import torch
import gradio as gr
import numpy as np
from torch import autocast
from PIL import Image
from diffusers import StableDiffusionImg2ImgPipeline

# load the pipeline
device = "cuda"
model_id_or_path = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
model_id_or_path,
    revision="fp16", 
    torch_dtype=torch.float16,
    use_auth_token='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'
)
# or download via git clone https://huggingface.co/CompVis/stable-diffusion-v1-4
# and pass `model_id_or_path="./stable-diffusion-v1-4"` without having to use `use_auth_token=True`.
pipe = pipe.to(device)

def diffuse(x, param):
    print('in callback')
    x = Image.fromarray(np.uint8(x))
    init_image = x.resize((768, 512))
    prompt = 'st petersburg logo'
    if param == 'Эрмитаж':
        prompt = "st petersburg logo winter palace image on background hermitage vector style"
    elif param == 'Казанский собор':
        prompt = "st petersburg logo kazansky sobor image on background"
    elif param == 'Мосты':
        prompt = 'st petersburg logo bridges over neva image on background beutiful high quality'

    with autocast("cuda"):
         images = pipe(prompt=prompt, init_image=init_image, strength=0.7, guidance_scale=7.5).images
    return [images[0], param]

def flip_image(x, param):
    return [np.fliplr(x), 'функция приняла на вход ' + param]

with gr.Blocks() as demo:
    gr.Markdown("Слово 'Санкт-Петербург'")
    with gr.Tab("Санкт-Петербург"):
        with gr.Row():
            image_input = gr.Image()
            param_input = gr.Radio(["Эрмитаж", "Мосты", "Казанский собор"], label='Что для тебя Санкт-Петербург?')
            image_output = gr.Image()
            param_out = gr.Markdown()
        image_button = gr.Button("GET IMAGE")

    image_button.click(diffuse, [image_input, param_input], [image_output, param_out])

demo.launch()

#def greet(name):
#    return "Hello " + name + "!!"

#iface = gr.Interface(fn=greet, inputs="text", outputs="text")
#iface.launch()