import os
import datetime
import toml
import gradio as gr
from PIL import Image, PngImagePlugin
from huggingface_hub import HfApi, snapshot_download
from pnginfo import read_info_from_image, send_paras
from images_history import img_history_ui
from utils import set_token, generate_novelai_image, image_from_bytes
client_config = toml.load("config.toml")['client']
today_count = 0
today = datetime.date.today().strftime('%Y-%m-%d')
api = HfApi()
def get_count():
global today_count, today
now = datetime.date.today().strftime('%Y-%m-%d')
if now != today:
today = now
today_count = 0
return today_count
def control_ui():
prompt = gr.TextArea(value="{breast expansion},{gigantic breasts},[artist:ningen_mame],artist:ciloranko,[artist:sho_(sho_lwlw)],[artist:foifoi (marfoyfoyfoy)],1girl,skinny,narrow waist,loli,", elem_id='txt2img_prompt', label="提示词", lines=3)
quality_tags = gr.TextArea(
elem_id='txt2img_qua_prompt', label="质量词", lines=1,
value=client_config['default_quality'],
)
neg_prompt = gr.TextArea(
elem_id='txt2img_neg_prompt', label="负面词", lines=1,
value=client_config['default_neg'],
)
with gr.Row():
sampler = gr.Dropdown(
choices=[
"k_euler", "k_euler_ancestral", "k_dpmpp_2s_ancestral",
"k_dpmpp_2m", "k_dpmpp_sde", "ddim_v3"
],
value="k_euler",
label="采样器",
interactive=True
)
scale = gr.Slider(label="CFG Scale", value=5.0, minimum=1, maximum=10, step=0.1)
steps = gr.Slider(label="步数", value=28, minimum=1, maximum=28, step=1)
with gr.Row():
seed = gr.Number(label="种子", value=-1, step=1, maximum=2**32-1, minimum=-1, scale=3)
rand_seed = gr.Button('🎲️', scale=1)
reuse_seed = gr.Button('♻️', scale=1)
with gr.Row():
width = gr.Slider(label="宽度", value=1024, minimum=64, maximum=2048, step=64)
height = gr.Slider(label="高度", value=1024, minimum=64, maximum=2048, step=64)
with gr.Row():
with gr.Column():
with gr.Accordion('风格迁移', open=False):
ref_image = gr.Image(label="上传图片", value=None, sources=["upload", "clipboard", "webcam"], interactive=True, type="pil")
info_extract = gr.Slider(label='参考信息提取', value=1, minimum=0, maximum=1, step=0.1)
ref_str = gr.Slider(label='参考强度', value=0.6, minimum=0, maximum=1, step=0.1)
reuse_img_vibe = gr.Button(value='使用上一次生成的图片')
with gr.Accordion('附加输入', open=False):
with gr.Tab('图生图') as i2i:
i2i_image = gr.Image(label="上传图片", value=None, sources=["upload", "clipboard", "webcam"], interactive=True, type="pil")
i2i_str = gr.Slider(label='去噪强度', value=0.7, minimum=0, maximum=0.99, step=0.01)
i2i_noise = gr.Slider(label='噪声', value=0, minimum=0, maximum=1, step=0.1)
reuse_img_i2i = gr.Button(value='使用上一次生成的图片')
with gr.Tab('局部重绘') as inp:
overlay = gr.Checkbox(label='覆盖原图', value=True)
inp_img = gr.ImageMask(label="上传图片", value=None, sources=["upload", "clipboard", "webcam"], interactive=True, type="pil", eraser=False, transforms=None, brush=gr.Brush(colors=['#FFFFFF'], color_mode='fixed'), layers=False)
reuse_img_inp = gr.Button(value='使用上一次生成的图片')
selection = gr.Radio(choices=['i2i', 'inp'], value='i2i', visible=False)
with gr.Row():
with gr.Column():
with gr.Accordion('高级选项', open=False):
scheduler = gr.Dropdown(
choices=[
"native", "karras", "exponential", "polyexponential"
],
value="native",
label="Scheduler",
interactive=True
)
with gr.Row():
smea = gr.Checkbox(False, label="SMEA")
dyn = gr.Checkbox(False, label="SMEA DYN")
with gr.Row():
dyn_threshold = gr.Checkbox(False, label="Dynamic Thresholding")
cfg_rescale = gr.Slider(0, 1, 0, step=0.01, label="CFG rescale")
with gr.Column():
save = gr.Checkbox(value=True, label='云端保存图片')
gen_btn = gr.Button(value="生成", variant="primary")
rand_seed.click(fn=lambda: -1, inputs=None, outputs=seed)
width.change(lambda w, h: h if w*h<=1024*1024 else (1024*1024//w//64)*64, [width, height], height)
height.change(lambda w, h: w if w*h<=1024*1024 else (1024*1024//h//64)*64, [width, height], width)
i2i.select(lambda: 'i2i', inputs=None, outputs=selection)
inp.select(lambda: 'inp', inputs=None, outputs=selection)
return gen_btn, [prompt, quality_tags, neg_prompt, seed, scale, width, height, steps, sampler, scheduler, smea, dyn, dyn_threshold, cfg_rescale, ref_image, info_extract, ref_str, i2i_image, i2i_str, i2i_noise, overlay, inp_img, selection], [save, rand_seed, reuse_seed, reuse_img_vibe, reuse_img_i2i, reuse_img_inp]
def generate(prompt, quality_tags, neg_prompt, seed, scale, width, height, steps, sampler, scheduler, smea, dyn, dyn_threshold, cfg_rescale, ref_image, info_extract, ref_str, i2i_image, i2i_str, i2i_noise, overlay, inp_img, selection, save):
