nodes/MinusZoneAI-ComfyUI-Prompt-MZ
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

'update'

Browse Source
This commit is contained in:
wailovet
2024-06-01 03:42:52 +08:00
parent 288474f7b6
commit 9bb6e8919d
14 changed files with 708 additions and 472 deletions
Download Patch File Download Diff File Expand all files Collapse all files
__init__.py
+73 -169
View File
@@ -8,6 +8,7 @@ from nodes import MAX_RESOLUTION
import comfy.utils
import shutil
import comfy.samplers
import folder_paths
WEB_DIRECTORY = "./web"
@@ -15,12 +16,12 @@ WEB_DIRECTORY = "./web"
AUTHOR_NAME = u"MinusZone"
CATEGORY_NAME = f"{AUTHOR_NAME} - Prompt"
sys.path.append(os.path.join(os.path.dirname(__file__)))
# sys.path.append(os.path.join(os.path.dirname(__file__)))
import importlib
import mz_prompt_webserver
mz_prompt_webserver.start_server()
# import mz_prompt_webserver
# mz_prompt_webserver.start_server()
NODE_CLASS_MAPPINGS = {
}
@@ -29,11 +30,64 @@ NODE_CLASS_MAPPINGS = {
NODE_DISPLAY_NAME_MAPPINGS = {
}
from . import mz_llama_cpp
from . import mz_llava
import mz_llama3
import mz_phi3
import mz_llama_cpp
import mz_llava
class MZ_LLamaCPPModelConfig_ManualSelect:
@classmethod
def INPUT_TYPES(s):
gguf_files = folder_paths.get_filename_list("gguf")
return {
"required": {
"llama_cpp_model": (gguf_files,),
},
"optional": {
},
}
RETURN_TYPES = ("LLamaCPPModelConfig",)
RETURN_NAMES = ("llama_cpp_model_config",)
FUNCTION = "create"
CATEGORY = CATEGORY_NAME
def create(self, **kwargs):
return (kwargs,)
class MZ_LLamaCPPCLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
importlib.reload(mz_llama_cpp)
result = {
"required": {
},
"optional": {
"llama_cpp_model": ("LLamaCPPModelConfig",),
},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
from . import mz_llama_core_nodes
importlib.reload(mz_llama_core_nodes)
return mz_llama_core_nodes.llama_cpp_node_encode(kwargs)
class MZ_LLamaCPPOptions:
@@ -83,13 +137,12 @@ NODE_DISPLAY_NAME_MAPPINGS["MZ_LLamaCPPOptions"] = f"{AUTHOR_NAME} - LLamaCPPOpt
class MZ_CustomizeInstruct:
@classmethod
def INPUT_TYPES(s):
import mz_prompts
from . import mz_prompts
return {
"required": {
"system": ("STRING", {"multiline": True, "default": mz_prompts.Long_prompt}),
"instruct": ("STRING", {"multiline": True, "default": "Short: %text%"}),
"start_str": ("STRING", {"default": "Long: "}),
"end_str": ("STRING", {"default": ""}),
},
}
@@ -107,7 +160,8 @@ NODE_DISPLAY_NAME_MAPPINGS["MZ_CustomizeInstruct"] = f"{AUTHOR_NAME} - Customize
def getCommonCLIPTextEncodeInput():
style_presets = mz_llama_cpp.get_style_presets()
from . import mz_llama_core_nodes
style_presets = mz_llama_core_nodes.get_style_presets()
CommonCLIPTextEncodeInput = {
"required": {
"prompt_version": (["v1"], {"default": "v1"}),
@@ -129,162 +183,6 @@ def getCommonCLIPTextEncodeInput():
return CommonCLIPTextEncodeInput
class MZ_LLama3CLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
m_models = mz_llama3.llama3_models.copy()
for i in range(len(m_models)):
if mz_llama3.get_exist_model(m_models[i]) is not None:
m_models[i] += "[downloaded]"
importlib.reload(mz_llama_cpp)
result = {
"required": {
"llama_cpp_model": (m_models, {"default": m_models[0]}),
"download_source": (
["none", "modelscope", "hf-mirror.com",],
{"default": "none"}
),
},
"optional": {},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
importlib.reload(mz_llama3)
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_llama3.query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_LLama3CLIPTextEncode"] = MZ_LLama3CLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_LLama3CLIPTextEncode"] = f"{AUTHOR_NAME} - CLIPTextEncode(LLama3)"
class MZ_Phi3CLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
m_models = mz_phi3.phi3_models.copy()
for i in range(len(m_models)):
if mz_llama3.get_exist_model(m_models[i]) is not None:
m_models[i] += "[downloaded]"
importlib.reload(mz_phi3)
result = {
"required": {
"llama_cpp_model": (m_models, {"default": m_models[0]}),
"download_source": (
["none", "modelscope", "hf-mirror.com",],
{"default": "none"}
),
},
"optional": {},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
importlib.reload(mz_llama3)
