app.py

import discord
import gradio_client
from gradio_client import Client
import gradio as gr
import os
import threading

#for deepfloydif
import requests
import json
import random
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import time

# random + small llama #


#todos
#alert
#fix error on first command on bot startup
#stable diffusion upscale
#buttons for deepfloydIF (1,2,3,4)
#application commands instead of message content checks (more user-friendly)




DFIF_TOKEN = os.getenv('DFIF_TOKEN')

#deepfloydIF
df = Client("DeepFloyd/IF", DFIF_TOKEN) #not reliable at the moment
#df = Client("huggingface-projects/IF", DFIF_TOKEN)

#stable diffusion upscaler
sdlu = Client("huggingface-projects/stable-diffusion-latent-upscaler", DFIF_TOKEN)

# Set up discord bot 
class MyClient(discord.Client):
    async def on_ready(self):
        print('Logged on as', self.user)


    async def on_message(self, message):

        #safety checks----------------------------------------------------------------------------------------------------

        #   tldr, bot should run if
        #1) it does not have @offline role
        #2) user has @verified role
        #3) bot is in #bot-test channel
        
        # bot won't respond to itself, prevents feedback loop + API spam
        if message.author == self.user:
            return       

        # if the bot has this role, it won't run
        OFFLINE_ROLE_ID = 1103676632667017266  # 1103676632667017266 = @offline / under maintenance
        guild = message.guild
        bot_member = guild.get_member(self.user.id)
        if any(role.id == OFFLINE_ROLE_ID for role in bot_member.roles):
            return
        
        # the message author needs this role in order to use the bot
        REQUIRED_ROLE_ID = 897376942817419265 # 900063512829755413 = @verified, 897376942817419265 = @huggingfolks
        if not any(role.id == REQUIRED_ROLE_ID for role in message.author.roles):
            return            

        # channels where bot will accept commands
        ALLOWED_CHANNEL_IDS = [1100458786826747945] # 1100458786826747945 = #bot-test
        if message.channel.id not in ALLOWED_CHANNEL_IDS:
            return

        #deepfloydif----------------------------------------------------------------------------------------------------
            
        if message.content.startswith('!deepfloydif'): # change to application commands, more intuitive

            #(prompt, negative_prompt, seed, number_of_images, guidance_scale,custom_timesteps_1, number_of_inference_steps, api_name="/generate64") 
            #-> (stage_1_results, stage_1_param_path, stage_1_result_path)

            # input prompt
            prompt = message.content[12:].strip()   
            number_of_images = 4

            # random seed
            current_time = int(time.time())
            random.seed(current_time)
            seed = random.randint(0, 2**32 - 1)
            
            stage_1_results, stage_1_param_path, stage_1_result_path = df.predict(prompt, "blur", seed,number_of_images,7.0, 'smart100',50, api_name="/generate64")

            #stage_1_results, stage_1_param_path, stage_1_result_path = df.predict("gradio written on a wall", "blur", 1,1,7.0, 'smart100',50, api_name="/generate64")
            
            # stage_1_results -> path to directory with png files, so we isolate those
            png_files = [f for f in os.listdir(stage_1_results) if f.endswith('.png')]

            # merge images into larger, 2x2 image the way midjourney does it
            if png_files:
                first_png = png_files[0]
                second_png = png_files[1]
                third_png = png_files[2]
                fourth_png = png_files[3]

                '''
                [],[],[],[] -> [][]
                               [][]
                
                '''
            
                first_png_path = os.path.join(stage_1_results, first_png)
                second_png_path = os.path.join(stage_1_results, second_png)
                third_png_path = os.path.join(stage_1_results, third_png)
                fourth_png_path = os.path.join(stage_1_results, fourth_png)
            
                img1 = Image.open(first_png_path)
                img2 = Image.open(second_png_path)
                img3 = Image.open(third_png_path)
                img4 = Image.open(fourth_png_path)
            
                # create a new blank image with the size of the combined images (2x2)
                combined_image = Image.new('RGB', (img1.width * 2, img1.height * 2))
            
                # paste the individual images into the combined image
                combined_image.paste(img1, (0, 0))
                combined_image.paste(img2, (img1.width, 0))
                combined_image.paste(img3, (0, img1.height))
                combined_image.paste(img4, (img1.width, img1.height))
            
                # save the combined image
                combined_image_path = os.path.join(stage_1_results, 'combined_image.png')
                combined_image.save(combined_image_path)
            
            # send the combined image file as a discord attachment
            with open(combined_image_path, 'rb') as f:
                await message.reply('Here is the combined image', file=discord.File(f, 'combined_image.png'))








            '''
            # stage 2
            selected_index_for_stage_2 = -1
            custom_timesteps_2 = 'smart100' # could reset to smart50 if index was the issue
            seed = 362572064 # again, could randomize this        seed = 362572064      seed = random.randint(0, 2**32 - 1)
            
            # predict(stage_1_result_path, selected_index_for_stage_2, seed, guidance_scale, custom_timesteps_2, number_of_inference_steps, api_name="/upscale256") -> result
            
            img = df.predict(stage_1_result_path, selected_index_for_stage_2, seed, guidance_scale, custom_timesteps_2, number_of_inference_steps, api_name="/upscale256")
    
            # Save the generated image to a file
            img_path = "/tmp/generated_image.png"
            img.save(img_path)
            
            # Send the image file as a Discord attachment
            with open(img_path, 'rb') as f:
                await message.reply(f'Here is the generated image', file=discord.File(f, 'generated_image.png'))                
            '''
       
            

DISCORD_TOKEN = os.environ.get("GRADIOTEST_TOKEN", None)
intents = discord.Intents.default()
intents.message_content = True
client = MyClient(intents=intents)

def run_bot():
  client.run(DISCORD_TOKEN)

threading.Thread(target=run_bot).start()

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

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