app.py

import gradio as gr
import requests
import io
import random
import os
from PIL import Image
from deep_translator import GoogleTranslator
from langdetect import detect

API_URL = "https://api-inference.huggingface.co/models/openskyml/dalle-3-xl"
API_TOKEN = os.getenv("HF_READ_TOKEN") # it is free
headers = {"Authorization": f"Bearer {API_TOKEN}"}

def query(prompt, model, is_negative=False, steps=20, cfg_scale=7, seed=None):
    language = detect(prompt)
    
    if language == 'ru':
        prompt = GoogleTranslator(source='ru', target='en').translate(prompt)
    print(f'\033[1mГенерация:\033[0m {prompt}')
    
    if model == 'DALL-E 3 XL':
        API_URL = "https://api-inference.huggingface.co/models/openskyml/dalle-3-xl"
    if model == 'SDXL Turbo':
        API_URL = "https://api-inference.huggingface.co/models/stabilityai/sdxl-turbo"

    payload = {
        "inputs": prompt,
        "is_negative": is_negative,
        "steps": steps,
        "cfg_scale": cfg_scale,
        "seed": seed if seed is not None else random.randint(-1, 2147483647)
        }

    image_bytes = requests.post(API_URL, headers=headers, json=payload).content
    image = Image.open(io.BytesIO(image_bytes))
    return image


css = """
footer {visibility: hidden !important;}
"""

with gr.Blocks(css=css) as dalle:
    with gr.Tab("Базовые настройки"):
        with gr.Row():
            with gr.Column(elem_id="prompt-container"):
                text_prompt = gr.Textbox(label="Prompt", placeholder="Описание изображения", lines=3, elem_id="prompt-text-input")
                model = gr.Radio(label="Модель", value="DALL-E 3 XL", choices=["DALL-E 3 XL", "SDXL Turbo"])
                

    with gr.Tab("Расширенные настройки"):
        negative_prompt = gr.Textbox(label="Negative Prompt", placeholder="Чего не должно быть на изображении", value="[deformed | disfigured], poorly drawn, [bad : wrong] anatomy, [extra | missing | floating | disconnected] limb, (mutated hands and fingers), blurry, text, fuzziness", lines=3, elem_id="negative-prompt-text-input")

    with gr.Row():
        text_button = gr.Button("Генерация", variant='primary', elem_id="gen-button")
    with gr.Row():
        image_output = gr.Image(type="pil", label="Изображение", elem_id="gallery")
        
    text_button.click(query, inputs=[text_prompt, model, negative_prompt], outputs=image_output)

dalle.launch(show_api=False)