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_phi3.query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_Phi3CLIPTextEncode"] = MZ_Phi3CLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_Phi3CLIPTextEncode"] = f"{AUTHOR_NAME} - CLIPTextEncode(Phi3)"
class MZ_BaseLLamaCPPCLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
importlib.reload(mz_llama_cpp)
result = {
"required": {
"llama_cpp_model": ("STRING", {"default": "", "placeholder": "model_path"}),
},
"optional": {
},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
importlib.reload(mz_llama3)
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_llama_cpp.base_query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_BaseLLamaCPPCLIPTextEncode"] = MZ_BaseLLamaCPPCLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_BaseLLamaCPPCLIPTextEncode"] = f"{AUTHOR_NAME} - CLIPTextEncode(BaseLLamaCPP)"
class MZ_LLavaImageInterrogator:
@classmethod
def INPUT_TYPES(s):
@@ -518,6 +416,12 @@ NODE_CLASS_MAPPINGS["MZ_OpenAIApiCLIPTextEncode"] = MZ_OpenAIApiCLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_OpenAIApiCLIPTextEncode"] = f"{AUTHOR_NAME} - CLIPTextEncode(OpenAIApi)"
import mz_gen_translate
from . import mz_gen_translate
mz_gen_translate.gen_translate(NODE_DISPLAY_NAME_MAPPINGS, NODE_CLASS_MAPPINGS)
from .v1.init import NODE_CLASS_MAPPINGS as DEPRECATED_NODE_CLASS_MAPPINGS
from .v1.init import NODE_DISPLAY_NAME_MAPPINGS as DEPRECATED_NODE_DISPLAY_NAME_MAPPINGS
NODE_CLASS_MAPPINGS.update(DEPRECATED_NODE_CLASS_MAPPINGS)
NODE_DISPLAY_NAME_MAPPINGS.update(DEPRECATED_NODE_DISPLAY_NAME_MAPPINGS)
mz_gen_translate.py
+1
View File
@@ -26,6 +26,7 @@ ZH_Replace_Map = {
"ImageInterrogator": "图像反推",
"customize_instruct": "自定义指令",
"CustomizeInstruct": "自定义指令",
"deprecated": "已废弃",
}
mz_llama_core_nodes.py
+216
View File
@@ -0,0 +1,216 @@
import importlib
import json
from . import mz_prompt_utils
from . import mz_llama_cpp
from . import mz_prompts
def get_schema_base_type(t):
return {
"type": t,
}
def get_schema_obj(keys_type={}, required=[]):
item = {}
for key, value in keys_type.items():
if type(value) == str:
value = get_schema_base_type(value)
item[key] = value
return {
"type": "object",
"properties": item,
"required": required
}
def get_schema_array(item_type="string"):
if type(item_type) == str:
item_type = get_schema_base_type(item_type)
return {
"type": "array",
"items": item_type,
}
high_quality_prompt = "((high quality:1.4), (best quality:1.4), (masterpiece:1.4), (8K resolution), (2k wallpaper))"
style_presets_prompt = {
"none": "",
"high_quality": high_quality_prompt,
"photography": f"{high_quality_prompt}, (RAW photo, best quality), (realistic, photo-realistic:1.2), (bokeh, cinematic shot, dynamic composition, incredibly detailed, sharpen, details, intricate detail, professional lighting, film lighting, 35mm, anamorphic, lightroom, cinematography, bokeh, lens flare, film grain, HDR10, 8K)",
"illustration": f"{high_quality_prompt}, ((detailed matte painting, intricate detail, splash screen, complementary colors), (detailed),(intricate details),illustration,an extremely delicate and beautiful,ultra-detailed,highres,extremely detailed)",
}
def get_style_presets():
return [
"none",
"high_quality",
"photography",
"illustration",
]
def llama_cpp_node_encode(args_dict):
importlib.reload(mz_prompts)
importlib.reload(mz_llama_cpp)
model_file = args_dict.get("llama_cpp_model", "")
text = args_dict.get("text", "")
style_presets = args_dict.get("style_presets", "")
options = args_dict.get("llama_cpp_options", {})
keep_device = args_dict.get("keep_device", False)
seed = args_dict.get("seed", -1)
options["seed"] = seed
customize_instruct = args_dict.get("customize_instruct", None)
mz_prompt_utils.Utils.print_log(
f"customize_instruct: {customize_instruct}")
try:
schema = None
if customize_instruct is None:
schema = get_schema_obj(
keys_type={
"description": get_schema_base_type("string"),
"long_prompt": get_schema_base_type("string"),
"main_color_word": get_schema_base_type("string"),
"camera_angle_word": get_schema_base_type("string"),
"style_words": get_schema_array("string"),
"subject_words": get_schema_array("string"),
