# -*- coding: utf-8 -*- """Test_gradio_push.ipynb Automatically generated by Colaboratory. Original file is located at https://colab.research.google.com/drive/1mlZpAq-EWRmmLHH4Ok533awreqtJwzzW """ """# HF Script """ # -*- coding: utf-8 -*- """Copy of Anime_Pack_Gradio.ipynb Automatically generated by Colaboratory. Original file is located at https://colab.research.google.com/drive/1RxVCwOkq3Q5qlEkQxhFGeUxICBujjEjR """ import os from transformers import AutoTokenizer, AutoModelForSeq2SeqLM tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-zh-en") model = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-zh-en") import gradio as gr import numpy as np from PIL import Image from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, DPMSolverMultistepScheduler, StableDiffusionImg2ImgPipeline import torch from controlnet_aux import HEDdetector from diffusers.utils import load_image import concurrent.futures from threading import Thread from compel import Compel from transformers import pipeline model_ckpt = "papluca/xlm-roberta-base-language-detection" pipe = pipeline("text-classification", model=model_ckpt) HF_TOKEN = os.environ.get("HUGGING_FACE_HUB_TOKEN") device="cuda" if torch.cuda.is_available() else "cpu" pipe_scribble, pipe_depth, pipe_img2img = None, None, None hidden_booster_text = "masterpiece++, best quality++, ultra-detailed+ +, unity 8k wallpaper+, illustration+, anime style+, intricate, fluid simulation, sharp edges. glossy++, Smooth++, detailed eyes++, best quality++,4k++,8k++,highres++,masterpiece++,ultra- detailed,realistic++,photorealistic++,photo-realistic++,depth of field, ultra-high definition, highly detailed, natural lighting, sharp focus, cinematic, hyperrealism,extremely detailed" hidden_negative = "bad anatomy, disfigured, poorly drawn,deformed, mutation, malformation, deformed, mutated, disfigured, deformed eyes+, bad face++, bad hands, poorly drawn hands, malformed hands, extra arms++, extra legs++, Fused body+, Fused hands+, Fused legs+, missing arms, missing limb, extra digit+, fewer digits, floating limbs, disconnected limbs, inaccurate limb, bad fingers, missing fingers, ugly face, long body++" hidden_cn_booster_text = ",漂亮的脸" hidden_cn_negative = "" hed = HEDdetector.from_pretrained('lllyasviel/ControlNet') controlnet_scribble = ControlNetModel.from_pretrained( "lllyasviel/sd-controlnet-scribble", torch_dtype=torch.float16, safety_checker=None, requires_safety_checker=False, ) depth_estimator = pipeline('depth-estimation') controlnet_depth = ControlNetModel.from_pretrained( "lllyasviel/sd-controlnet-depth", torch_dtype=torch.float16 ) def translate(prompt): trans_text = prompt translated = model.generate(**tokenizer(trans_text, return_tensors="pt", padding=True)) tgt_text = [tokenizer.decode(t, skip_special_tokens=True) for t in translated] tgt_text = ''.join(tgt_text)[:-1] return tgt_text def load_pipe_scribble(): global pipe_scribble if pipe_scribble is None: pipe_scribble = StableDiffusionControlNetPipeline.from_single_file( "https://huggingface.co/shellypeng/anime-god/blob/main/animeGod_v10.safetensors", controlnet=controlnet_scribble, safety_checker=None, requires_safety_checker=False, torch_dtype=torch.float16, token=HF_TOKEN ) pipe_scribble.load_lora_weights("shellypeng/lora2") pipe_scribble.fuse_lora(lora_scale=0.1) pipe_scribble.load_textual_inversion("shellypeng/textinv1") pipe_scribble.load_textual_inversion("shellypeng/textinv2") pipe_scribble.load_textual_inversion("shellypeng/textinv3") pipe_scribble.load_textual_inversion("shellypeng/textinv4") pipe_scribble.scheduler = DPMSolverMultistepScheduler.from_config(pipe_scribble.scheduler.config, use_karras_sigmas=True) pipe_scribble.safety_checker = None pipe_scribble.requires_safety_checker = False pipe_scribble.to(device) pipe_scribble.safety_checker = lambda images, **kwargs: (images, [False] * len(images)) def load_pipe_depth(): global pipe_depth if pipe_depth is None: pipe_depth = StableDiffusionControlNetPipeline.from_single_file( "https://huggingface.co/shellypeng/anime-god/blob/main/animeGod_v10.safetensors", controlnet=controlnet_depth, torch_dtype=torch.float16, ) pipe_depth.load_lora_weights("shellypeng/lora1") pipe_depth.fuse_lora(lora_scale=0.3) pipe_depth.load_textual_inversion("shellypeng/textinv1") pipe_depth.load_textual_inversion("shellypeng/textinv2") pipe_depth.load_textual_inversion("shellypeng/textinv3") pipe_depth.load_textual_inversion("shellypeng/textinv4") pipe_depth.scheduler = DPMSolverMultistepScheduler.from_config(pipe_depth.scheduler.config, use_karras_sigmas=True) def dummy(images, **kwargs): return images, False pipe_depth.safety_checker = lambda images, **kwargs: (images, [False] * len(images)) pipe_depth.to(device) def load_pipe_img2img(): global pipe_img2img if pipe_img2img is None: pipe_img2img = StableDiffusionImg2ImgPipeline.from_single_file("https://huggingface.co/shellypeng/anime-god/blob/main/animeGod_v10.safetensors", torch_dtype=torch.float16, safety_checker=None, requires_safety_checker=False, token=HF_TOKEN) pipe_img2img.load_lora_weights("shellypeng/lora1") pipe_img2img.fuse_lora(lora_scale=0.1) pipe_img2img.load_lora_weights("shellypeng/lora2", token=HF_TOKEN) pipe_img2img.fuse_lora(lora_scale=0.1) pipe_img2img.load_textual_inversion("shellypeng/textinv1") pipe_img2img.load_textual_inversion("shellypeng/textinv2") pipe_img2img.load_textual_inversion("shellypeng/textinv3") pipe_img2img.load_textual_inversion("shellypeng/textinv4") pipe_img2img.scheduler = DPMSolverMultistepScheduler.from_config(pipe_img2img.scheduler.config, use_karras_sigmas=True) pipe_img2img.safety_checker = None pipe_img2img.requires_safety_checker = False pipe_img2img.to(device) pipe_img2img.safety_checker = lambda images, **kwargs: (images, [False] * len(images)) def real_to_anime(text, input_img): """ pass the sd model and do scribble to image include Adetailer, detail tweaker lora, prompt backend include: beautiful eyes, beautiful face, beautiful hand, (maybe infer from user's prompt for gesture and facial expression to improve hand) """ load_pipe_depth() input_img = Image.fromarray(input_img) input_img = load_image(input_img) input_img = depth_estimator(input_img)['depth'] res_image0 = pipe_depth(text, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] res_image1 = pipe_depth(text, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] res_image2 = pipe_depth(text, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] res_image3 = pipe_depth(text, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] return res_image0, res_image1, res_image2, res_image3 def scribble_to_image(text, neg_prompt_box, input_img): """ pass the sd model and do scribble to image include Adetailer, detail tweaker lora, prompt backend include: beautiful eyes, beautiful face, beautiful hand, (maybe infer from user's prompt for gesture and facial expression to improve hand) """ load_pipe_scribble() # if auto detect detects chinese => auto turn on chinese prompting checkbox # change param "bag" below to text, image param below to input_img input_img = Image.fromarray(input_img) input_img = hed(input_img, scribble=True) input_img = load_image(input_img) # global prompt lang_check_label = pipe(text, top_k=1, truncation=True)[0]['label'] lang_check_score = pipe(text, top_k=1, truncation=True)[0]['score'] if lang_check_label == 'zh' and lang_check_score >= 0.85: text = translate(text) compel_proc = Compel(tokenizer=pipe_scribble.tokenizer, text_encoder=pipe_scribble.text_encoder) prompt = text + hidden_booster_text prompt_embeds = compel_proc(prompt) negative_prompt = neg_prompt_box + hidden_negative negative_prompt_embeds = compel_proc(negative_prompt) res_image0 = pipe_scribble(image=input_img, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image1 = pipe_scribble(image=input_img, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image2 = pipe_scribble(image=input_img, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image3 = pipe_scribble(image=input_img, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] return res_image0, res_image1, res_image2, res_image3 def real_img2img_to_anime(text, neg_prompt_box, input_img): """ pass the sd model and do scribble to image include Adetailer, detail tweaker lora, prompt backend include: beautiful eyes, beautiful face, beautiful hand, (maybe infer from user's prompt for gesture and facial expression to improve hand) """ load_pipe_img2img() input_img = Image.fromarray(input_img) input_img = load_image(input_img) lang_check_label = pipe(text, top_k=1, truncation=True)[0]['label'] lang_check_score = pipe(text, top_k=1, truncation=True)[0]['score'] if lang_check_label == 'zh' and lang_check_score >= 0.85: text = translate(text) compel_proc = Compel(tokenizer=pipe_img2img.tokenizer, text_encoder=pipe_img2img.text_encoder) prompt = text + hidden_booster_text prompt_embeds = compel_proc(prompt) negative_prompt = neg_prompt_box + hidden_negative negative_prompt_embeds = compel_proc(negative_prompt) # input_img = depth_estimator(input_img)['depth'] res_image0 = pipe_img2img(image=input_img, strength=0.8, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image1 = pipe_img2img(image=input_img, strength=0.8, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image2 = pipe_img2img(image=input_img, strength=0.8, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] res_image3 = pipe_img2img(image=input_img, strength=0.8, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] return res_image0, res_image1, res_image2, res_image3 theme = gr.themes.Soft( primary_hue="orange", secondary_hue="orange", ).set( block_background_fill='*primary_50' ) def zh_prompt_info(text, neg_text, chinese_check): can_raise_info = "" lang_check_label = pipe(text, top_k=1, truncation=True)[0]['label'] lang_check_score = pipe(text, top_k=1, truncation=True)[0]['score'] neg_lang_check_label = pipe(neg_text, top_k=1, truncation=True)[0]['label'] neg_lang_check_score = pipe(neg_text, top_k=1, truncation=True)[0]['score'] print(lang_check_label) if lang_check_label == 'zh' and lang_check_score >= 0.85: if not chinese_check: chinese_check = True can_raise_info = "zh" if neg_lang_check_label == 'en' and neg_lang_check_score >= 0.85: can_raise_info = "invalid" return True, can_raise_info elif lang_check_label == 'en' and lang_check_score >= 0.85: if chinese_check: chinese_check = False can_raise_info = "en" if neg_lang_check_label == 'zh' and neg_lang_check_score >= 0.85: can_raise_info = "invalid" return False, can_raise_info return chinese_check, can_raise_info def mult_thread_img2img(prompt_box, neg_prompt_box, image_box): with concurrent.futures.ThreadPoolExecutor(max_workers=12000) as executor: future = executor.submit(real_img2img_to_anime, prompt_box, neg_prompt_box, image_box) image1, image2, image3, image4 = future.result() return image1, image2, image3, image4 def mult_thread_scribble(prompt_box, neg_prompt_box, image_box): with concurrent.futures.ThreadPoolExecutor(max_workers=12000) as executor: future = executor.submit(scribble_to_image, prompt_box, neg_prompt_box, image_box) image1, image2, image3, image4 = future.result() return image1, image2, image3, image4 def mult_thread_live_scribble(prompt_box, neg_prompt_box, image_box): image_box = image_box["composite"] with concurrent.futures.ThreadPoolExecutor(max_workers=12000) as executor: future = executor.submit(scribble_to_image, prompt_box, neg_prompt_box, image_box) image1, image2, image3, image4 = future.result() return image1, image2, image3, image4 def