ehristoforu's picture
Upload folder using huggingface_hub
0163a2c verified
import requests
import json
import gradio as gr
import urllib.request
import urllib.parse
import urllib.error
import os
import re
import datetime
import platform
from PIL import Image
from io import BytesIO
from collections import defaultdict
from modules.images import read_info_from_image
from modules.shared import cmd_opts, opts
from modules.paths import models_path, extensions_dir, data_path
from html import escape
from scripts.civitai_global import print
import scripts.civitai_global as gl
import scripts.civitai_download as _download
try:
from fake_useragent import UserAgent
except ImportError:
print("Python module 'fake_useragent' has not been imported correctly, please try to restart or install it manually.")
gl.init()
def contenttype_folder(content_type, desc=None, fromCheck=False, custom_folder=None):
use_LORA = getattr(opts, "use_LORA", False)
folder = None
if desc:
desc = desc.upper()
else:
desc = "PLACEHOLDER"
if custom_folder:
main_models = custom_folder
main_data = custom_folder
else:
main_models = models_path
main_data = data_path
if content_type == "modelFolder":
folder = os.path.join(main_models)
if content_type == "Checkpoint":
if cmd_opts.ckpt_dir and not custom_folder:
folder = cmd_opts.ckpt_dir
else:
folder = os.path.join(main_models,"Stable-diffusion")
elif content_type == "Hypernetwork":
if cmd_opts.hypernetwork_dir and not custom_folder:
folder = cmd_opts.hypernetwork_dir
else:
folder = os.path.join(main_models, "hypernetworks")
elif content_type == "TextualInversion":
if cmd_opts.embeddings_dir and not custom_folder:
folder = cmd_opts.embeddings_dir
else:
folder = os.path.join(main_data, "embeddings")
elif content_type == "AestheticGradient":
if not custom_folder:
folder = os.path.join(extensions_dir, "stable-diffusion-webui-aesthetic-gradients", "aesthetic_embeddings")
else:
folder = os.path.join(custom_folder, "aesthetic_embeddings")
elif content_type == "LORA":
if cmd_opts.lora_dir and not custom_folder:
folder = cmd_opts.lora_dir
else:
folder = folder = os.path.join(main_models, "Lora")
elif content_type == "LoCon":
folder = os.path.join(main_models, "LyCORIS")
if use_LORA and not fromCheck:
if cmd_opts.lora_dir and not custom_folder:
folder = cmd_opts.lora_dir
else:
folder = folder = os.path.join(main_models, "Lora")
elif content_type == "VAE":
if cmd_opts.vae_dir and not custom_folder:
folder = cmd_opts.vae_dir
else:
folder = os.path.join(main_models, "VAE")
elif content_type == "Controlnet":
folder = os.path.join(main_models, "ControlNet")
elif content_type == "Poses":
folder = os.path.join(main_models, "Poses")
elif content_type == "Upscaler":
if "SWINIR" in desc:
if cmd_opts.swinir_models_path and not custom_folder:
folder = cmd_opts.swinir_models_path
else:
folder = os.path.join(main_models, "SwinIR")
elif "REALESRGAN" in desc:
if cmd_opts.realesrgan_models_path and not custom_folder:
folder = cmd_opts.realesrgan_models_path
else:
folder = os.path.join(main_models, "RealESRGAN")
elif "GFPGAN" in desc:
if cmd_opts.gfpgan_models_path and not custom_folder:
folder = cmd_opts.gfpgan_models_path
else:
folder = os.path.join(main_models, "GFPGAN")
elif "BSRGAN" in desc:
if cmd_opts.bsrgan_models_path and not custom_folder:
folder = cmd_opts.bsrgan_models_path
else:
folder = os.path.join(main_models, "BSRGAN")
else:
if cmd_opts.esrgan_models_path and not custom_folder:
folder = cmd_opts.esrgan_models_path
else:
folder = os.path.join(main_models, "ESRGAN")
elif content_type == "MotionModule":
folder = os.path.join(extensions_dir, "sd-webui-animatediff", "model")
elif content_type == "Workflows":
folder = os.path.join(main_models, "Workflows")
elif content_type == "Other":
if "ADETAILER" in desc:
folder = os.path.join(main_models, "adetailer")
else:
folder = os.path.join(main_models, "Other")
elif content_type == "Wildcards":
folder = os.path.join(extensions_dir, "UnivAICharGen", "wildcards")
if not os.path.exists(folder):
folder = os.path.join(extensions_dir, "sd-dynamic-prompts", "wildcards")
return folder
def api_to_data(content_type, sort_type, period_type, use_search_term, current_page, base_filter, only_liked, tile_count, search_term=None, nsfw=None, timeOut=None, isNext=None, inputs_changed=None):
if current_page in [0, None, ""]:
current_page = 1
if inputs_changed:
gl.file_scan = False
api_url = f"https://civitai.com/api/v1/models?limit={tile_count}&page=1"
else:
api_url = f"https://civitai.com/api/v1/models?limit={tile_count}&page={current_page}"
if timeOut:
if isNext:
next_page = str(int(current_page) + 1)
else:
if current_page not in [1, 0, None, ""]:
next_page = str(int(current_page) - 1)
api_url = f"https://civitai.com/api/v1/models?limit={tile_count}&page={next_page}"
if period_type:
period_type = period_type.replace(" ", "")
query = {'sort': sort_type, 'period': period_type}
types_query_str = ""
if content_type:
types_query_str = "".join([f"&types={type}" for type in content_type])
query_str = urllib.parse.urlencode(query, quote_via=urllib.parse.quote)
if types_query_str:
query_str += types_query_str
if use_search_term != "None" and search_term:
search_term = search_term.replace("\\","\\\\")