"light_words": get_schema_array("string"),
"environment_words": get_schema_array("string"),
},
required=[
"description",
"long_prompt",
"main_color_word",
"camera_angle_word",
"style_words",
"subject_words",
"light_words",
"environment_words",
]
)
question = f"IDEA: {style_presets},{text}"
if style_presets == "none":
question = f"IDEA: {text}"
system_prompt = mz_prompts.Beautify_Prompt + mz_prompts.Long_prompt + "\n"
else:
system_prompt = customize_instruct.get("system", "")
question = customize_instruct.get("instruct", "%text%")
system_prompt = system_prompt.replace("%text%", text)
question = question.replace("%text%", text)
mz_prompt_utils.Utils.print_log(f"system_prompt: {system_prompt}")
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(
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 = []
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']})")
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'])})")
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)
else:
full_response = mz_llama_cpp.llama_cpp_simple_interrogator(
model_file=model_file,
system=system_prompt,
question=question,
options=options,
)
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)
# 删除第一个元素
full_response_list.pop(0)
full_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)
# 删除最后一个元素
full_response_list.pop()
full_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", " ")
# 句号换成逗号
while full_response.find(".") != -1:
full_response = full_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(", ,", ",")
full_response = mz_prompt_utils.Utils.prompt_zh_to_en(full_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}"
except Exception as e:
mz_llama_cpp.freed_gpu_memory(model_file=model_file)
raise e
conditionings = None
clip = args_dict.get("clip", None)
if clip is not None:
conditionings = mz_prompt_utils.Utils.a1111_clip_text_encode(
clip, full_response, )
return {"ui": {"string": [full_response,]}, "result": (full_response, conditionings)}
mz_llama_cpp.py
+28 -274
View File
@@ -6,40 +6,12 @@ import subprocess
import sys
import torch
try:
import mz_prompt_utils
import mz_prompt_webserver
from . import mz_prompt_utils
from . import mz_prompt_webserver
except ImportError:
pass
def LlamaCppOptions():
return {
"n_ctx": 2048,
"n_batch": 2048,
"n_threads": 0,
"n_threads_batch": 0,
"split_mode": ["LLAMA_SPLIT_MODE_NONE", "LLAMA_SPLIT_MODE_LAYER", "LLAMA_SPLIT_MODE_ROW",],
"main_gpu": 0,
"n_gpu_layers": -1,
"max_tokens": 4096,
"temperature": 1.6,
"top_p": 0.95,
"min_p": 0.05,
"typical_p": 1.0,
"stop": "",
"frequency_penalty": 0.0,
"presence_penalty": 0.0,
"repeat_penalty": 1.1,
"top_k": 50,
"tfs_z": 1.0,
"mirostat_mode": ["none", "mirostat", "mirostat_v2"],
"mirostat_tau": 5.0,
"mirostat_eta": 0.1,
}
def check_llama_cpp_requirements():
last_version = "0.2.63"
try:
@@ -103,6 +75,32 @@ def check_llama_cpp_requirements():
print("llama_cpp installed successfully. (llama_cpp安装成功)")
def LlamaCppOptions():
return {
"n_ctx": 2048,
"n_batch": 2048,
"n_threads": 0,
"n_threads_batch": 0,
"split_mode": ["LLAMA_SPLIT_MODE_NONE", "LLAMA_SPLIT_MODE_LAYER", "LLAMA_SPLIT_MODE_ROW",],
"main_gpu": 0,
"n_gpu_layers": -1,
"max_tokens": 4096,
"temperature": 1.6,
"top_p": 0.95,
"min_p": 0.05,
"typical_p": 1.0,
"stop": "",
"frequency_penalty": 0.0,
"presence_penalty": 0.0,
"repeat_penalty": 1.1,
"top_k": 50,
"tfs_z": 1.0,
"mirostat_mode": ["none", "mirostat", "mirostat_v2"],
"mirostat_tau": 5.0,
"mirostat_eta": 0.1,
}
def freed_gpu_memory(model_file):
check_llama_cpp_requirements()
@@ -237,33 +235,6 @@ def llama_cpp_messages(model_file, chat_handler=None, messages=[], options={}):
return result
def get_schema_base_type(t):
return {
"type": t,
}
def get_schema_obj(keys_type={}, required=[]):
item = {}
for key, value in keys_type.items():
if type(value) == str:
value = get_schema_base_type(value)
item[key] = value
return {
"type": "object",
"properties": item,
"required": required
}
def get_schema_array(item_type="string"):
if type(item_type) == str:
item_type = get_schema_base_type(item_type)
return {
"type": "array",
"items": item_type,
}
def llama_cpp_simple_interrogator_to_json(model_file, use_system=True, system=None, question="", schema={}, options={}):