mult_thread_lang_class(prompt_box, neg_prompt_box, chinese_check): with concurrent.futures.ThreadPoolExecutor(max_workers=12000) as executor: future = executor.submit(zh_prompt_info, prompt_box, neg_prompt_box, chinese_check) chinese_check, can_raise_info = future.result() if can_raise_info == "zh": gr.Info("Chinese Language Detected, Switching to Chinese Prompt Mode") elif can_raise_info == "en": gr.Info("English Language Detected, Disabling Chinese Prompt Mode") return chinese_check with gr.Blocks(theme=theme, css="footer {visibility: hidden}", title="ShellAI Apps") as iface: with gr.Tab("AnimeDepth（安妮深度）"): gr.Markdown( """ # AnimeDepth（安妮深度） Turns pictures into one in the anime style with depth-to-image controlnet. 将图片用深度图的方式转为动漫风图片。 """ ) with gr.Row(equal_height=True): with gr.Column(): with gr.Row(equal_height=True): with gr.Column(scale=4): prompt_box = gr.Textbox(label="Prompt（提示词）", placeholder="Enter a prompt\n输入提示词", lines=3) neg_prompt_box = gr.Textbox(label="Negative Prompt（负面提示词）", placeholder="Enter a negative prompt(things you don't want to include in the generated image)\n输入负面提示词：输入您不想生成的部分", lines=3) with gr.Row(equal_height=True): chinese_check = gr.Checkbox(label="Chinese Prompt Mode（中文提示词模式）", info="Click here to enable Chinese Prompting（点此触发中文提示词输入）") image_box = gr.Image(label="Input Image（上传图片）", height=400) gen_btn = gr.Button(value="Generate（生成）") with gr.Row(equal_height=True): image1 = gr.Image(label="Result 1（结果图 1）") image2 = gr.Image(label="Result 2（结果图 2）") image3 = gr.Image(label="Result 3（结果图 3）") image4 = gr.Image(label="Result 4（结果图 4）") example_img2img = [ ["漂亮的女孩，微笑，长发，黑发，粉色外套，白色内衬，优雅，红色背景，红色窗帘", "低画质", "sunmi.jpg"], ["Beautiful girl, smiling, bun, bun hair, black hair, beautiful eyes, black dress, elegant, red carpet photo","ugly, bad quality", "emma.jpg"] ] # gr.Examples(examples=example_img2img, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4], fn=mult_thread_img2img, cache_examples=True) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_lang_class, inputs=[prompt_box, neg_prompt_box, chinese_check], outputs=[chinese_check], show_progress=False) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=real_to_anime, inputs=[prompt_box, image_box], outputs=[image1, image2, image3, image4]) with gr.Tab("Animefier（安妮漫风）"): gr.Markdown( """ # Animefier（安妮漫风） Turns realistic photos into one in the anime style. 将真实图片转为动漫风图片。 """ ) with gr.Row(equal_height=True): with gr.Column(): with gr.Row(equal_height=True): with gr.Column(scale=4): prompt_box = gr.Textbox(label="Prompt（提示词）", placeholder="Enter a prompt\n输入提示词", lines=3) neg_prompt_box = gr.Textbox(label="Negative Prompt（负面提示词）", placeholder="Enter a negative prompt(things you don't want to include in the generated image)\n输入负面提示词：输入您不想生成的部分", lines=3) with gr.Row(equal_height=True): chinese_check = gr.Checkbox(label="Chinese Prompt Mode（中文提示词模式）", info="Click here to enable Chinese Prompting（点此触发中文提示词输入）") image_box = gr.Image(label="Input Image（上传图片）", height=400) gen_btn = gr.Button(value="Generate（生成）") with gr.Row(equal_height=True): image1 = gr.Image(label="Result 1（结果图 1）") image2 = gr.Image(label="Result 2（结果图 2）") image3 = gr.Image(label="Result 3（结果图 3）") image4 = gr.Image(label="Result 4（结果图 4）") example_img2img = [ ["漂亮的女孩，微笑，长发，黑发，粉色外套，白色内衬，优雅，红色背景，红色窗帘", "低画质", "sunmi.jpg"], ["Beautiful girl, smiling, bun, bun hair, black hair, beautiful eyes, black dress, elegant, red carpet photo","ugly, bad quality", "emma.jpg"] ] # gr.Examples(examples=example_img2img, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4], fn=mult_thread_img2img, cache_examples=True) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_lang_class, inputs=[prompt_box, neg_prompt_box, chinese_check], outputs=[chinese_check], show_progress=False) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_img2img, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4]) with gr.Tab("Live Sketch（实时涂鸦）"): gr.Markdown( """ # Live Sketch（实时涂鸦） Live draw sketches/scribbles and turns into one in the anime style. 