if "civitai.com" in search_term:
match = re.search(r'models/(\d+)', search_term)
model_number = match.group(1)
query_str = f"&ids={urllib.parse.quote(model_number)}"
elif use_search_term == "User name":
query_str += f"&username={urllib.parse.quote(search_term)}"
elif use_search_term == "Tag":
query_str += f"&tag={urllib.parse.quote(search_term)}"
else:
query_str += f"&query={urllib.parse.quote(search_term)}"
if base_filter:
for base in base_filter:
query_str += f"&baseModels={urllib.parse.quote(base)}"
if only_liked:
query_str += f"&favorites=true"
if nsfw == False:
query_str += f"&nsfw=false"
full_url = f"{api_url}&{query_str}"
if gl.file_scan:
highest_number = max(gl.url_list_with_numbers.keys())
full_url = gl.url_list_with_numbers.get(int(current_page))
nextPage = int(current_page) + 1
prevPage = int(current_page) - 1
data = request_civit_api(full_url)
data["metadata"]["currentPage"] = current_page
data["metadata"]["totalPages"] = highest_number
if not nextPage > highest_number:
data["metadata"]["nextPage"] = gl.url_list_with_numbers.get(nextPage)
if not prevPage == 0:
data["metadata"]["prevPage"] = gl.url_list_with_numbers.get(prevPage)
else:
data = request_civit_api(full_url)
return data
def model_list_html(json_data):
video_playback = getattr(opts, "video_playback", True)
playback = ""
if video_playback: playback = "autoplay loop"
hide_early_access = getattr(opts, "hide_early_access", True)
filtered_items = []
current_time = datetime.datetime.utcnow()
for item in json_data['items']:
versions_to_keep = []
for version in item['modelVersions']:
if not version['files']:
continue
if hide_early_access:
early_access_days = version['earlyAccessTimeFrame']
if early_access_days != 0:
published_at_str = version.get('publishedAt')
if published_at_str is not None:
published_at = datetime.datetime.strptime(version['publishedAt'], "%Y-%m-%dT%H:%M:%S.%fZ")
adjusted_date = published_at + datetime.timedelta(days=early_access_days)
if not current_time > adjusted_date or not published_at_str:
continue
versions_to_keep.append(version)
if versions_to_keep:
item['modelVersions'] = versions_to_keep
filtered_items.append(item)
json_data['items'] = filtered_items
HTML = '<div class="column civmodellist">'
sorted_models = {}
existing_files = set()
existing_files_sha256 = set()
model_folders = set()
for item in json_data['items']:
model_folder = os.path.join(contenttype_folder(item['type'], item['description']))
model_folders.add(model_folder)
for folder in model_folders:
for root, dirs, files in os.walk(folder, followlinks=True):
for file in files:
existing_files.add(file)
if file.endswith('.json'):
json_path = os.path.join(root, file)
with open(json_path, 'r', encoding="utf-8") as f:
try:
json_file = json.load(f)
if isinstance(json_file, dict):
sha256 = json_file.get('sha256')
if sha256:
existing_files_sha256.add(sha256.upper())
else:
print(f"Invalid JSON data in {json_path}. Expected a dictionary.")
except Exception as e:
print(f"Error decoding JSON in {json_path}: {e}")
for item in json_data['items']:
model_id = item.get('id')
model_name = item.get('name')
nsfw = ""
installstatus = ""
baseModel = ""
try:
if 'baseModel' in item['modelVersions'][0]:
baseModel = item['modelVersions'][0]['baseModel']
except:
baseModel = "Not Found"
try:
if 'updatedAt' in item['modelVersions'][0]:
date = item['modelVersions'][0]['updatedAt'].split('T')[0]
except:
baseModel = "Not Found"
if gl.sortNewest:
if date not in sorted_models:
sorted_models[date] = []
if any(item['modelVersions']):
if len(item['modelVersions'][0]['images']) > 0:
if item["modelVersions"][0]["images"][0]['nsfw'] not in ["None", "Soft"]:
nsfw = "civcardnsfw"
media_type = item["modelVersions"][0]["images"][0]["type"]
image = item["modelVersions"][0]["images"][0]["url"]
if media_type == "video":
image = image.replace("width=", "transcode=true,width=")
imgtag = f'<video class="video-bg" {playback} muted playsinline><source src="{image}" type="video/mp4"></video>'
else:
imgtag = f'<img src="{image}"></img>'
else:
imgtag = f'<img src="./file=html/card-no-preview.png"></img>'
installstatus = None
for version in reversed(item['modelVersions']):
for file in version.get('files', []):
file_name = file['name']
file_sha256 = file.get('hashes', {}).get('SHA256', "").upper()
name_match = file_name in existing_files
sha256_match = file_sha256 in existing_files_sha256
if name_match or sha256_match:
if version == item['modelVersions'][0]:
installstatus = "civmodelcardinstalled"
else:
installstatus = "civmodelcardoutdated"
model_name_js = model_name.replace("'", "\\'")
model_string = escape(f"{model_name_js} ({model_id})")
model_card = f'<figure class="civmodelcard {nsfw} {installstatus}" base-model="{baseModel}" date="{date}" onclick="select_model(\'{model_string}\', event)">'
if installstatus != "civmodelcardinstalled":
model_card += f'<input type="checkbox" class="model-checkbox" id="checkbox-{model_string}" onchange="multi_model_select(\'{model_string}\', \'{item["type"]}\', this.checked)" style="opacity: 0; position: absolute; top: 10px; right: 10px;">' \
+ f'<label for="checkbox-{model_string}" class="custom-checkbox"></label>'
if len(item["name"]) > 40:
display_name = item["name"][:40] + '...'