if system is None:
@@ -409,220 +380,3 @@ def llava_cpp_simple_interrogator_to_json(
model_file, mmproj_file, system=system, question=question, image=image, options=options)
return result
high_quality_prompt = "((high quality:1.4), (best quality:1.4), (masterpiece:1.4), (8K resolution), (2k wallpaper))"
style_presets_prompt = {
"none": "",
"high_quality": high_quality_prompt,
"photography": f"{high_quality_prompt}, (RAW photo, best quality), (realistic, photo-realistic:1.2), (bokeh, cinematic shot, dynamic composition, incredibly detailed, sharpen, details, intricate detail, professional lighting, film lighting, 35mm, anamorphic, lightroom, cinematography, bokeh, lens flare, film grain, HDR10, 8K)",
"illustration": f"{high_quality_prompt}, ((detailed matte painting, intricate detail, splash screen, complementary colors), (detailed),(intricate details),illustration,an extremely delicate and beautiful,ultra-detailed,highres,extremely detailed)",
}
def get_style_presets():
return [
"none",
"high_quality",
"photography",
"illustration",
]
def base_query_beautify_prompt_text(args_dict):
model_file = args_dict.get("llama_cpp_model", "")
text = args_dict.get("text", "")
style_presets = args_dict.get("style_presets", "")
options = args_dict.get("llama_cpp_options", {})
keep_device = args_dict.get("keep_device", False)
seed = args_dict.get("seed", -1)
options["seed"] = seed
import mz_prompts
importlib.reload(mz_prompts)
customize_instruct = args_dict.get("customize_instruct", None)
mz_prompt_utils.Utils.print_log(
f"customize_instruct: {customize_instruct}")
try:
schema = None
if customize_instruct is None:
schema = get_schema_obj(
keys_type={
"description": get_schema_base_type("string"),
"long_prompt": get_schema_base_type("string"),
"main_color_word": get_schema_base_type("string"),
"camera_angle_word": get_schema_base_type("string"),
"style_words": get_schema_array("string"),
"subject_words": get_schema_array("string"),
"light_words": get_schema_array("string"),
"environment_words": get_schema_array("string"),
},
required=[
"description",
"long_prompt",
"main_color_word",
"camera_angle_word",
"style_words",
"subject_words",
"light_words",
"environment_words",
]
)
question = f"IDEA: {style_presets},{text}"
if style_presets == "none":
question = f"IDEA: {text}"
system_prompt = mz_prompts.Beautify_Prompt + mz_prompts.Long_prompt + "\n"
else:
system_prompt = customize_instruct.get("system", "")
question = customize_instruct.get("instruct", "%text%")
system_prompt = system_prompt.replace("%text%", text)
question = question.replace("%text%", text)
mz_prompt_utils.Utils.print_log(f"system_prompt: {system_prompt}")
mz_prompt_utils.Utils.print_log(f"question: {question}")
if schema is not None:
response_json = 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 = []
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']})")
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'])})")
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)
else:
full_response = llama_cpp_simple_interrogator(
model_file=model_file,
system=system_prompt,
question=question,
options=options,
)
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)
# 删除第一个元素
full_response_list.pop(0)
full_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)
# 删除最后一个元素
full_response_list.pop()
full_response = end_str.join(full_response_list)
if keep_device is False:
freed_gpu_memory(model_file=model_file)
# 去除换行
while full_response.find("\n") != -1:
full_response = full_response.replace("\n", " ")
# 句号换成逗号
while full_response.find(".") != -1:
full_response = full_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(", ,", ",")
full_response = mz_prompt_utils.Utils.prompt_zh_to_en(full_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}"
return full_response
except Exception as e:
freed_gpu_memory(model_file=model_file)
# mz_utils.Utils.print_log(f"Error in auto_prompt_text: {e}")
raise e
def get_image_composition_json_schema():
system = "Please make a specific composition of the proposed content. You need to have the width and height of the overall picture and a list of specific element positions."