实时涂鸦，将粗线条涂鸦转为动漫风图片。 """ ) with gr.Row(equal_height=True): with gr.Column(): with gr.Row(equal_height=True): with gr.Column(scale=4): prompt_box = gr.Textbox(label="Prompt（提示词）", placeholder="Enter a prompt\n输入提示词", lines=3) neg_prompt_box = gr.Textbox(label="Negative Prompt（负面提示词）", placeholder="Enter a negative prompt(things you don't want to include in the generated image)\n输入负面提示词：输入您不想生成的部分", lines=3) with gr.Row(equal_height=True): chinese_check = gr.Checkbox(label="Chinese Prompt Mode（中文提示词模式）", info="Click here to enable Chinese Prompting（点此触发中文提示词输入）") image_box = gr.ImageEditor(sources=(), brush=gr.Brush(default_size="5", color_mode="fixed", colors=["#000000"]), height=400) gen_btn = gr.Button(value="Generate（生成）") with gr.Row(equal_height=True): image1 = gr.Image(label="Result 1（结果图 1）") image2 = gr.Image(label="Result 2（结果图 2）") image3 = gr.Image(label="Result 3（结果图 3）") image4 = gr.Image(label="Result 4（结果图 4）") # sketch_image_box.change(fn=mult_thread_scribble, inputs=[prompt_box, neg_prompt_box, sketch_image_box], outputs=[image1, image2, image3, image4]) example_scribble_live2img = [ ["帅气的男孩，橙色头发++，皱眉，闭眼，深蓝色开襟毛衣，白色内衬，酷，冷漠，帅气，硝烟背景", "劣质", "sketch_boy.png"], ["a beautiful girl spreading her arms, blue hair, long hair, hat with flowers on its edge, smiling++, dynamic, black dress, park background, birds, trees, flowers, grass","ugly, worst quality", "girl_spread.jpg"] ] # gr.Examples(examples=example_scribble_live2img, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4], fn=mult_thread_live_scribble, cache_examples=True) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_lang_class, inputs=[prompt_box, neg_prompt_box, chinese_check], outputs=[chinese_check], show_progress=False) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_live_scribble, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4]) with gr.Tab("AniSketch（安妮涂鸦）"): gr.Markdown( """ # AniSketch（安妮涂鸦） Turns sketches/scribbles into one in the anime style. 将草图、粗线条涂鸦转为动漫风图片。 """ ) with gr.Row(equal_height=True): with gr.Column(): with gr.Row(equal_height=True): with gr.Column(scale=4): prompt_box = gr.Textbox(label="Prompt（提示词）", placeholder="Enter a prompt\n输入提示词", lines=3) neg_prompt_box = gr.Textbox(label="Negative Prompt（负面提示词）", placeholder="Enter a negative prompt(things you don't want to include in the generated image)\n输入负面提示词：输入您不想生成的部分", lines=3) with gr.Row(equal_height=True): chinese_check = gr.Checkbox(label="Chinese Prompt Mode（中文提示词模式）", info="Click here to enable Chinese Prompting（点此触发中文提示词输入）") image_box = gr.Image(label="Input Image（上传图片）", height=400) gen_btn = gr.Button(value="Generate（生成）") with gr.Row(equal_height=True): image1 = gr.Image(label="Result 1（结果图 1）") image2 = gr.Image(label="Result 2（结果图 2）") image3 = gr.Image(label="Result 3（结果图 3）") image4 = gr.Image(label="Result 4（结果图 4）") example_scribble2img = [ ["漂亮的女人，散开的长发，巫师，巫师袍，微笑，拍手，优雅，成熟，月夜背景", "水印", "final_witch.jpg"], ["a man wearing a chinese clothes, closed eyes, handsome face, dragon on the clothes, expressionless face, indifferent, chinese building background","poor quality", "chinese_man.jpg"] ] # gr.Examples(examples=example_scribble2img, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4], fn=mult_thread_scribble, cache_examples=True) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_lang_class, inputs=[prompt_box, neg_prompt_box, chinese_check], outputs=[chinese_check], show_progress=False) gr.on(triggers=[prompt_box.submit, gen_btn.click],fn=mult_thread_scribble, inputs=[prompt_box, neg_prompt_box, image_box], outputs=[image1, image2, image3, image4]) def run(): iface.queue(default_concurrency_limit=20).launch(debug=True, share=True) run() """# Separator """