else:
display_name = item["name"]
display_name = escape(display_name)
full_name = escape(item['name'])
model_card += imgtag \
+ f'<figcaption title="{full_name}">{display_name}</figcaption></figure>'
if gl.sortNewest:
sorted_models[date].append(model_card)
else:
HTML += model_card
if gl.sortNewest:
for date, cards in sorted(sorted_models.items(), reverse=True):
HTML += f'<div class="date-section"><h4>{date}</h4><hr style="margin-bottom: 5px; margin-top: 5px;">'
HTML += '<div class="card-row">'
for card in cards:
HTML += card
HTML += '</div></div>'
HTML += '</div>'
return HTML
def update_prev_page(content_type, sort_type, period_type, use_search_term, search_term, current_page, base_filter, only_liked, nsfw, tile_count):
return update_next_page(content_type, sort_type, period_type, use_search_term, search_term, current_page, base_filter, only_liked, nsfw, tile_count, isNext=False)
def update_next_page(content_type, sort_type, period_type, use_search_term, search_term, current_page, base_filter, only_liked, nsfw, tile_count, isNext=True):
use_LORA = getattr(opts, "use_LORA", False)
if content_type:
if use_LORA and 'LORA & LoCon' in content_type:
content_type.remove('LORA & LoCon')
if 'LORA' not in content_type:
content_type.append('LORA')
if 'LoCon' not in content_type:
content_type.append('LoCon')
if gl.json_data is None or gl.json_data == "timeout":
timeOut = True
return_values = update_model_list(content_type, sort_type, period_type, use_search_term, search_term, current_page, base_filter, only_liked, nsfw, timeOut=timeOut, isNext=isNext)
timeOut = False
return return_values
current_inputs = (content_type, sort_type, period_type, use_search_term, search_term, tile_count, base_filter, nsfw)
if current_inputs != gl.previous_inputs and gl.previous_inputs != None:
inputs_changed = True
else:
inputs_changed = False
if inputs_changed:
return_values = update_model_list(content_type, sort_type, period_type, use_search_term, search_term, current_page, base_filter, only_liked, nsfw, tile_count)
return return_values
if not gl.file_scan:
if isNext:
if gl.json_data['metadata']['nextPage'] is not None:
gl.json_data = request_civit_api(gl.json_data['metadata']['nextPage'])
else:
gl.json_data = None
else:
if gl.json_data['metadata']['prevPage'] is not None:
gl.json_data = request_civit_api(gl.json_data['metadata']['prevPage'])
else:
gl.json_data = None
else:
highest_number = max(gl.url_list_with_numbers.keys())
if isNext:
if gl.json_data['metadata']['nextPage'] is not None:
currentPage = int(gl.json_data['metadata']['currentPage'])
nextPage = currentPage + 2
prevPage = currentPage
pageCount = currentPage + 1
gl.json_data = request_civit_api(gl.json_data['metadata']['nextPage'])
gl.json_data["metadata"]["totalPages"] = highest_number
if not nextPage > highest_number:
gl.json_data["metadata"]["nextPage"] = gl.url_list_with_numbers.get(nextPage)
if not prevPage == 0:
gl.json_data["metadata"]["prevPage"] = gl.url_list_with_numbers.get(prevPage)
gl.json_data["metadata"]["currentPage"] = pageCount
else:
gl.json_data = None
else:
if gl.json_data['metadata']['prevPage'] is not None:
currentPage = int(gl.json_data['metadata']['currentPage'])
nextPage = currentPage
prevPage = currentPage - 2
pageCount = currentPage - 1
gl.json_data = request_civit_api(gl.json_data['metadata']['prevPage'])
gl.json_data["metadata"]["totalPages"] = highest_number
if not nextPage > highest_number:
gl.json_data["metadata"]["nextPage"] = gl.url_list_with_numbers.get(nextPage)
if not prevPage == 0:
gl.json_data["metadata"]["prevPage"] = gl.url_list_with_numbers.get(prevPage)
gl.json_data["metadata"]["currentPage"] = pageCount
else:
gl.json_data = None
if gl.json_data is None:
return
if gl.json_data == "timeout":
HTML = '<div style="font-size: 24px; text-align: center; margin: 50px !important;">The Civit-API has timed out, please try again.<br>The servers might be too busy or down if the issue persists.</div>'
hasPrev = current_page not in [0, 1]
hasNext = current_page == 1 or hasPrev
model_dict = {}
if gl.json_data != None and gl.json_data != "timeout":
(hasPrev, hasNext, current_page, total_pages) = pagecontrol(gl.json_data)
model_dict = {}
try:
gl.json_data['items']
except TypeError:
return gr.Dropdown.update(choices=[], value=None)
HTML = model_list_html(gl.json_data)
page_string = f"Page: {current_page}/{total_pages}"
return (
gr.Dropdown.update(choices=[v for k, v in model_dict.items()], value="", interactive=True), # Model List
gr.Dropdown.update(choices=[], value=""), # Version List
gr.HTML.update(value=HTML), # HTML Tiles
gr.Button.update(interactive=hasPrev), # Prev Page Button
gr.Button.update(interactive=hasNext), # Next Page Button
gr.Slider.update(value=current_page, maximum=total_pages, label=page_string), # Page Count
gr.Button.update(interactive=False), # Save Tags
gr.Button.update(interactive=False), # Save Images
gr.Button.update(interactive=False, visible=False if gl.isDownloading else True), # Download Button
gr.Button.update(interactive=False, visible=False), # Delete Button
gr.Textbox.update(interactive=False, value=None), # Install Path
gr.Dropdown.update(choices=[], value="", interactive=False), # Sub Folder List
gr.Dropdown.update(choices=[], value="", interactive=False), # File List
gr.HTML.update(value='<div style="min-height: 0px;"></div>'), # Preview HTML
gr.Textbox.update(value=None), # Trained Tags
gr.Textbox.update(value=None), # Base Model
gr.Textbox.update(value=None) # Model Filename
)
def pagecontrol(json_data):
current_page = f"{json_data['metadata']['currentPage']}"