return system, {
"type": "object",
"properties": {
"width": {
"type": "number"
},
"height": {
"type": "number"
},
"bbox": {
"type": "array",
"items": {
"type": "object",
"properties": {
"english_name": {
"type": "string"
},
"left": {
"type": "number"
},
"top": {
"type": "number"
},
"width": {
"type": "number"
},
"height": {
"type": "number"
}
},
"required": ["english_name", "left", "top", "width", "height"]
}
}
},
"required": ["width", "height", "bbox"]
}
mz_llava.py
+4 -4
View File
@@ -1,12 +1,12 @@
import json
import os
import mz_prompt_utils
import mz_llama_cpp
from . import mz_prompt_utils
from . import mz_llama_cpp
from . import mz_prompts
from . import mz_prompt_webserver
import importlib
import mz_prompts
import mz_prompt_webserver
LLava_models = [
"llava-1.6-mistral-7b-gguf/llava-v1.6-mistral-7b.Q5_K_M.gguf",
mz_openaiapi.py
+3 -3
View File
@@ -3,9 +3,9 @@ import sys
import json
import subprocess
import mz_prompt_utils
import mz_llama_cpp
import mz_prompts
from . import mz_prompt_utils
from . import mz_llama_cpp
from . import mz_prompts
def zhipu_json_fix(input_data):
mz_prompt_utils.py
+6
View File
@@ -165,6 +165,12 @@ class Utils:
os.makedirs(models_path, exist_ok=True)
return models_path
def get_gguf_models_path():
models_path = os.path.join(
folder_paths.models_dir, "gguf")
os.makedirs(models_path, exist_ok=True)
return models_path
def translate_text(text, from_code, to_code):
try:
import argostranslate
mz_prompt_webserver.py
+1 -1
View File
@@ -2,7 +2,7 @@
import asyncio
import uuid
import mz_prompt_utils
from . import mz_prompt_utils
web_msg_pool = {
mz_prompts.py
+15
View File
@@ -18,3 +18,18 @@ Standardize_Prompt = """
Extract the content about Stable Diffusion style from the following input and combine it into a json array. Note that the output will be directly used in the program.
Please output the standardized json content.
"""
# 来自https://github.com/jiayev/GPT4V-Image-Captioner
GPT4VImageCaptioner_Prompt = """
As an AI image tagging expert, please provide precise tags for these images to enhance CLIP model's understanding of the content.
Employ succinct keywords or phrases, 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.
For other image categories, apply appropriate and common descriptive tags as well.
Recognize and tag any celebrities, well-known landmark or IPs if clearly featured in the image.
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.
