = subprocess.run([
-            sys.executable, "-m",
-            "pip", "install", wheel_url], check=True)
+        modelscope_url = f"https://www.modelscope.cn/api/v1/models/wailovet/MinusZoneAIModels/repo?Revision=master&FilePath=llama-cpp-python-win%2F{cuda_version}%2F{wheel_name}"
+        if mz_prompt_utils.Utils.testDownloadSpeed(wheel_url):
+            ret = subprocess.run([
+                sys.executable, "-m",
+                "pip", "install", wheel_url], check=True)
+        elif mz_prompt_utils.Utils.testDownloadSpeed(modelscope_url):
+            import tempfile
+            whl_download_file = os.path.join(
+                tempfile.gettempdir(), wheel_name)
+            mz_prompt_utils.Utils.download_file(
+                modelscope_url, whl_download_file)
+            print(f"pip install {whl_download_file}")
+            ret = subprocess.run([
+                sys.executable, "-m",
+                "pip", "install", whl_download_file], check=True)
+        else:
+
+            # 兜底方案
+            modelscope_url = f"https://www.modelscope.cn/api/v1/models/wailovet/MinusZoneAIModels/repo?Revision=master&FilePath=llama-cpp-python-win%2Fcu121%2Fllama_cpp_python-0.2.76-cp310-cp310-win_amd64.whl"
+            if py_version == "310" and system_name == "win_amd64" and mz_prompt_utils.Utils.testDownloadSpeed(modelscope_url):
+                import tempfile
+                whl_download_file = os.path.join(
+                    tempfile.gettempdir(), wheel_name)
+                mz_prompt_utils.Utils.download_file(
+                    modelscope_url, whl_download_file)
+                print(f"pip install {whl_download_file}")
+                ret = subprocess.run([
+                    sys.executable, "-m",
+                    "pip", "install", whl_download_file], check=True)
+            else:
+                ret = subprocess.run([
+                    sys.executable, "-m",
+                    "pip", "install", wheel_url], check=True)
 
         if ret.returncode != 0:
             raise ValueError("Failed to install llama_cpp. (安装llama_cpp失败)")
@@ -357,7 +391,8 @@ def llava_cpp_simple_interrogator(
     check_llama_cpp_requirements()
     from llama_cpp.llama_chat_format import Llava15ChatHandler
     if mmproj_file is not None:
-        mz_prompt_utils.Utils.print_log(f"llava_cpp_simple_interrogator mmproj_file: {mmproj_file}")
+        mz_prompt_utils.Utils.print_log(
+            f"llava_cpp_simple_interrogator mmproj_file: {mmproj_file}")
         chat_handler = Llava15ChatHandler(clip_model_path=mmproj_file)
 
     return llava_cpp_messages(model_file, chat_handler, [

mz_openaiapi.py

@@ -5,6 +5,7 @@ import subprocess
 
 from . import mz_prompt_utils
 from . import mz_llama_cpp
+from . import mz_llama_core_nodes
 from . import mz_prompts
 
 
@@ -63,16 +64,16 @@ def query_beautify_prompt_text(args_dict):
 
     schema = None
     if customize_instruct is None:
-        schema = mz_llama_cpp.get_schema_obj(
+        schema = mz_llama_core_nodes.get_schema_obj(
             keys_type={
-                "description": mz_llama_cpp.get_schema_base_type("string"),
-                "long_prompt": mz_llama_cpp.get_schema_base_type("string"),
-                "main_color_word": mz_llama_cpp.get_schema_base_type("string"),
-                "camera_angle_word": mz_llama_cpp.get_schema_base_type("string"),
-                "style_words": mz_llama_cpp.get_schema_array("string"),
-                "subject_words": mz_llama_cpp.get_schema_array("string"),
-                "light_words": mz_llama_cpp.get_schema_array("string"),
-                "environment_words": mz_llama_cpp.get_schema_array("string"),
+                "description": mz_llama_core_nodes.get_schema_base_type("string"),
+                "long_prompt": mz_llama_core_nodes.get_schema_base_type("string"),
+                "main_color_word": mz_llama_core_nodes.get_schema_base_type("string"),
+                "camera_angle_word": mz_llama_core_nodes.get_schema_base_type("string"),
+                "style_words": mz_llama_core_nodes.get_schema_array("string"),
+                "subject_words": mz_llama_core_nodes.get_schema_array("string"),
+                "light_words": mz_llama_core_nodes.get_schema_array("string"),
+                "environment_words": mz_llama_core_nodes.get_schema_array("string"),
             },
             required=[
                 "description",
@@ -184,7 +185,7 @@ def query_beautify_prompt_text(args_dict):
         full_response = full_response.replace(",,", ",")
     while full_response.find(", ,") != -1:
         full_response = full_response.replace(", ,", ",")
-    style_presets_prompt_text = mz_llama_cpp.style_presets_prompt.get(
+    style_presets_prompt_text = mz_llama_core_nodes.style_presets_prompt.get(
         style_presets, "")
     if style_presets_prompt_text != "":
         full_response = f"{style_presets_prompt_text}, {full_response}"

