init

app.py

@@ -0,0 +1,386 @@
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
+from sqlmodel import Field, Session, SQLModel, create_engine, select
+from typing import Optional, List, Tuple
+import hashlib
+from datetime import datetime
+from reportlab.lib.pagesizes import A4
+from reportlab.lib.units import cm
+from reportlab.platypus import SimpleDocTemplate, Paragraph, Image, PageBreak, Spacer
+from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle
+from reportlab.pdfbase import pdfmetrics
+from reportlab.pdfbase.ttfonts import TTFont
+from reportlab.lib.enums import TA_CENTER, TA_JUSTIFY
+from PIL import Image as PILImage
+import os
+
+# 캐릭터 일관성을 위한 고정 속성 추가 (한국어)
+CHARACTER_DESCRIPTION = "young korean man with blue hoodie"
+
+# 데이터베이스 모델
+class Story(SQLModel, table=True):
+    id: Optional[int] = Field(default=None, primary_key=True)
+    prompt: str
+    content: str
+    created_at: datetime = Field(default_factory=datetime.now)
+
+class ImageCache(SQLModel, table=True):
+    id: Optional[int] = Field(default=None, primary_key=True)
+    prompt_hash: str = Field(index=True)
+    image_path: str
+    created_at: datetime = Field(default_factory=datetime.now)
+
+# 데이터베이스 초기화
+engine = create_engine("sqlite:///storybook.db")
+SQLModel.metadata.create_all(engine)
+
+# 모델 초기화
+print("모델 로딩 중...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# LLM 모델
+llm_model_name = "Bllossom/llama-3.2-Korean-Bllossom-AICA-5B"
+tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
+llm_model = AutoModelForCausalLM.from_pretrained(
+    llm_model_name,
+    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
+    device_map="auto"
+)
+
+# Stable Diffusion 모델
+sd_model_name = "Lykon/DreamShaper"
+sd_pipe = StableDiffusionPipeline.from_pretrained(
+    sd_model_name,
+    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
+    use_safetensors=False
+)
+sd_pipe = sd_pipe.to(device)
+
+# 모델 로드 후 스케줄러 변경
+sd_pipe.scheduler = DPMSolverMultistepScheduler.from_config(
+    sd_pipe.scheduler.config,
+    use_karras_sigmas=True,  # Karras schedule
+    algorithm_type="dpmsolver++"
+)
+
+# 이미지 저장 디렉토리
+os.makedirs("generated_images", exist_ok=True)
+
+
+def generate_story(prompt: str) -> Tuple[str, List[str]]:
+    """프롬프트로부터 스토리 생성"""
+    system_prompt = f"""당신은 뛰어난 스토리텔러입니다.
+다음 주제를 바탕으로, 5개의 문단으로 구성된 흥미로운 이야기를 작성하세요.
+
+규칙:
+- 주인공은 '청년' 또는 '그'로만 지칭하세요 (이름 사용 금지)
+- 주인공은 안경을 쓴 20대 청년입니다
+- 각 문단은 2~4개의 문장으로 구성
+- 시각적으로 표현 가능한 구체적인 장면 묘사 포함
+- 순수 한국어만 사용
+- 각 문단마다 명확한 장소와 행동 묘사
+
+주제: {prompt}
+
+이야기:"""
+    
+    inputs = tokenizer(system_prompt, return_tensors="pt").to(device)
+    
+    with torch.no_grad():
+        outputs = llm_model.generate(
+            **inputs,
+            max_new_tokens=1000,
+            temperature=0.7,
+            do_sample=True,
+            top_p=0.92,
+            repetition_penalty=1.1,
+        )
+    
+    story = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    story = story.replace(system_prompt, "").strip()
+    
+    # 문단 분리
+    paragraphs = []
+    raw_paragraphs = story.split("\n\n")
+    for p in raw_paragraphs:
+        p = p.strip()
+        if p and len(p) > 20:
+            paragraphs.append(p)
+    
+    paragraphs = paragraphs[:5]
+    
+    # DB 저장
+    with Session(engine) as session:
+        db_story = Story(prompt=prompt, content="\n\n".join(paragraphs))
+        session.add(db_story)
+        session.commit()
+    
+    return "\n\n".join(paragraphs), paragraphs
+
+
+def analyze_text_for_english_scene(text: str, paragraph_num: int = 1) -> str:
+    """텍스트를 분석하여 영어 씬 추출 (기본 10개 키워드)"""
+    
+    # 디버깅용 출력
+    print(f"[{paragraph_num}] 텍스트 분석 중: {text[:60]}...")
