app.py

import gradio as gr
from urllib.parse import quote
import re
import os
import openai
from gradio_client import Client

OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
clipi_client = Client("https://fffiloni-clip-interrogator-2.hf.space/")

chapter_num = 1
story_intro = ""
current_story = ""
current_image_url = ""
chapters = []

def generate_image_url(keywords):
    truncated_keywords = keywords[:20]
    return f"https://image.pollinations.ai/prompt/{quote(truncated_keywords)}"

def next_chapter(audience, keyword, protagonist):
    global chapter_num, current_story, current_image_url
    current_image_url = generate_image_url(current_story)
    gr.Info(f'Chapter {chapter_num}를 생성하고 있습니다...')
    chapter_prompt = f"{story_intro}\n\nKeyword: {keyword}\nProtagonist: {protagonist}\n\n![Chapter {chapter_num} Image]({current_image_url})\n\nChapter {chapter_num} 내용을 만들어줘."
    chat_completion = openai.ChatCompletion.create(model="gpt-3.5-turbo-16k", messages=[{"role": "user", "content": chapter_prompt}])
    current_story = chat_completion.choices[0].message.content
    chapters.append({"story": current_story, "image": current_image_url})
    chapter_num += 1
    chapter_table = "| 챕터 | 이야기 | 그림 |\n|------|--------|------|\n" + "\n".join([f"| {i+1} | {c['story']} | ![image]({c['image']}) |" for i, c in enumerate(chapters)])
    return current_story, current_image_url, chapter_table

def infer(image_input, audience, keyword, protagonist):
    global story_intro, current_story, current_image_url, chapter_num, chapters
    chapter_num = 1
    chapters = []
    gr.Info('Calling CLIP Interrogator, 이미지를 해석하고 있습니다...')
    clipi_result = clipi_client.predict(image_input, "best", 4, api_name="/clipi2")[0]

    story_intro = f"""
    # Illustrated Tales
    ## Created by [Sigkawat Pengnoo](https://flowgpt.com/prompt/qzv2D3OvHkzkfSE4rQCqv) at FlowGPT
    Keyword: {keyword}
    Protagonist: {protagonist}
    한국어로 답변해줘.
    STORY : "{{ {clipi_result} }}"
    Let's begin with Chapter 1!
    """

    current_story = clipi_result
    return next_chapter(audience, keyword, protagonist)

css = """
#col-container {max-width: 910px; margin-left: auto; margin-right: auto;}
a {text-decoration-line: underline; font-weight: 600;}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(
            """
            <h1 style="text-align: center">Illustrated Tales - Korean</h1>
            <p style="text-align: center">이미지를 업로드하세요, ChatGPT를 통해 한국어로 이야기와 그림을 만들어 줍니다!</p>
            """
        )
        with gr.Row():
            with gr.Column():
                image_in = gr.Image(label="이미지 입력", type="filepath", elem_id="image-in", height=420)
                audience = gr.Radio(label="대상", choices=["Children", "Adult"], value="Children")
                keyword_in = gr.Textbox(label="핵심 키워드")
                protagonist_in = gr.Textbox(label="주인공")
                submit_btn = gr.Button('이야기와 그림을 만들어 주세요')
                next_chapter_btn = gr.Button('다음 이야기')
            with gr.Column():
                chapter_story = gr.Markdown(label="이야기", elem_id="chapter_story")
                chapter_image = gr.Image(label="그림", elem_id="chapter_image")
                chapter_table_md = gr.Markdown(label="모든 챕터")

    submit_btn.click(fn=infer, inputs=[image_in, audience, keyword_in, protagonist_in], 
                     outputs=[chapter_story, chapter_image, chapter_table_md])
    next_chapter_btn.click(fn=lambda: next_chapter(audience=audience.value, keyword=keyword_in.value, protagonist=protagonist_in.value), 
                           outputs=[chapter_story, chapter_image, chapter_table_md])

    demo.queue(max_size=12).launch()