Spaces:
Runtime error
Runtime error
| # -*- 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 | |
| """ | |
| 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) | |
| device="cuda" if torch.cuda.is_available() else "cpu" | |
| hidden_booster_text = ", loraeyes, beautiful face, small boobs, a cup" | |
| hidden_negative = "big boobs, huge boobs, sexy, dirty, d cup, e cup, g cup, slutty, badhandv4, ng_deepnegative_v1_75t, worst quality, low quality, extra digits, text, signature, bad anatomy, mutated hand, error, missing finger, cropped, worse quality, bad quality, lowres, floating limbs, bad hands, anatomical nonsense" | |
| hidden_cn_booster_text = ",loraeyes漂亮的脸,小胸,贫乳,a罩杯" | |
| hidden_cn_negative = "大胸, ，, ！, 。, ；,巨乳,性感,脏,d罩杯,e罩杯,g罩杯,骚,骚气,badhandv4, ng_deepnegative_v1_75t" | |
| 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 | |
| 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, ) | |
| 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, | |
| ) | |
| # pipe.load_lora_weights("shellypeng/detail-tweaker") | |
| # pipe.load_lora_weights("shellypeng/midjourney-anime") | |
| # pipe.load_lora_weights("shellypeng/animetarot") | |
| # pipe.load_lora_weights("shellypeng/anime-stickers-v3") | |
| # pipe.load_lora_weights("shellypeng/anime-magazine") | |
| # pipe_img2img.load_lora_weights("yenojunie/slit-pupils") | |
| # pipe_scribble.load_lora_weights("shellypeng/detail-tweaker") | |
| # pipe_scribble.fuse_lora(lora_scale=0.1) | |
| pipe_scribble.load_lora_weights("shellypeng/lora-eyes") | |
| pipe_scribble.fuse_lora(lora_scale=0.1) | |
| # pipe_scribble.load_lora_weights("shellypeng/beautiful-eyes") | |
| # pipe_scribble.fuse_lora(lora_scale=0.1) | |
| pipe_scribble.load_textual_inversion("shellypeng/bad-prompt") | |
| pipe_scribble.load_textual_inversion("shellypeng/badhandv4") | |
| # pipe.load_textual_inversion("shellypeng/easynegative") | |
| pipe_scribble.load_textual_inversion("shellypeng/deepnegative") | |
| pipe_scribble.load_textual_inversion("shellypeng/verybadimagenegative") | |
| pipe_scribble.scheduler = DPMSolverMultistepScheduler.from_config(pipe_scribble.scheduler.config, use_karras_sigmas=True) | |
| # pipe.enable_model_cpu_offload() | |
| pipe_scribble.safety_checker = None | |
| pipe_scribble.requires_safety_checker = False | |
| pipe_scribble.to(device) | |
| 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) | |
| """ | |
| # 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) | |
| """ | |
| 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_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) | |
| # input_img = depth_estimator(input_img)['depth'] | |
| res_image0 = pipe_img2img(image=input_img, strength=0.6, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] | |
| res_image1 = pipe_img2img(image=input_img, strength=0.6, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] | |
| res_image2 = pipe_img2img(image=input_img, strength=0.6, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, num_inference_steps=40).images[0] | |
| res_image3 = pipe_img2img(image=input_img, strength=0.6, 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' | |
| ) | |
| # %cd /content/drive/MyDrive/stable-diffusion-webui-colab/stable-diffusion-webui | |
| 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) | |
| # pipe_img2img.load_lora_weights("shellypeng/detail-tweaker") | |
| # pipe_img2img.fuse_lora(lora_scale=0.1) | |
| pipe_img2img.load_lora_weights("shellypeng/lora-eyes") | |
| pipe_img2img.fuse_lora(lora_scale=0.1) | |
| # pipe_img2img.load_lora_weights("shellypeng/beautiful-eyes") | |
| # pipe_img2img.fuse_lora(lora_scale=0.1) | |
| pipe_img2img.load_textual_inversion("shellypeng/bad-prompt") | |
| pipe_img2img.load_textual_inversion("shellypeng/badhandv4") | |
| # pipe.load_textual_inversion("shellypeng/easynegative") | |
| pipe_img2img.load_textual_inversion("shellypeng/deepnegative") | |
| pipe_img2img.load_textual_inversion("shellypeng/verybadimagenegative") | |
| pipe_img2img.scheduler = DPMSolverMultistepScheduler.from_config(pipe_img2img.scheduler.config, use_karras_sigmas=True) | |
| # pipe.enable_model_cpu_offload() | |
| 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)) | |
| # depth_estimator = pipeline('depth-estimation') | |
| # controlnet_depth = ControlNetModel.from_pretrained( | |
