app.py

import os
import json
import base64
import requests
import gradio as gr
import matplotlib.pyplot as plt

from PIL import Image
from io import BytesIO
from dotenv import load_dotenv
from huggingface_hub import HfFileSystem, hf_hub_download

def request_to_endpoint (request:dict, endpoint_id='s4i4um7xakaq37'):
    api_key = os.getenv("RUNPOD_API_KEY")
    url = f"https://api.runpod.ai/v2/{endpoint_id}/runsync"
    headers = {
        "accept": "application/json",
        "authorization": api_key,
        "content-type": "application/json"
    }
    response = requests.post(url, headers=headers, data=json.dumps(request)) 

    return response

# Собирает только sdxl версии
def prepare_characters(hf_path="OnMoon/loras"):
    fs = HfFileSystem()

    files = fs.ls(hf_path, detail=False)
    character_names = []
    character_configs = {}
    for file in files:
        if file.endswith(".safetensors") and file.startswith(f"{hf_path}/sdxl_"):
            character = file[len(f"{hf_path}/sdxl_"): -len(".safetensors")]
            character_names.append(character)
        elif file.endswith(".json") and file.startswith(f"{hf_path}/sdxl_"):
            character = file[len(f"{hf_path}/sdxl_"): -len(".json")]
            with fs.open(file, 'r', encoding='utf-8') as file_json:
                character_configs[character] = json.load(file_json)  
        
    return character_names, character_configs

character_names, character_configs = prepare_characters()


def process_input (name, scale, triggers, tech_prompt, tech_negative_prompt, prompts_count, *args):    
    # Если хотим задать свои триггерные слова или технический промпт
    triggers = triggers if triggers != "" else ",".join(character_configs[name]["trigger_words"])
    tech_prompt = tech_prompt if tech_prompt != "" else character_configs[name]["tech_prompt"]
    tech_negative_prompt = tech_negative_prompt if tech_negative_prompt != "" else character_configs[name]["tech_negative_prompt"]

    model = character_configs[name]["model"]
    model['loras'] = {name: scale}
    params = character_configs[name]["params"]
    params["cross_attention_kwargs"]['scale'] = scale

    boxes = list(args)
    prompts = []
    negative_prompts = []
    for n in range(prompts_count):
        prompts.append(f'{triggers}, {tech_prompt}, {boxes[n]}')
        negative_prompts.append(f"{tech_negative_prompt}")
    
    request_data = {
        "input": {
            "model": model,
            "params": params,
            "prompt": prompts,
            "negative_prompt": negative_prompts,
            "height": 1216,
            "width": 832,
        }
    }
    
    response = request_to_endpoint(request_data)

    images = []
    for base64_string in response.json()['output']['images']:
        img = Image.open(BytesIO(base64.b64decode(base64_string)))
        images.append(img)

    gallery = [[images[i], f"{prompts[i]}"] for i in range(prompts_count)]

    return gallery






                            ######################################################
                            #   ____               _ _         _                 #
                            #  / ___|_ __ __ _  __| (_) ___   / \   _ __  _ __   #
                            # | |  _| '__/ _` |/ _` | |/ _ \ / _ \ | '_ \| '_ \  #
                            # | |_| | | | (_| | (_| | | (_) / ___ \| |_) | |_) | #
                            #  \____|_|  \__,_|\__,_|_|\___/_/   \_\ .__/| .__/  #
                            #                                      |_|   |_|     #
############################################################################################################
with gr.Blocks() as demo:
    with gr.Group():
        name = gr.Radio(
            character_names,
            label="Select character:",
            interactive=True,
            visible=True,
        )

        scale = gr.Slider(
            minimum=0, 
            maximum=2.0,
            value=0.75, 
            step=0.01, 
            label="Selected LoRA scale:", 
            interactive=True,
        )

        with gr.Accordion(open=False):
            triggers = gr.Textbox(label=f"Trigger words:")
            tech_prompt = gr.Textbox(label=f"Technical prompt:")
            tech_negative_prompt = gr.Textbox(label=f"Negative technical prompt:")

    prompts_count = gr.State(1)

    with gr.Group():
        add_btn = gr.Button("Add prompt")
        del_btn = gr.Button("Delete prompt")
            
        add_btn.click(lambda x: x + 1, prompts_count, prompts_count)
        del_btn.click(lambda x: x - 1, prompts_count, prompts_count)

        @gr.render(inputs=prompts_count)
        def render_count(count):
            boxes = []
            for i in range(count):
                with gr.Group():
                    prompt = gr.Textbox(key=str(i), label=f"Prompt {i+1}")
                boxes.append(prompt)
            
            generate_btn.click(
                process_input, 
                [name, scale, triggers, tech_prompt, tech_negative_prompt, prompts_count]+boxes,
                output
            )

    generate_btn = gr.Button("Generate!")

    output = gr.Gallery(
        label="Generation results:",
        object_fit="contain", 
        height="auto",
    )

demo.launch()
############################################################################################################