import gradio as gr
import random, os, shutil
from PIL import Image
import pandas as pd

def open_ims(adj, group, seed):
    if group != '':
        if adj != '':
            prompt=adj+'_'+group.replace(' ','_')
            if os.path.isdir(prompt) == False:
                shutil.unpack_archive('zipped_images/'+ prompt.replace(' ', '_') +'.zip', prompt, 'zip')
        else:
            prompt=group
            if os.path.isdir(prompt) == False:
                shutil.unpack_archive('zipped_images/'+ prompt.replace(' ', '_') +'.zip', prompt, 'zip')
        imnames= os.listdir(prompt+'/Seed_'+ str(seed)+'/')
        images = [(Image.open(prompt+'/Seed_'+ str(seed)+'/'+name)) for name in imnames]
        return images[:9]


vowels = ["a","e","i","o","u"]
prompts = pd.read_csv('promptsadjectives.csv')

seeds = [46267, 48040, 51237, 54325, 60884, 64830, 67031, 72935, 92118, 93109]

m_adjectives = prompts['Masc-adj'].tolist()[:10]
f_adjectives = prompts['Fem-adj'].tolist()[:10]
adjectives = sorted(m_adjectives+f_adjectives)
#adjectives = ['attractive','strong']
adjectives.insert(0, '')
professions = sorted([p.lower() for p in prompts['Occupation-Noun'].tolist()])

with gr.Blocks() as demo:
    gr.Markdown("# Stable Diffusion Explorer")
    gr.Markdown("## Choose from the prompts below to explore how the Stable Diffusion model represents different professions and adjectives")
#    seed_choice = gr.State(0)
#    seed_choice = 93109
#    print("Seed choice is: " + str(seed_choice))
    with gr.Row():
        with gr.Column():
            adj1 = gr.Dropdown(adjectives, label = "Choose a first adjective (or leave this blank!)", interactive=True)
            choice1 = gr.Dropdown(professions, label = "Choose a first group", interactive=True)
            seed1= gr.Dropdown(seeds, label = "Choose a random seed to compare results", value = seeds[1], interactive=True)
            images1 = gr.Gallery(label="Images").style(grid=[3], height="auto")
        with gr.Column():
            adj2 = gr.Dropdown(adjectives, label = "Choose a second adjective (or leave this blank!)", interactive=True)
            choice2 = gr.Dropdown(professions, label = "Choose a second group", interactive=True)
            seed2= gr.Dropdown(seeds, label = "Choose a random seed to compare results", value= seeds[1], interactive=True)
            images2 = gr.Gallery(label="Images").style(grid=[3], height="auto")

    gr.Markdown("### [Research](http://gender-decoder.katmatfield.com/static/documents/Gaucher-Friesen-Kay-JPSP-Gendered-Wording-in-Job-ads.pdf) has shown that \
                    certain words are considered more masculine- or feminine-coded based on how appealing job descriptions containing these words \
                    seemed to male and female research participants and to what extent the participants felt that they 'belonged' in that occupation.")


    #demo.load(random_image, None, [images])
    choice1.change(open_ims, [adj1,choice1,seed1], [images1])
    choice2.change(open_ims, [adj2,choice2,seed2], [images2])
    adj1.change(open_ims, [adj1,choice1,seed1], [images1])
    adj2.change(open_ims, [adj2,choice2,seed2], [images2])
    seed1.change(open_ims, [adj1,choice1,seed1], [images1])
    seed2.change(open_ims, [adj2,choice2,seed2], [images2])

demo.launch()