total_pages = f"{json_data['metadata']['totalPages']}"
hasNext = False
hasPrev = False
if 'nextPage' in json_data['metadata']:
hasNext = True
if 'prevPage' in json_data['metadata']:
hasPrev = True
return hasPrev, hasNext, current_page, total_pages
def update_model_list(content_type=None, sort_type=None, period_type=None, use_search_term=None, search_term=None, current_page=None, base_filter=None, only_liked=None, nsfw=None, tile_count=None, timeOut=None, isNext=None, from_ver=False, from_installed=False):
use_LORA = getattr(opts, "use_LORA", False)
model_list = []
id_list = []
if content_type:
if use_LORA and 'LORA & LoCon' in content_type:
content_type.remove('LORA & LoCon')
if 'LORA' not in content_type:
content_type.append('LORA')
if 'LoCon' not in content_type:
content_type.append('LoCon')
if not from_ver and not from_installed:
gl.ver_json = None
current_inputs = (content_type, sort_type, period_type, use_search_term, search_term, tile_count, base_filter, nsfw)
if current_inputs != gl.previous_inputs and gl.previous_inputs != None:
inputs_changed = True
else:
inputs_changed = False
gl.previous_inputs = current_inputs
gl.json_data = api_to_data(content_type, sort_type, period_type, use_search_term, current_page, base_filter, only_liked, tile_count, search_term, nsfw, timeOut, isNext, inputs_changed)
if gl.json_data == "timeout":
HTML = '<div style="font-size: 24px; text-align: center; margin: 50px !important;">The Civit-API has timed out, please try again.<br>The servers might be too busy or down if the issue persists.</div>'
hasPrev = current_page not in [0, 1]
hasNext = current_page == 1 or hasPrev
if gl.json_data is None:
return
if from_installed or from_ver:
gl.json_data = gl.ver_json
if gl.json_data != None and gl.json_data != "timeout":
if not from_ver:
(hasPrev, hasNext, current_page, total_pages) = pagecontrol(gl.json_data)
else:
current_page = 1
total_pages = 1
hasPrev = False
hasNext = False
for item in gl.json_data['items']:
model_list.append(f"{item['name']} ({item['id']})")
HTML = model_list_html(gl.json_data)
else:
current_page = 1
total_pages = 1
page_string = f"Page: {current_page}/{total_pages}"
return (
gr.Dropdown.update(choices=model_list, value="", interactive=True), # Model List
gr.Dropdown.update(choices=[], value=""), # Version List
gr.HTML.update(value=HTML), # HTML Tiles
gr.Button.update(interactive=hasPrev), # Prev Page Button
gr.Button.update(interactive=hasNext), # Next Page Button
gr.Slider.update(value=current_page, maximum=total_pages, label=page_string), # Page Count
gr.Button.update(interactive=False), # Save Tags
gr.Button.update(interactive=False), # Save Images
gr.Button.update(interactive=False, visible=False if gl.isDownloading else True), # Download Button
gr.Button.update(interactive=False, visible=False), # Delete Button
gr.Textbox.update(interactive=False, value=None, visible=True), # Install Path
gr.Dropdown.update(choices=[], value="", interactive=False), # Sub Folder List
gr.Dropdown.update(choices=[], value="", interactive=False), # File List
gr.HTML.update(value='<div style="min-height: 0px;"></div>'), # Preview HTML
gr.Textbox.update(value=None), # Trained Tags
gr.Textbox.update(value=None), # Base Model
gr.Textbox.update(value=None) # Model Filename
)
def update_model_versions(model_id, json_input=None):
if json_input:
api_json = json_input
else:
api_json = gl.json_data
for item in api_json['items']:
if int(item['id']) == int(model_id):
content_type = item['type']
desc = item.get('description', "None")
versions_dict = defaultdict(list)
installed_versions = set()
model_folder = os.path.join(contenttype_folder(content_type, desc))
gl.main_folder = model_folder
versions = item['modelVersions']
version_files = set()
for version in versions:
versions_dict[version['name']].append(item["name"])
for version_file in version['files']:
file_sha256 = version_file.get('hashes', {}).get('SHA256', "").upper()
version_filename = version_file['name']
version_files.add((version['name'], version_filename, file_sha256))
for root, _, files in os.walk(model_folder, followlinks=True):
for file in files:
if file.endswith('.json'):
try:
json_path = os.path.join(root, file)
with open(json_path, 'r', encoding="utf-8") as f:
json_data = json.load(f)
if isinstance(json_data, dict):
if 'sha256' in json_data and json_data['sha256']:
sha256 = json_data.get('sha256', "").upper()
for version_name, _, file_sha256 in version_files:
if sha256 == file_sha256:
installed_versions.add(version_name)
break
except Exception as e:
print(f"failed to read: \"{file}\": {e}")
for version_name, version_filename, _ in version_files:
if file == version_filename:
installed_versions.add(version_name)
break
version_names = list(versions_dict.keys())
display_version_names = [f"{v} [Installed]" if v in installed_versions else v for v in version_names]
default_installed = next((f"{v} [Installed]" for v in installed_versions), None)
default_value = default_installed or next(iter(version_names), None)
return gr.Dropdown.update(choices=display_version_names, value=default_value, interactive=True) # Version List
return gr.Dropdown.update(choices=[], value=None, interactive=False) # Version List
def cleaned_name(file_name):
if platform.system() == "Windows":
illegal_chars_pattern = r'[\\/:*?"<>|]'
else:
illegal_chars_pattern = r'/'
name, extension = os.path.splitext(file_name)
clean_name = re.sub(illegal_chars_pattern, '', name)
return f"{clean_name}{extension}"
def fetch_and_process_image(image_url):
response = requests.get(image_url)
if response.status_code == 200:
