import torch # from torch import autocast from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler,AutoencoderKL,DPMSolverMultistepScheduler,StableDiffusionLatentUpscalePipeline import matplotlib.pyplot as plt from PIL import Image import time from diffusers import DiffusionPipeline import os import sys import requests import utils import json import torch def notify_discord(output): config = utils.get_config() discord_webhook_url = config['discord_webhook_url'] #発行したトークンを入力 data = {"content": "img"} requests.post( discord_webhook_url, data=data, files={'file': open(output, 'rb')} ) def notify_line(putput): config = utils.get_config() line_notify_token = config['line_notify_token'] #発行したトークンを入力 line_notify_api = "https://notify-api.line.me/api/notify" current_time = time.strftime("%Y%m%d%H%M%S") payload = {"message":current_time} #メッセージの本文 # payload = {} #メッセージの本文 headers = {"Authorization":"Bearer " + line_notify_token} requests.post( line_notify_api, data = payload, headers = headers, files={'imageFile': open(output, 'rb')} ) if __name__ == '__main__': # SendLineMessage("Start") # notify_discord("Start") # sys.exit() # cuDNN 自動チューナーを有効にする # torch.backends.cudnn.benchmark = True # fp32 の代わりに tf32 を使用する (Ampere とそれ以後の CUDA デバイス上で) # torch.backends.cuda.matmul.allow_tf32 = True base_dir = os.path.join('C:' + os.sep,'dev','ai','stable-diffusion-webui-forge') lora_dir = os.path.join(base_dir, 'models', 'lora') model_dir = os.path.join(base_dir, 'models', 'diffusers') vae_dir = os.path.join(base_dir, 'models', 'VAE') config = utils.get_config() headers = { # "User-Agent": "user-agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36", "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/116.0.0.0 Safari/537.36", } width =config['width'] height = config['height'] create_count = config['create_count'] url = config['check_url'] r = requests.get(url, headers=headers,verify=False) json_data = r.json() # print(json.dumps(json_data, indent=2)) # sys.exit() for data in json_data: for i in range(create_count): try: model = data['model'] lora = data['lora'] prompt = data['prompt'] negative_prompt = data['negative_prompt'] if(model.find('/') == -1): model = os.path.join(model_dir , model) # vae_path = os.path.join(vae_dir , "vae-ft-mse-840000-ema-pruned.safetensors") # vae = AutoencoderKL.from_single_file(vae_path) # ダウンロードしたVAEファイルを指定 # vae = AutoencoderKL.from_single_file("./vae/sdxl_vae.safetensors") print(f"■Model: {model}") print(f"■Lora: {lora}") print(f"■Prompt: {prompt}") print(f"■Negative Prompt: {negative_prompt}") # print(f"■VAE: {vae_path}") print(f"■Width: {width}") print(f"■Height: {height}") # print(width * 4) # sys.exit() # sys.exit() print("---------------start--------------") # ダウンロードしたファイル名を指定して、Pipeline作成 # pipe = DiffusionPipeline.from_single_file( # "Counterfeit-V3.0_fp16.safetensors", torch_dtype=torch.float16 # ).to("cuda") pipe = StableDiffusionXLPipeline.from_pretrained( model, torch_dtype=torch.float16, # torch_dtype=torch.float32, # vae=vae, # use_safetensors=True, # safety_checker = None, # requires_safety_checker = False ) # pipe = DiffusionPipeline.from_pretrained( # model, # torch_dtype=torch.float16, # # torch_dtype=torch.float32, # # vae=vae, # # use_safetensors=True, # # safety_checker = None, # # requires_safety_checker = False # ) # try: # pipe = StableDiffusionXLPipeline.from_pretrained( # model, # torch_dtype=torch.float16, # # torch_dtype=torch.float32, # # vae=vae, # # use_safetensors=True, # # safety_checker = None, # # requires_safety_checker = False # ) # except Exception as e: # pipe = DiffusionPipeline.from_pretrained( # model, # torch_dtype=torch.float16, # # torch_dtype=torch.float32, # # vae=vae, # # use_safetensors=True, # # safety_checker = None, # # requires_safety_checker = False # ) # pipe = DiffusionPipeline.from_single_file( # model, # vae=AutoencoderKL.from_single_file("vae/Counterfeit-V2.5.vae.pt"), # load_safety_checker=False, # extract_ema=True # ) # 黒画像回避 pipe.safety_checker = None pipe.requires_safety_checker = False # pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config) pipe.to('cuda') # 更なるメモリ節約のための sliced アテンション # pipe.enable_attention_slicing() # CPU でのシーケンシャルなオフロードを有効にする # pipe.enable_sequential_cpu_offload() if(lora != None): pipe.load_lora_weights( lora_dir, weight_name= lora ) # if pipe.safety_checker is not None: # pipe.safety_checker = lambda images, **kwargs: (images, False) # if pipe.safety_checker is not None: # pipe.safety_checker = lambda images, **kwargs: (images, False) current_time = time.strftime("%Y%m%d%H%M%S") output = "content/"+str(current_time)+".png" generator = torch.Generator() generator.manual_seed(2833846766) image = pipe( prompt, negative_prompt=negative_prompt, # width=832, # height=1216, width=width, height=height, # width=1024, # height=1024, guidance_scale=5, # num_inference_steps=28 num_inference_steps=50, # generator=generator, target_size=((width * 4),(height * 4)), original_size=((width * 8),(height * 8)), ).images[0] # -------------------------- # model_id = "stabilityai/sd-x2-latent-upscaler" # upscaler = StableDiffusionLatentUpscalePipeline.from_pretrained(model_id, torch_dtype=torch.float16) # upscaler.to("cuda") # image = upscaler( # prompt=prompt, # image=image, # num_inference_steps=20, # guidance_scale=0, # # generator=generator, # ).images[0] # -------------------------- image.save(output) image = Image.open(output) plt.imshow(image) plt.axis('off') # to hide the axis # notify_discord(output) print("line notify") notify_line(output) print("line notify end") url = config['update_url'] obj = { "id": data['id'], } print("post") # response = requests.post(url,json=obj,verify=False) response = requests.post( url, data=obj, verify=False, files={'file': open(output, 'rb')} ) if response.status_code == 200: print("Success!") print(response.json()) # サーバーからのレスポンスを出力 else: print("POST request failed with status code:", response.status_code) print("posted") # sys.exit() except Exception as e: print(f"Error: {e}") continue finally: print("---------------end--------------") print("")