+    
+    # 핵심 키워드 10개만 처리
+    # 1. 카페 + 노트북/컴퓨터
+    if "카페" in text and ("노트북" in text or "컴퓨터" in text):
+        return "working on laptop in coffee shop"
+    
+    # 2. 카페 (일반)
+    elif "카페" in text:
+        return "in a coffee shop"
+    
+    # 3. 프로그래밍/코딩
+    elif "프로그래밍" in text or "코딩" in text or "코드" in text:
+        return "coding on laptop"
+    
+    # 4. 회의/미팅
+    elif "회의" in text or "미팅" in text:
+        return "in a meeting"
+    
+    # 5. 발표/프레젠테이션
+    elif "발표" in text or "프레젠테이션" in text:
+        return "giving presentation"
+    
+    # 6. 동료/팀
+    elif "��료" in text or "팀" in text:
+        return "with team members"
+    
+    # 7. 성공/축하
+    elif "성공" in text or "축하" in text:
+        return "celebrating success"
+    
+    # 8. 계획
+    elif "계획" in text:
+        return "planning"
+    
+    # 9. 사무실
+    elif "사무실" in text:
+        return "in office"
+    
+    # 10. 투자/투자자
+    elif "투자" in text:
+        return "meeting investors"
+    
+    # 기본값 (문단별)
+    defaults = {
+        1: "young entrepreneur working",
+        2: "developing project",
+        3: "collaborating with others",
+        4: "business presentation",
+        5: "successful achievement"
+    }
+    
+    return defaults.get(paragraph_num, "at work")
+
+    
+def generate_image(text: str, paragraph_num: int = 1) -> str:
+    """텍스트로부터 이미지 생성"""
+    # 프롬프트 해시 생성
+    prompt_hash = hashlib.md5(text.encode()).hexdigest()
+    
+    # 캐시 확인
+    with Session(engine) as session:
+        cached = session.exec(
+            select(ImageCache).where(ImageCache.prompt_hash == prompt_hash)
+        ).first()
+        
+        if cached:
+            return cached.image_path
+    
+    # 씬 추출
+    print(f"\n[{paragraph_num}/5] 이미지 생성 중...")
+    scene = analyze_text_for_english_scene(text)
+    
+    # 최종 프롬프트 생성 
+    final_prompt = f"{CHARACTER_DESCRIPTION} {scene}"
+    
+    
+    print("최종 프롬프트: ", final_prompt)
+    print(f"프롬프트 길이: {len(final_prompt)} 글자")
+    
+    # 네거티브 프롬프트
+    negative_prompt = "realistic, photo, multiple people, crowd"
+    
+    # Seed 고정
+    base_seed = 396135060
+    # Seed 미세 변화
+    seed = base_seed + (paragraph_num * 10)  # 10, 20, 30, 40, 50
+    # Seed 다변화
+    #text_hash = int(hashlib.md5(text.encode()).hexdigest()[:8], 16)
+    #seed = base_seed + (text_hash % 1000)
+    generator = torch.Generator(device=device).manual_seed(seed)
+    #generator = torch.Generator(device=device).manual_seed(
+    #    torch.randint(0, 100000, (1,)).item()
+    #)
+    
+    # 이미지 생성
+    with torch.no_grad():
+        image = sd_pipe(
+            prompt=final_prompt,
+            negative_prompt=negative_prompt,
+            num_inference_steps=20,
+            guidance_scale=6.0,
+            height=512,
+            width=512,
+            generator=generator,
+            safety_checker=None,
+            requires_safety_checker=False
+        ).images[0]
+    
+    # 이미지 저장
+    image_path = f"generated_images/{prompt_hash}.png"
+    image.save(image_path)
+    
+    # 캐시 저장
+    with Session(engine) as session:
+        cache_entry = ImageCache(prompt_hash=prompt_hash, image_path=image_path)
+        session.add(cache_entry)
+        session.commit()
+    
+    return image_path
+
+def create_pdf(story_text: str, image_paths: List[str], output_path: str = "storybook.pdf"):
+    """스토리와 이미지로 PDF 생성"""
+    doc = SimpleDocTemplate(output_path, pagesize=A4)
+    story = []
+
+    font_path = "malgun.ttf"
+    pdfmetrics.registerFont(TTFont('맑은고딕', font_path))
+
+    # 스타일 설정
+    styles = getSampleStyleSheet()
+    title_style = ParagraphStyle(
+        'CustomTitle',
+        parent=styles['Heading1'],
+        fontName="맑은고딕",
+        fontSize=24,
+        textColor='black',
+        alignment=TA_CENTER,