image = Image.open(BytesIO(response.content))
geninfo, _ = read_info_from_image(image)
return geninfo
return None
def extract_model_info(input_string):
last_open_parenthesis = input_string.rfind("(")
last_close_parenthesis = input_string.rfind(")")
name = input_string[:last_open_parenthesis].strip()
id_number = input_string[last_open_parenthesis + 1:last_close_parenthesis]
return name, int(id_number)
def update_model_info(model_string=None, model_version=None, only_html=False, input_id=None, json_input=None, from_preview=False):
video_playback = getattr(opts, "video_playback", True)
meta_btn = getattr(opts, "individual_meta_btn", True)
playback = ""
if video_playback: playback = "autoplay loop"
if json_input:
api_data = json_input
else:
api_data = gl.json_data
BtnDownInt = True
BtnDel = False
BtnImage = False
model_id = None
if not input_id:
_, model_id = extract_model_info(model_string)
else:
model_id = input_id
if model_version and "[Installed]" in model_version:
model_version = model_version.replace(" [Installed]", "")
if model_id:
output_html = ""
output_training = ""
output_basemodel = ""
img_html = ""
dl_dict = {}
is_LORA = False
file_list = []
file_dict = []
default_file = None
model_filename = None
sha256_value = None
for item in api_data['items']:
if int(item['id']) == int(model_id):
content_type = item['type']
if content_type == "LORA":
is_LORA = True
desc = item['description']
model_name = item['name']
model_folder = os.path.join(contenttype_folder(content_type, desc))
model_uploader = item['creator']['username']
uploader_avatar = item['creator']['image']
if uploader_avatar is None:
uploader_avatar = ''
else:
uploader_avatar = f'<div class="avatar"><img src={uploader_avatar}></div>'
tags = item.get('tags', "")
model_desc = item.get('description', "")
if model_desc:
model_desc = model_desc.replace('<img', '<img style="max-width: -webkit-fill-available;"')
model_desc = model_desc.replace('<code>', '<code style="text-wrap: wrap">')
if model_version is None:
selected_version = item['modelVersions'][0]
else:
for model in item['modelVersions']:
if model['name'] == model_version:
selected_version = model
break
if selected_version['trainedWords']:
output_training = ",".join(selected_version['trainedWords'])
output_training = re.sub(r'<[^>]*:[^>]*>', '', output_training)
output_training = re.sub(r', ?', ', ', output_training)
output_training = output_training.strip(', ')
if selected_version['baseModel']:
output_basemodel = selected_version['baseModel']
for file in selected_version['files']:
dl_dict[file['name']] = file['downloadUrl']
if not model_filename:
model_filename = file['name']
dl_url = file['downloadUrl']
gl.json_info = item
sha256_value = file['hashes'].get('SHA256', 'Unknown')
size = file['metadata'].get('size', 'Unknown')
format = file['metadata'].get('format', 'Unknown')
fp = file['metadata'].get('fp', 'Unknown')
sizeKB = file.get('sizeKB', 0) * 1024
filesize = _download.convert_size(sizeKB)
unique_file_name = f"{size} {format} {fp} ({filesize})"
is_primary = file.get('primary', False)
file_list.append(unique_file_name)
file_dict.append({
"format": format,
"sizeKB": sizeKB
})
if is_primary:
default_file = unique_file_name
model_filename = file['name']
dl_url = file['downloadUrl']
gl.json_info = item
sha256_value = file['hashes'].get('SHA256', 'Unknown')
safe_tensor_found = False
pickle_tensor_found = False
if is_LORA and file_dict:
for file_info in file_dict:
file_format = file_info.get("format", "")
if "SafeTensor" in file_format:
safe_tensor_found = True
if "PickleTensor" in file_format:
pickle_tensor_found = True
if safe_tensor_found and pickle_tensor_found:
if "PickleTensor" in file_dict[0].get("format", ""):
if file_dict[0].get("sizeKB", 0) <= 100:
model_folder = os.path.join(contenttype_folder("TextualInversion"))
model_url = selected_version.get('downloadUrl', '')
model_main_url = f"https://civitai.com/models/{item['id']}"
img_html = '<div class="sampleimgs"><input type="radio" name="zoomRadio" id="resetZoom" class="zoom-radio" checked>'
for index, pic in enumerate(selected_version['images']):
meta_button = False
meta = pic['meta']
if meta and meta.get('prompt'):
meta_button = True
BtnImage = True
# Change width value in URL to original image width
image_url = re.sub(r'/width=\d+', f'/width={pic["width"]}', pic["url"])
if pic['type'] == "video":
image_url = image_url.replace("width=", "transcode=true,width=")
nsfw = 'class="model-block"'
if pic['nsfw'] not in ["None", "Soft"]:
nsfw = 'class="civnsfw model-block"'
img_html += f'''
<div {nsfw} style="display:flex;align-items:flex-start;">
<div class="civitai-image-container">
<input type="radio" name="zoomRadio" id="zoomRadio{index}" class="zoom-radio">
<label for="zoomRadio{index}" class="zoom-img-container">
'''
# Check if the pic is an image or video
if pic['type'] == "video":
img_html += f'<video data-sampleimg="true" {playback} muted playsinline><source src="{image_url}" type="video/mp4"></video>'
meta_button = False
else:
img_html += f'<img data-sampleimg="true" src="{image_url}">'
img_html += '''
</label>
<label for="resetZoom" class="zoom-overlay"></label>
'''
if meta_button:
img_html += f'''
<div class="civitai_txt2img" style="margin-top:30px;margin-bottom:30px;">
<label onclick='sendImgUrl("{escape(image_url)}")' class="civitai-txt2img-btn" style="max-width:fit-content;cursor:pointer;">Send to txt2img</label>
</div></div>
'''
else:
img_html += '</div>'
if meta:
img_html += '<div style="margin:1em 0em 1em 1em;text-align:left;line-height:1.5em;" id="image_info"><dl style="gap:10px; display:grid;">'