mz_prompt_utils.py

@@ -675,6 +675,12 @@ class Utils:
         return Utils.file_hash(file_path, hashlib.sha256)
 
     def get_auto_model_fullpath(model_name):
+        fullpath = Utils.cache_get(f"get_auto_model_fullpath_{model_name}")
+        Utils.print_log(f"get_auto_model_fullpath_{model_name} => {fullpath}")
+        if fullpath is not None:
+            if os.path.exists(fullpath):
+                return fullpath
+
         find_paths = []
         target_sha256 = ""
         file_path = ""
@@ -703,12 +709,31 @@ class Utils:
                 for root, dirs, files in os.walk(find_fullpath):
                     for file in files:
                         if target_sha256 == Utils.file_sha256(os.path.join(root, file)):
+                            Utils.cache_set(f"get_auto_model_fullpath_{model_name}", os.path.join(root, file))
                             return os.path.join(root, file)
                         else:
                             Utils.print_log(
                                 f"Model {os.path.join(root, file)} file hash not match, {target_sha256} != {Utils.file_sha256(os.path.join(root, file))}")
 
-        return Utils.download_model({"url": download_url, "output": file_path})
+        result = Utils.download_model({"url": download_url, "output": file_path})
+        Utils.cache_set(f"get_auto_model_fullpath_{model_name}", result)
+        return result
+
+
+    def testDownloadSpeed(url):
+        try:
+            print(f"Testing download speed for {url}")
+            start = time.time()
+            # 下载2M数据
+            headers = {"Range": "bytes=0-2097151"}
+            _ = requests.get(url, headers=headers, timeout=5)
+            end = time.time()
+            print(
+                f"Download speed: {round(5.00 / (float(end) - float(start)) / 1024, 2)} KB/s")
+            return float(end) - float(start) < 4
+        except Exception as e:
+            print(f"Test download speed failed: {e}")
+            return False
 
 
 MODEL_ZOO = [

mz_prompts.py

@@ -26,7 +26,7 @@ As an AI image tagging expert, please provide precise tags for these images to e
 
 # 来自https://github.com/jiayev/GPT4V-Image-Captioner
 GPT4VImageCaptioner_Prompt = """
-Employ succinct keywords or phrases, steering clear of elaborate sentences and extraneous conjunctions. 
+Employ succinct keywords or phrases or sentence, steering clear of elaborate sentences and extraneous conjunctions. 
 Prioritize the tags by relevance. 
 Your tags should capture key elements such as the main subject, setting, artistic style, composition, image quality, color tone, filter, and camera specifications, and any other tags crucial for the image. 
 When tagging photos of people, include specific details like gender, nationality, attire, actions, pose, expressions, accessories, makeup, composition type, age, etc. 
@@ -35,4 +35,15 @@ Recognize and tag any celebrities, well-known landmark or IPs if clearly feature
 Your tags should be accurate, non-duplicative, and within a 20-75 word count range. 
 These tags will use for image re-creation, so the closer the resemblance to the original image, the better the tag quality. 
 Tags should be comma-separated.
-"""
+"""
+
+
+M_ImageCaptioner_System = """
+Long prompt version should consist of 3 to 5 sentences. Long prompt version must sepcify the color, shape, texture or spatial relation of the included objects. DO NOT generate sentences that describe any atmosphere!!!  
+"""
+
+M_ImageCaptioner_Prompt = """
+Describe this image in detail please.
+The language of reply is English only!!!
+Starts with "In the image," 
+"""