global today_count
set_token(os.environ.get('token'))
img_data, payload = generate_novelai_image(
f"{prompt}, {quality_tags}", neg_prompt, seed, scale,
width, height, steps, sampler, scheduler,
smea, dyn, dyn_threshold, cfg_rescale, ref_image, info_extract, ref_str,
i2i_image, i2i_str, i2i_noise, overlay, inp_img, selection
)
if not isinstance(img_data, bytes):
return gr.Image(value=None), payload
today_count = get_count() + 1
img = image_from_bytes(img_data)
if save:
save_path = client_config['save_path']
today = datetime.date.today().strftime('%Y-%m-%d')
today_path = os.path.join(save_path, today)
os.makedirs(today_path, exist_ok=True)
filename = str(today_count).rjust(5, '0') + '-' + str(payload['parameters']['seed']) + '.png'
pnginfo_data = PngImagePlugin.PngInfo()
for k, v in img.info.items():
pnginfo_data.add_text(k, str(v))
img.save(os.path.join(today_path, filename), pnginfo=pnginfo_data)
api.upload_folder(folder_path=today_path, path_in_repo=today_path, repo_id="P01yH3dr0n/naimages", repo_type="dataset", token=os.environ.get("hf_token"))
return img, payload
def preview_ui():
with gr.Blocks(css='#preview_image { height: 100%;}'):
image = gr.Image(elem_id='preview_image', interactive=False, type='pil')
info = gr.JSON(value={}, label="生成信息")
return image, info
def main_ui():
with gr.Blocks() as page:
with gr.Row(variant="panel"):
with gr.Column():
gen_btn, paras, others = control_ui()
with gr.Column():
image, info = preview_ui()
gen_btn.click(generate, paras + [others[0]], [image, info], concurrency_limit=1)
others[2].click(lambda o, s: o if len(s) == 0 else s['parameters']['seed'], inputs=[paras[3], info], outputs=paras[3])
others[3].click(lambda i: i, inputs=image, outputs=paras[14])
others[4].click(lambda i: i, inputs=image, outputs=paras[17])
others[5].click(lambda i: gr.ImageEditor(value={'background': i, 'layers': [Image.new('RGBA', i.size)], 'composite': i}), inputs=image, outputs=paras[21])
return page, paras[:14]
def util_ui():
with gr.Blocks(analytics_enabled=False) as page:
with gr.Row(equal_height=False):
with gr.Column(variant='panel'):
image = gr.Image(label="上传图片", sources=["upload"], interactive=True, type="pil")
with gr.Column(variant='panel'):
info = gr.HTML()
items = gr.JSON(value=None, visible=False)
png2main = gr.Button('参数发送到文生图')
return page, png2main, items, info, image
def load_javascript():
head = ''
for f in sorted(os.listdir('./tagcomplete/javascript')):
head += f'\n'
head += '\n'
share = gr.routes.templates.TemplateResponse
def template_response(*args, **kwargs):
res = share(*args, **kwargs)
res.body = res.body.replace(b'', f'{head}'.encode("utf8"))
res.init_headers()
return res
gr.routes.templates.TemplateResponse = template_response
def ui():
load_javascript()
with gr.Blocks(title="NAI Client", theme=gr.themes.Base(), js="() => {document.body.classList.toggle('dark');}") as website:
with gr.Tabs():
with gr.TabItem("图片生成", elem_id="client_ui_main"):
_, paras = main_ui()
with gr.TabItem("图片信息读取"):
_, png2main, png_items, info, image = util_ui()
with gr.TabItem("云端图片浏览") as tab:
gal2main, gal_items = img_history_ui(tab)
with gr.TabItem("设置"):
switchLightDark = gr.Button(value="切换浅色/深色模式")
switchLightDark.click(fn=None, js="() => {document.body.classList.toggle('dark');}")
loadTagComplete = gr.Button(value="重新加载tag补全")
loadTagComplete.click(fn=None, js="() => {run();}")
clearTagCache = gr.Button(value="清除tag补全缓存")
clearTagCache.click(fn=None, js="() => {localStorage.clear();}")
png2main.click(fn=send_paras,
inputs=[png_items] + paras,
outputs=paras)
png2main.click(fn=None,
js="(x) => { if (x !== null) document.getElementById('client_ui_main-button').click(); return null; }",
inputs=image)
gal2main.click(fn=send_paras,
inputs=[gal_items] + paras,
outputs=paras)
gal2main.click(fn=None,
js="(x) => { if (x !== null) document.getElementById('client_ui_main-button').click(); return null; }",
inputs=gal_items)
image.change(read_info_from_image, inputs=image, outputs=[info, png_items])
return website
if __name__ == '__main__':
snapshot_download(repo_id="P01yH3dr0n/naimages", repo_type="dataset", local_dir="./", token=os.environ.get("hf_token"), allow_patterns="*.png")
website = ui()
website.queue(default_concurrency_limit=5)
website.launch(auth=(os.environ.get('account'), os.environ.get('password')), allowed_paths=['tagcomplete'], debug=True)