# Define the preferred order of keys and convert them to lowercase
preferred_order = ["prompt", "negativePrompt", "seed", "Size", "Model", "clipSkip", "sampler", "steps", "cfgScale"]
preferred_order_lower = [key.lower() for key in preferred_order]
# Loop through the keys in the preferred order and add them to the HTML
for key in preferred_order:
if key in meta:
value = meta[key]
if meta_btn:
img_html += f'<div class="civitai-meta-btn" onclick="metaToTxt2Img(\'{escape(str(key))}\', this)"><dt>{escape(str(key).capitalize())}</dt><dd>{escape(str(value))}</dd></div>'
else:
img_html += f'<div class="civitai-meta"><dt>{escape(str(key).capitalize())}</dt><dd>{escape(str(value))}</dd></div>'
# Check if there are remaining keys in meta
remaining_keys = [key for key in meta if key.lower() not in preferred_order_lower]
# Add the rest
if remaining_keys:
img_html += f"""
<div class="tabs">
<div class="tab">
<input type="checkbox" class="accordionCheckbox" id="chck{index}">
<label class="tab-label" for="chck{index}">More details...</label>
<div class="tab-content" style="gap:10px;display:grid;margin-left:1px;">
"""
for key in remaining_keys:
value = meta[key]
img_html += f'<div class="civitai-meta"><dt>{escape(str(key).capitalize())}</dt><dd>{escape(str(value))}</dd></div>'
img_html = img_html + '</div></div></div>'
img_html += '</dl></div>'
img_html = img_html + '</div>'
img_html = img_html + '</div>'
tags_html = ''.join([f'<span class="civitai-tag">{escape(str(tag))}</span>' for tag in tags])
def perms_svg(color):
return f'<span style="display:inline-block;vertical-align:middle;">'\
f'<svg width="15" height="15" viewBox="0 1.5 24 24" stroke-width="4" stroke-linecap="round" stroke="{color}">'
allow_svg = f'{perms_svg("lime")}<path d="M5 12l5 5l10 -10"></path></svg></span>'
deny_svg = f'{perms_svg("red")}<path d="M18 6l-12 12"></path><path d="M6 6l12 12"></path></svg></span>'
perms_html= '<p style="line-height: 2; font-weight: bold;">'\
f'{allow_svg if item.get("allowNoCredit") else deny_svg} Use the model without crediting the creator<br/>'\
f'{allow_svg if item.get("allowCommercialUse") in ["Image", "Rent", "RentCivit", "Sell"] else deny_svg} Sell images they generate<br/>'\
f'{allow_svg if item.get("allowCommercialUse") in ["Rent", "Sell"] else deny_svg} Run on services that generate images for money<br/>'\
f'{allow_svg if item.get("allowCommercialUse") in ["RentCivit", "Rent", "Sell"] else deny_svg} Run on Civitai<br/>'\
f'{allow_svg if item.get("allowDerivatives") else deny_svg} Share merges using this model<br/>'\
f'{allow_svg if item.get("allowCommercialUse") == "Sell" else deny_svg} Sell this model or merges using this model<br/>'\
f'{allow_svg if item.get("allowDifferentLicense") else deny_svg} Have different permissions when sharing merges'\
'</p>'
output_html = f'''
<div class="model-block">
<h2><a href={model_main_url} target="_blank" id="model_header">{escape(str(model_name))}</a></h2>
<h3 class="model-uploader">Uploaded by <a href="https://civitai.com/user/{escape(str(model_uploader))}" target="_blank">{escape(str(model_uploader))}</a>{uploader_avatar}</h3>
<div class="civitai-version-info" style="display:flex; flex-wrap:wrap; justify-content:space-between;">
<dl id="info_block">
<dt>Version</dt>
<dd>{escape(str(model_version))}</dd>
<dt>Base Model</dt>
<dd>{escape(str(output_basemodel))}</dd>
<dt>CivitAI Tags</dt>
<dd>
<div class="civitai-tags-container">
{tags_html}
</div>
</dd>
{"<dt>Download Link</dt>" if model_url else ''}
{f'<dd><a href={model_url} target="_blank">{model_url}</a></dd>' if model_url else ''}
</dl>
<div style="align-self:center; min-width:320px;">
<div>
{perms_html}
</div>
</div>
</div>
<div class="model-description" style="overflow-wrap: break-word;">
<h2>Description</h2>
{model_desc}
</div>
</div>
<div align=center>{img_html}</div>
'''
if only_html:
return output_html
folder_location = "None"
default_subfolder = "None"
sub_folders = ["None"]
for root, dirs, files in os.walk(model_folder, followlinks=True):
for filename in files:
if filename.endswith('.json'):
json_file_path = os.path.join(root, filename)
with open(json_file_path, 'r', encoding="utf-8") as f:
try:
data = json.load(f)
sha256 = data.get('sha256')
if sha256:
sha256 = sha256.upper()
if sha256 == sha256_value:
folder_location = root
BtnDownInt = False
BtnDel = True
break
except Exception as e:
print(f"Error decoding JSON: {str(e)}")
else:
for filename in files:
if filename == model_filename or filename == cleaned_name(model_filename):
folder_location = root
BtnDownInt = False
BtnDel = True
break
if folder_location != "None":
break
insert_sub_1 = getattr(opts, "insert_sub_1", False)
insert_sub_2 = getattr(opts, "insert_sub_2", False)
insert_sub_3 = getattr(opts, "insert_sub_3", False)
insert_sub_4 = getattr(opts, "insert_sub_4", False)
insert_sub_5 = getattr(opts, "insert_sub_5", False)
insert_sub_6 = getattr(opts, "insert_sub_6", False)
insert_sub_7 = getattr(opts, "insert_sub_7", False)
insert_sub_8 = getattr(opts, "insert_sub_8", False)
insert_sub_9 = getattr(opts, "insert_sub_9", False)
insert_sub_10 = getattr(opts, "insert_sub_10", False)
insert_sub_11 = getattr(opts, "insert_sub_11", False)
insert_sub_12 = getattr(opts, "insert_sub_12", False)
insert_sub_13 = getattr(opts, "insert_sub_13", False)
insert_sub_14 = getattr(opts, "insert_sub_14", False)
dot_subfolders = getattr(opts, "dot_subfolders", True)
try:
sub_folders = ["None"]
for root, dirs, _ in os.walk(model_folder, followlinks=True):
if dot_subfolders:
dirs = [d for d in dirs if not d.startswith('.')]