"""
v1/init.py
+195
View File
@@ -0,0 +1,195 @@
from ..mz_prompt_utils import Utils
NODE_CLASS_MAPPINGS = {
}
NODE_DISPLAY_NAME_MAPPINGS = {
}
import importlib
from . import mz_llama3
from . import mz_phi3
from .. import mz_llama_cpp
from .. import mz_llama_core_nodes
AUTHOR_NAME = u"MinusZone"
CATEGORY_NAME = f"{AUTHOR_NAME} - Prompt/deprecated"
def getCommonCLIPTextEncodeInput():
style_presets = mz_llama_core_nodes.get_style_presets()
CommonCLIPTextEncodeInput = {
"required": {
"prompt_version": (["v1"], {"default": "v1"}),
"style_presets": (
style_presets, {"default": style_presets[1]}
),
"text": ("STRING", {"multiline": True, }),
"keep_device": ([False, True], {"default": False}),
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
},
"optional": {
"clip": ("CLIP", ),
"llama_cpp_options": ("LLamaCPPOptions", ),
"customize_instruct": ("CustomizeInstruct", ),
# "customize_json_schema": ("STRING", ),
}
}
return CommonCLIPTextEncodeInput
class MZ_LLama3CLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
m_models = mz_llama3.llama3_models.copy()
for i in range(len(m_models)):
if mz_llama3.get_exist_model(m_models[i]) is not None:
m_models[i] += "[downloaded]"
result = {
"required": {
"llama_cpp_model": (m_models, {"default": m_models[0]}),
"download_source": (
["none", "modelscope", "hf-mirror.com",],
{"default": "none"}
),
},
"optional": {},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_llama3.query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_LLama3CLIPTextEncode"] = MZ_LLama3CLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_LLama3CLIPTextEncode"] = f"{AUTHOR_NAME} - deprecated - CLIPTextEncode(LLama3)"
class MZ_Phi3CLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
m_models = mz_phi3.phi3_models.copy()
for i in range(len(m_models)):
if mz_llama3.get_exist_model(m_models[i]) is not None:
m_models[i] += "[downloaded]"
importlib.reload(mz_phi3)
result = {
"required": {
"llama_cpp_model": (m_models, {"default": m_models[0]}),
"download_source": (
["none", "modelscope", "hf-mirror.com",],
{"default": "none"}
),
},
"optional": {},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
importlib.reload(mz_llama3)
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_phi3.query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_Phi3CLIPTextEncode"] = MZ_Phi3CLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_Phi3CLIPTextEncode"] = f"{AUTHOR_NAME} - deprecated - CLIPTextEncode(Phi3)"
class MZ_BaseLLamaCPPCLIPTextEncode:
@classmethod
def INPUT_TYPES(s):
importlib.reload(mz_llama_cpp)
result = {
"required": {
"llama_cpp_model": ("STRING", {"default": "", "placeholder": "model_path"}),
},
"optional": {
},
}
common_input = getCommonCLIPTextEncodeInput()
for key in common_input["required"]:
result["required"][key] = common_input["required"][key]
for key in common_input["optional"]:
result["optional"][key] = common_input["optional"][key]
return result
RETURN_TYPES = ("STRING", "CONDITIONING",)
RETURN_NAMES = ("text", "conditioning",)
OUTPUT_NODE = True
FUNCTION = "encode"
CATEGORY = CATEGORY_NAME
def encode(self, **kwargs):
importlib.reload(mz_llama3)
kwargs["llama_cpp_model"] = kwargs.get(
"llama_cpp_model", "").replace("[downloaded]", "")
text = mz_llama_cpp.base_query_beautify_prompt_text(kwargs)
conditionings = None
clip = kwargs.get("clip", None)
if clip is not None:
conditionings = Utils.a1111_clip_text_encode(clip, text, )
return {"ui": {"string": [text,]}, "result": (text, conditionings)}
NODE_CLASS_MAPPINGS["MZ_BaseLLamaCPPCLIPTextEncode"] = MZ_BaseLLamaCPPCLIPTextEncode
NODE_DISPLAY_NAME_MAPPINGS[
"MZ_BaseLLamaCPPCLIPTextEncode"] = f"{AUTHOR_NAME} - deprecated - CLIPTextEncode(BaseLLamaCPP)"
v1/mz_deprecated.py
+160
View File
@@ -0,0 +1,160 @@
from ..mz_prompt_utils import Utils
from ..mz_llama_cpp import *
from ..mz_llama_core_nodes import *
from ..mz_prompts import *
def base_query_beautify_prompt_text(args_dict):
model_file = args_dict.get("llama_cpp_model", "")
text = args_dict.get("text", "")
style_presets = args_dict.get("style_presets", "")
options = args_dict.get("llama_cpp_options", {})
keep_device = args_dict.get("keep_device", False)
seed = args_dict.get("seed", -1)
options["seed"] = seed
customize_instruct = args_dict.get("customize_instruct", None)