| # "lllyasviel/sd-controlnet-depth", torch_dtype=torch.float16 | |
| # ) | |
| # # models that worked well: anime god, pastel dream, | |
| # # https://huggingface.co/shellypeng/featureless/tree/main | |
| # 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 = StableDiffusionControlNetPipeline.from_pretrained("furusu/SSD-1B-anime", | |
| # # torch_dtype=torch.float16 | |
| # # ) | |
| # pipe_depth.load_lora_weights("shellypeng/detail-tweaker") | |
| # pipe_depth.fuse_lora(lora_scale=0.1) | |
| # # pipe.load_lora_weights("shellypeng/stylized-3d") | |
| # # pipe.load_lora_weights("shellypeng/midjourney-anime") | |
| # # pipe.load_lora_weights("shellypeng/animetarot") | |
| # # pipe.load_lora_weights("shellypeng/anime-stickers-v3") | |
| # # pipe.load_lora_weights("shellypeng/anime-magazine") | |
| # pipe_depth.load_textual_inversion("shellypeng/bad-prompt") | |
| # pipe_depth.load_textual_inversion("shellypeng/badhandv4") | |
| # # pipe.load_textual_inversion("shellypeng/easynegative") | |
| # pipe_depth.load_textual_inversion("shellypeng/deepnegative") | |
| # pipe_depth.load_textual_inversion("shellypeng/verybadimagenegative") | |
| # pipe_depth.scheduler = DPMSolverMultistepScheduler.from_config(pipe_depth.scheduler.config, use_karras_sigmas=True) | |
| # # pipe.enable_model_cpu_offload() | |
| # def dummy(images, **kwargs): | |
| # return images, False | |
| # pipe_depth.safety_checker = lambda images, **kwargs: (images, [False] * len(images)) | |
| # pipe_depth.to(device) | |
| # # pipe.load_lora_weights("shellypeng/detail-tweaker", weight_name="add_detail.safetensors") | |
| # # load textual inversion negative embeddings!!!: pipe.load_textual_inversion("sd-concepts-library/cat-toy") | |
| # 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) | |
| # """ | |
| # input_img = Image.fromarray(input_img) | |
| # input_img = load_image(input_img) | |
| # input_img = depth_estimator(input_img)['depth'] | |
| # res_image0 = pipe_depth(prompt, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] | |
| # res_image1 = pipe_depth(prompt, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] | |
| # res_image2 = pipe_depth(prompt, input_img, negative_prompt=hidden_negative, num_inference_steps=40).images[0] | |
| # res_image3 = pipe_depth(prompt, input_img, negative_prompt=hidden_negative, 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_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("Animefier"): | |
| 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", placeholder="Enter a prompt", lines=3) | |
| neg_prompt_box = gr.Textbox(label="Negative Prompt", placeholder="Enter a negative prompt(things you don't want to include in the generated image)", lines=3) | |
| with gr.Row(equal_height=True): | |
| chinese_check = gr.Checkbox(label="Chinese Prompt Mode", 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", height=350) | |
| gen_btn = gr.Button(value="Generate") | |
| with gr.Row(equal_height=True): | |
| global image1 | |
| global image2 | |
| global image3 | |
| global image4 | |
| image1 = gr.Image(label="Result 1") | |
| image2 = gr.Image(label="Result 2") | |
| image3 = gr.Image(label="Result 3") | |
| image4 = gr.Image(label="Result 4") | |
| 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("AniSketch"): | |
| 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", placeholder="Enter a prompt", lines=3) | |
| neg_prompt_box = gr.Textbox(label="Negative Prompt", placeholder="Enter a negative prompt(things you don't want to include in the generated image)", lines=3) | |
| with gr.Row(equal_height=True): | |
| chinese_check = gr.Checkbox(label="Chinese Prompt Mode", 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", height=350) | |
| gen_btn = gr.Button(value="Generate") | |
| with gr.Row(equal_height=True): | |
| image1 = gr.Image(label="Result 1") | |
| image2 = gr.Image(label="Result 2") | |
| image3 = gr.Image(label="Result 3") | |
| image4 = gr.Image(label="Result 4") | |
| 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]) | |
| # gen_btn.click(mult_thread, [prompt_box, image_box, chinese_check], [image1, image2, image3, image4, chinese_check]) | |
| iface.dependencies[0]["show_progress"] = "hidden" | |
| iface.launch(debug=True, share=True, auth=["shenrym", "shjdqw%23-sw2&"]) | |