dirs = [d for d in dirs if not any(part.startswith('.') for part in os.path.join(root, d).split(os.sep))]
for d in dirs:
sub_folder = os.path.relpath(os.path.join(root, d), model_folder)
if sub_folder:
sub_folders.append(f'{os.sep}{sub_folder}')
sub_folders.remove("None")
sub_folders = sorted(sub_folders, key=lambda x: (x.lower(), x))
sub_folders.insert(0, "None")
base = cleaned_name(model_uploader)
author = cleaned_name(model_uploader)
name = cleaned_name(model_name)
ver = cleaned_name(model_version)
if insert_sub_1:
sub_folders.insert(1, os.path.join(os.sep, base))
if insert_sub_2:
sub_folders.insert(2, os.path.join(os.sep, base, author))
if insert_sub_3:
sub_folders.insert(3, os.path.join(os.sep, base, author, name))
if insert_sub_4:
sub_folders.insert(4, os.path.join(os.sep, base, author, name, ver))
if insert_sub_5:
sub_folders.insert(5, os.path.join(os.sep, base, name))
if insert_sub_6:
sub_folders.insert(6, os.path.join(os.sep, base, name, ver))
if insert_sub_7:
sub_folders.insert(7, os.path.join(os.sep, author))
if insert_sub_8:
sub_folders.insert(8, os.path.join(os.sep, author, base))
if insert_sub_9:
sub_folders.insert(9, os.path.join(os.sep, author, base, name))
if insert_sub_10:
sub_folders.insert(10, os.path.join(os.sep, author, base, name, ver))
if insert_sub_11:
sub_folders.insert(11, os.path.join(os.sep, author, name))
if insert_sub_12:
sub_folders.insert(12, os.path.join(os.sep, author, name, ver))
if insert_sub_13:
sub_folders.insert(13, os.path.join(os.sep, name))
if insert_sub_14:
sub_folders.insert(14, os.path.join(os.sep, name, ver))
list = set()
sub_folders = [x for x in sub_folders if not (x in list or list.add(x))]
except:
sub_folders = ["None"]
default_sub = sub_folder_value(content_type, desc)
variable_mapping = {
"Base model": base,
"Author name": author,
"Model name": name,
"Model version": ver
}
if any(key in default_sub for key in variable_mapping.keys()):
path_components = [variable_mapping.get(component.strip(os.sep), component.strip(os.sep)) for component in default_sub.split(os.sep)]
default_sub = os.path.join(*path_components)
if folder_location == "None":
folder_location = model_folder
if default_sub != "None":
folder_path = folder_location + default_sub
else:
folder_path = folder_location
else:
folder_path = folder_location
relative_path = os.path.relpath(folder_location, model_folder)
default_subfolder = f'{os.sep}{relative_path}' if relative_path != "." else default_sub if BtnDel == False else "None"
if gl.isDownloading:
item = gl.download_queue[0]
if int(model_id) == int(item['model_id']):
BtnDel = False
BtnDownTxt = "Download model"
if len(gl.download_queue) > 0:
BtnDownTxt = "Add to queue"
for item in gl.download_queue:
if item['version_name'] == model_version and int(item['model_id']) == int(model_id):
BtnDownInt = False
break
return (
gr.HTML.update(value=output_html), # Preview HTML
gr.Textbox.update(value=output_training, interactive=True), # Trained Tags
gr.Textbox.update(value=output_basemodel), # Base Model Number
gr.Button.update(visible=False if BtnDel else True, interactive=BtnDownInt, value=BtnDownTxt), # Download Button
gr.Button.update(interactive=BtnImage), # Images Button
gr.Button.update(visible=BtnDel, interactive=BtnDel), # Delete Button
gr.Dropdown.update(choices=file_list, value=default_file, interactive=True), # File List
gr.Textbox.update(value=cleaned_name(model_filename), interactive=True), # Model File Name
gr.Textbox.update(value=dl_url), # Download URL
gr.Textbox.update(value=model_id), # Model ID
gr.Textbox.update(value=sha256_value), # SHA256
gr.Textbox.update(interactive=True, value=folder_path if model_name else None), # Install Path
gr.Dropdown.update(choices=sub_folders, value=default_subfolder, interactive=True) # Sub Folder List
)
else:
return (
gr.HTML.update(value=None), # Preview HTML
gr.Textbox.update(value=None, interactive=False), # Trained Tags
gr.Textbox.update(value=''), # Base Model Number
gr.Button.update(visible=False if BtnDel else True, value="Download model"), # Download Button
gr.Button.update(interactive=False), # Images Button
gr.Button.update(visible=BtnDel, interactive=BtnDel), # Delete Button
gr.Dropdown.update(choices=None, value=None, interactive=False), # File List
gr.Textbox.update(value=None, interactive=False), # Model File Name
gr.Textbox.update(value=None), # Download URL
gr.Textbox.update(value=None), # Model ID
gr.Textbox.update(value=None), # SHA256
gr.Textbox.update(interactive=False, value=None), # Install Path
gr.Dropdown.update(choices=None, value=None, interactive=False) # Sub Folder List
)
def sub_folder_value(content_type, desc=None):
use_LORA = getattr(opts, "use_LORA", False)
if content_type in ["LORA", "LoCon"] and use_LORA:
folder = getattr(opts, "LORA_LoCon_subfolder", "None")
elif content_type == "Upscaler":
for upscale_type in ["SWINIR", "REALESRGAN", "GFPGAN", "BSRGAN"]:
if upscale_type in desc:
folder = getattr(opts, f"{upscale_type}_subfolder", "None")
folder = getattr(opts, "ESRGAN_subfolder", "None")
else:
folder = getattr(opts, f"{content_type}_subfolder", "None")
if folder == None:
return "None"
return folder
def update_file_info(model_string, model_version, file_metadata):