Utils.print_log(
f"customize_instruct: {customize_instruct}")
try:
schema = None
if customize_instruct is None:
schema = get_schema_obj(
keys_type={
"description": get_schema_base_type("string"),
"long_prompt": get_schema_base_type("string"),
"main_color_word": get_schema_base_type("string"),
"camera_angle_word": get_schema_base_type("string"),
"style_words": get_schema_array("string"),
"subject_words": get_schema_array("string"),
"light_words": get_schema_array("string"),
"environment_words": get_schema_array("string"),
},
required=[
"description",
"long_prompt",
"main_color_word",
"camera_angle_word",
"style_words",
"subject_words",
"light_words",
"environment_words",
]
)
question = f"IDEA: {style_presets},{text}"
if style_presets == "none":
question = f"IDEA: {text}"
system_prompt = mz_prompts.Beautify_Prompt + mz_prompts.Long_prompt + "\n"
else:
system_prompt = customize_instruct.get("system", "")
question = customize_instruct.get("instruct", "%text%")
system_prompt = system_prompt.replace("%text%", text)
question = question.replace("%text%", text)
Utils.print_log(f"system_prompt: {system_prompt}")
Utils.print_log(f"question: {question}")
if schema is not None:
response_json = llama_cpp_simple_interrogator_to_json(
model_file=model_file,
system=system_prompt,
question=question,
schema=schema,
options=options,
)
Utils.print_log(f"response_json: {response_json}")
response = json.loads(response_json)
full_responses = []
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']})")
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'])})")
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)
else:
full_response = llama_cpp_simple_interrogator(
model_file=model_file,
system=system_prompt,
question=question,
options=options,
)
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)
# 删除第一个元素
full_response_list.pop(0)
full_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)
# 删除最后一个元素
full_response_list.pop()
full_response = end_str.join(full_response_list)
if keep_device is False:
freed_gpu_memory(model_file=model_file)
# 去除换行
while full_response.find("\n") != -1:
full_response = full_response.replace("\n", " ")
# 句号换成逗号
while full_response.find(".") != -1:
full_response = full_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(", ,", ",")
full_response = Utils.prompt_zh_to_en(full_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}"
return full_response
except Exception as e:
freed_gpu_memory(model_file=model_file)
# mz_utils.Utils.print_log(f"Error in auto_prompt_text: {e}")
raise e
mz_llama3.py → v1/mz_llama3.py
+4 -8
View File
@@ -1,7 +1,7 @@
import json
import mz_prompt_utils
import mz_llama_cpp
from .. import mz_prompt_utils
from .. import mz_llama_cpp
from . import mz_deprecated
import importlib
@@ -46,10 +46,6 @@ def query_beautify_prompt_text(args_dict):
model_name = args_dict.get("llama_cpp_model", "")
download_source = args_dict.get("download_source", None)
import mz_prompts
importlib.reload(mz_prompts)
importlib.reload(mz_llama_cpp)
try:
model_file = get_exist_model(model_name)
@@ -66,7 +62,7 @@ def query_beautify_prompt_text(args_dict):
model_file = mz_prompt_utils.Utils.hf_download_model(model_url)
args_dict["llama_cpp_model"] = model_file
full_response = mz_llama_cpp.base_query_beautify_prompt_text(
full_response = mz_deprecated.base_query_beautify_prompt_text(
args_dict=args_dict)
return full_response
mz_phi3.py → v1/mz_phi3.py
+2 -9
View File
@@ -1,9 +1,6 @@
import json
import mz_prompt_utils
import mz_llama_cpp
import importlib
from .. import mz_prompt_utils
from .. import mz_llama_cpp
phi3_models = [
"Phi-3-mini-4k-instruct-q4.gguf"
@@ -33,10 +30,6 @@ def query_beautify_prompt_text(args_dict):
model_name = args_dict.get("llama_cpp_model", "")
download_source = args_dict.get("download_source", None)
import mz_prompts
importlib.reload(mz_prompts)
importlib.reload(mz_llama_cpp)
try:
model_file = get_exist_model(model_name)
web/prompt_mz.js
-4
View File
@@ -41,14 +41,10 @@ async function waitMessage() {
};
websocket.onclose = async (event) => {
console.log("Connection closed", event);
await sleep(1000);
await waitMessage();
};
websocket.onerror = async (event) => {
console.log("Connection error", event);
await sleep(1000);
await waitMessage();
};
for (;;) {