app.py

import random
import re
import torch
import requests
from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler
from torch import autocast
import gradio as gr

def txt2img(prompt):

  image = pipe(prompt +",　realistic, highly detailed, high quality", height=512, width=512, negative_prompt = "((low quality)),((poor quality)),((clone)),retro style, bad anatomy,((lowres)), blurry, (worst quality), ((low quality)), normal quality,bad anatomy, disfigured, deformed, mutation, mutilated, ugly, totem pole,(poorly drawn face), cloned face, several faces, long neck, mutated hands, bad hands, poorly drawn hands,extra limbs, malformed limbs, missing arms, missing fingers, extra fingers, fused fingers, too many fingers,missing legs, extra legs, malformed legs, extra digit, fewer digits, glitchy, cropped, jpeg artifacts, signature, watermark, username, text, errorretro style ,bad anatomy,((lowres)), blurry, (worst quality), normal quality,bad anatomy, disfigured, deformed, mutation, mutilated, ugly, totem pole,(poorly drawn face), cloned face, several faces, long neck, mutated hands, bad hands, poorly drawn hands,extra limbs, malformed limbs, missing arms, missing fingers, extra fingers, fused fingers, too many fingers,missing legs, extra legs, malformed legs, extra digit, fewer digits, glitchy, cropped, jpeg artifacts, signature, watermark, username, text, error", guidance_scale = 7.5,num_inference_steps = 50).images[0]
    
  image.save("sd_image.png")

  return image

pipe = StableDiffusionPipeline.from_pretrained("Fung804/makoto-shinkai-v2", torch_dtype=torch.float16)  
pipe = pipe.to("cuda")    
generate = gr.Interface(fn = txt2img, inputs="text",outputs="image",allow_flagging="never")

generate.launch()