file_list = []
is_LORA = False
embed_check = False
model_name = None
model_id = None
model_name, model_id = extract_model_info(model_string)
if model_version and "[Installed]" in model_version:
model_version = model_version.replace(" [Installed]", "")
if model_id and model_version:
for item in gl.json_data['items']:
if int(item['id']) == int(model_id):
content_type = item['type']
if content_type == "LORA":
is_LORA = True
desc = item['description']
for model in item['modelVersions']:
if model['name'] == model_version:
for file in model['files']:
size = file['metadata'].get('size', 'Unknown')
format = file['metadata'].get('format', 'Unknown')
unique_file_name = f"{size} {format}"
file_list.append(unique_file_name)
pass
if is_LORA and file_list:
extracted_formats = [file.split(' ')[1] for file in file_list]
if "SafeTensor" in extracted_formats and "PickleTensor" in extracted_formats:
embed_check = True
for file in model['files']:
model_id = item['id']
file_name = file.get('name', 'Unknown')
sha256 = file['hashes'].get('SHA256', 'Unknown')
metadata = file.get('metadata', {})
file_size = metadata.get('size', 'Unknown')
file_format = metadata.get('format', 'Unknown')
file_fp = metadata.get('fp', 'Unknown')
sizeKB = file.get('sizeKB', 0)
sizeB = sizeKB * 1024
filesize = _download.convert_size(sizeB)
if f"{file_size} {file_format} {file_fp} ({filesize})" == file_metadata:
installed = False
folder_location = "None"
model_folder = os.path.join(contenttype_folder(content_type, desc))
if embed_check and file_format == "PickleTensor":
if sizeKB <= 100:
model_folder = os.path.join(contenttype_folder("TextualInversion"))
dl_url = file['downloadUrl']
gl.json_info = item
for root, _, files in os.walk(model_folder, followlinks=True):
if file_name in files:
installed = True
folder_location = root
break
if not installed:
for root, _, files in os.walk(model_folder, followlinks=True):
for filename in files:
if filename.endswith('.json'):
with open(os.path.join(root, filename), 'r', encoding="utf-8") as f:
try:
data = json.load(f)
sha256_value = data.get('sha256')
if sha256_value != None and sha256_value.upper() == sha256:
folder_location = root
installed = True
break
except Exception as e:
print(f"Error decoding JSON: {str(e)}")
default_sub = sub_folder_value(content_type, desc)
if folder_location == "None":
folder_location = model_folder
if default_sub != "None":
folder_path = folder_location + default_sub
else:
folder_path = folder_location
else:
folder_path = folder_location
relative_path = os.path.relpath(folder_location, model_folder)
default_subfolder = f'{os.sep}{relative_path}' if relative_path != "." else default_sub if installed == False else "None"
BtnDownInt = not installed
BtnDownTxt = "Download model"
if len(gl.download_queue) > 0:
BtnDownTxt = "Add to queue"
for item in gl.download_queue:
if item['version_name'] == model_version:
BtnDownInt = False
break
return (
gr.Textbox.update(value=cleaned_name(file['name']), interactive=True), # Model File Name Textbox
gr.Textbox.update(value=dl_url), # Download URL Textbox
gr.Textbox.update(value=model_id), # Model ID Textbox
gr.Textbox.update(value=sha256), # sha256 textbox
gr.Button.update(interactive=BtnDownInt, visible=False if installed else True, value=BtnDownTxt), # Download Button
gr.Button.update(interactive=True if installed else False, visible=True if installed else False), # Delete Button
gr.Textbox.update(interactive=True, value=folder_path if model_name else None), # Install Path
gr.Dropdown.update(value=default_subfolder, interactive=True) # Sub Folder List
)
return (
gr.Textbox.update(value=None, interactive=False), # Model File Name Textbox
gr.Textbox.update(value=None), # Download URL Textbox
gr.Textbox.update(value=None), # Model ID Textbox
gr.Textbox.update(value=None), # sha256 textbox
gr.Button.update(interactive=False, visible=True), # Download Button
gr.Button.update(interactive=False, visible=False), # Delete Button
gr.Textbox.update(interactive=False, value=None), # Install Path
gr.Dropdown.update(choices=None, value=None, interactive=False) # Sub Folder List
)
def get_headers():
api_key = getattr(opts, "custom_api_key", "")
try:
user_agent = UserAgent().chrome
except ImportError:
user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/117.0.0.0 Safari/537.36"
headers = {
'User-Agent': user_agent,
'Sec-Ch-Ua': '"Brave";v="119", "Chromium";v="119", "Not?A_Brand";v="24"',
'Sec-Ch-Ua-Mobile': '?0',
'Sec-Ch-Ua-Platform': '"Windows"',
'Sec-Fetch-Dest': 'document',
'Sec-Fetch-Mode': 'navigate',
'Sec-Fetch-Site': 'none',
'Sec-Fetch-User': '?1',
'Sec-Gpc': '1',
'Upgrade-Insecure-Requests': '1',
}
if api_key:
headers['Authorization'] = f'Bearer {api_key}'
return headers
def request_civit_api(api_url=None):
headers = get_headers()
try:
response = requests.get(api_url, headers=headers, timeout=(10, 30))
response.raise_for_status()
except requests.exceptions.RequestException as e:
print(f"Error: {e}")
return "timeout"
else:
response.encoding = "utf-8"
try:
data = json.loads(response.text)
except json.JSONDecodeError:
print("The CivitAI servers are currently offline. Please try again later.")
return "timeout"
return data