+        spaceAfter=30
+    )
+
+    text_style = ParagraphStyle(
+        'CustomText',
+        parent=styles['Normal'],
+        fontName="맑은고딕",
+        fontSize=12,
+        leading=18,
+        alignment=TA_JUSTIFY,
+        spaceAfter=20
+    )
+
+    story.append(Paragraph("AI 스토리북", title_style))
+    story.append(Spacer(1, 1*cm))
+
+    paragraphs = story_text.strip().split("\n\n")
+    for i, para in enumerate(paragraphs):
+        story.append(Paragraph(para.strip(), text_style))
+        
+        if i < len(image_paths) and os.path.exists(image_paths[i]):
+            img = Image(image_paths[i], width=15*cm, height=10*cm)
+            story.append(img)
+            story.append(Spacer(1, 1*cm))
+
+        if i < len(paragraphs) - 1:
+            story.append(PageBreak())
+
+    doc.build(story)
+    return output_path
+
+# Gradio 인터페이스
+def process_story(prompt: str):
+    """스토리 생성 처리"""
+    story, paragraphs = generate_story(prompt)
+    return story, gr.update(visible=True), paragraphs
+
+def generate_images_batch(paragraphs: List[str]):
+    """배치로 이미지 생성 (진행률 표시)"""
+    from tqdm import tqdm
+    
+    image_paths = []
+    for i, para in tqdm(enumerate(paragraphs), total=len(paragraphs), desc="이미지 생성"):
+        img_path = generate_image(para, paragraph_num=i+1)
+        image_paths.append(img_path)
+        
+        if device == "cuda":
+            torch.cuda.empty_cache()
+    
+    return image_paths
+
+def create_storybook(story_text: str, paragraphs: List[str]):
+    """스토리북 PDF 생성"""
+    # 이미지 생성
+    image_paths = generate_images_batch(paragraphs)
+    
+    # PDF 생성
+    pdf_path = create_pdf(story_text, image_paths)
+    
+    # 이미지 갤러리용 데이터
+    images = [PILImage.open(path) for path in image_paths]
+    
+    return images, pdf_path
+
+# Gradio UI
+with gr.Blocks(title="AI 스토리북 저작 도구") as app:
+    gr.Markdown("# AI 스토리북 저작 도구")
+    
+    with gr.Row():
+        with gr.Column():
+            prompt_input = gr.Textbox(
+                label="스토리 주제 입력",
+                placeholder="예: 스타트업 창업 성공 스토리",
+                lines=2
+            )
+            generate_btn = gr.Button("스토리 생성", variant="primary")
+            
+            story_output = gr.Textbox(
+                label="생성된 스토리",
+                lines=15,
+                interactive=True
+            )
+            
+            create_book_btn = gr.Button(
+                "스토리북 생성 (이미지 + PDF)",
+                variant="secondary",
+                visible=False
+            )
+        
+        with gr.Column():
+            image_gallery = gr.Gallery(
+                label="생성된 이미지",
+                show_label=True,
+                elem_id="gallery",
+                columns=2,
+                rows=3,
+                height="auto"
+            )
+            
+            pdf_output = gr.File(
+                label="PDF 다운로드",
+                visible=True
+            )
+    
+    # 상태 저장
+    paragraphs_state = gr.State([])
+    
+    # 이벤트 핸들러
+    generate_btn.click(
+        fn=process_story,
+        inputs=[prompt_input],
+        outputs=[story_output, create_book_btn, paragraphs_state]
+    )
+    
+    create_book_btn.click(
+        fn=create_storybook,
+        inputs=[story_output, paragraphs_state],
+        outputs=[image_gallery, pdf_output]
+    )
+
+if __name__ == "__main__":
+    app.launch(share=True)

requirements.txt

@@ -0,0 +1,18 @@
+--extra-index-url https://download.pytorch.org/whl/cu128
+torch==2.7.1+cu128
+torchvision==0.22.1+cu128
+torchaudio==2.7.1+cu128
+# Hugging Face 및 관련 라이브러리
+transformers==4.52.4
+diffusers==0.33.1
+accelerate==1.7.0
+bitsandbytes==0.46.0
+
+# 기타 필수 패키지
+gradio==5.34.0
+sqlmodel==0.0.24
+reportlab==4.4.1
+pillow==11.2.1
+
+# character_lora_fine_tuning 사용 시 필요
+peft==0.15.2