image_generator.py

import os
import sys
import shutil
import uuid
import random
import logging
from PIL import Image, ImageDraw, ImageFont

# 設定日誌記錄，便於開發與除錯
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)

# ==============================================================================
# 改為從環境變數讀取金鑰，保護資安
# ==============================================================================
GEMINI_API_KEY = os.environ.get("GOOGLE_API_KEY")
HF_TOKEN = os.environ.get("HF_TOKEN")

# 固定的生圖寬度與高度 (1024x1024)
DEFAULT_WIDTH = 1024
DEFAULT_HEIGHT = 1024

# 預設的俗俗早安圖問候語，當 Gemini API 金鑰未提供時作為 Fallback 備用
PRESET_GREETINGS = [
    "早安！今天也要假裝系統正常！",
    "早安！人生卡住時，請先重開機！",
    "早安！努力不一定成功，但會很累！",
    "早安！腦袋尚未開機，請多包涵！",
    "早安！今天的你，比昨天更像待機中！",
    "早安！不要放棄，反正也沒人接住你！",
    "早安！上課像登入失敗，但人要有光！",
    "早安！作業不會寫，太陽還是會升起！",
    "早安！群組安靜不是冷漠，是大家都沒電！",
    "早安！願你今天的 Wi-Fi 比人生穩定！"
]

# 頂部大字 Banner 的預設選用標題
PRESET_HEADERS = [
    "早安",
    "平安喜樂",
    "吉祥如意",
    "福氣滿滿",
    "吉祥安康",
    "順心如意"
]

# 經典長輩圖必備霓虹彩虹漸層字體配色系列
RAINBOW_COLORS = [
    "#FF0033",  # 霓虹紅
    "#FF6600",  # 霓虹橘
    "#FFFF00",  # 鮮艷黃
    "#33FF33",  # 螢光綠
    "#00FFFF",  # 霓虹藍/青色
    "#CC33FF",  # 亮紫色
    "#FF3399"   # 桃紅色
]

def translate_to_english(text):
    try:
        import requests
        url = f"https://translate.googleapis.com/translate_a/single?client=gtx&sl=auto&tl=en&dt=t&q={requests.utils.quote(text)}"
        r = requests.get(url, timeout=5)
        if r.status_code == 200:
            translated = r.json()[0][0][0]
            logger.info(f"自動翻譯成功: '{text}' -> '{translated}'")
            return translated
    except Exception as e:
        logger.error(f"翻譯 API 呼叫失敗: {e}")
    return text

def get_chinese_font(font_size, font_name="kaiu"):
    windir = os.environ.get('WINDIR', 'C:\\Windows')
    paths = []
    
    paths.extend([
        "kaiu.ttf",                  
        "static/kaiu.ttf",           
        os.path.join(windir, 'Fonts', 'kaiu.ttf'),    
        os.path.join(windir, 'Fonts', 'msjhbd.ttc'),  
        os.path.join(windir, 'Fonts', 'msjh.ttc'),    
        os.path.join(windir, 'Fonts', 'simsun.ttc'),  
        "/usr/share/fonts/truetype/noto/NotoSansCJK-Regular.ttc",
        "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
        "/usr/share/fonts/truetype/wqy/wqy-microhei.ttc",
        "/usr/share/fonts/truetype/wqy/wqy-zenhei.ttc",
        "/usr/share/fonts/truetype/droid/DroidSansFallback.ttf"
    ])
    
    for path in paths:
        if os.path.exists(path):
            try:
                if path.endswith('.ttc'):
                    logger.info(f"成功載入 TTC 字型: {path}")
                    return ImageFont.truetype(path, font_size, index=0)
                else:
                    logger.info(f"成功載入 TTF 字型: {path}")
                    return ImageFont.truetype(path, font_size)
            except Exception as e:
                logger.warning(f"載入字型 {path} 失敗: {e}")
                continue
                
    logger.warning("未偵測到任何系統字型，將使用 Pillow 預設字型。")
    try:
        return ImageFont.load_default()
    except Exception:
        return None

def wrap_text(text, font, max_width, draw):
    lines = []
    current_line = ""
    for char in text:
        test_line = current_line + char
        w = draw.textlength(test_line, font=font)
        if w <= max_width:
            current_line = test_line
        else:
            if current_line:
                lines.append(current_line)
            current_line = char
    if current_line:
        lines.append(current_line)
    return lines

def generate_text_with_gemini(api_key, query):
    if not api_key:
        logger.info("未偵測到 Gemini API Key。將使用內建隨機賀詞。")
        return random.choice(PRESET_GREETINGS)
        
    try:
        import google.generativeai as genai
        genai.configure(api_key=api_key)
        model = genai.GenerativeModel('gemini-2.5-flash')
        
        prompt = (
             f"請針對「{query}」寫一句繁體中文早安圖文字。\n"
             f"風格要求：台灣長輩圖、荒謬、38、假勵志、欠吐槽、適合群組聊天。\n"
             f"語氣要像一個很有自信但其實很荒謬的早安機器人。\n"
             f"字數限制：12 到 22 字。\n"
             f"內容方向：可以嘲諷抽象人生、早八、上班、作業、拖延、群組沉默、機器人自己。\n"
             f"安全限制：不可攻擊特定真人、族群、性別、外貌、疾病、政治立場或宗教。\n"
             f"不要髒話，不要人身攻擊，不要太黑暗。\n"
             f"要讓群組成員看了想回：『這什麼鬼啦』。\n"
             f"範例：\n"
             f"- 早安，今天也要假裝系統正常\n"
             f"- 早安，努力不一定成功但很耗電\n"
             f"- 早安，人生卡住請重新整理\n"
             f"- 早安，腦袋離線也是一種生活\n"
             f"- 早安，群組沒人講話我先亂講\n"
             f"請只回傳一句中文早安文字，不要解釋，不要加引號。"
        )
        response = model.generate_content(prompt)
        text = response.text.strip().replace('"', '').replace('「', '').replace('」', '')
        logger.info(f"Gemini 生成的長輩賀詞: {text}")
        return text
    except Exception as e:
        logger.error(f"Gemini 賀詞生成錯誤: {e}")
        return random.choice(PRESET_GREETINGS)

def expand_prompt_with_gemini(api_key, query):
    query_en = translate_to_english(query)
    
    default_expanded = (
           f"A funny, absurd, and slightly stupid Taiwanese good morning meme image featuring {query_en} as the main subject, "
           f"placed clearly in the center of the image. "
           f"The image should look like an over-the-top Taiwanese elder-style morning greeting card mixed with chaotic internet meme humor. "
           f"Use extremely colorful flowers, dramatic golden sunrise, sparkles, glowing clouds, rainbow decorations, fake inspirational poster energy, "
           f"retro 1990s digital clip-art aesthetics, and an exaggerated ridiculous composition. "
           f"The mood should be harmless, silly, flamboyant, awkwardly cheerful, and easy for group chat members to roast. "
           f"No readable text inside the image, because Traditional Chinese text will be added later by the program. "
           f"High resolution, colorful, detailed, humorous, meme-like, absurd, ridiculous."
    )
    
    if not api_key:
        logger.info("未偵測到 Gemini API Key。使用預設封裝 Prompt。")
        return default_expanded
        
    try:
        import google.generativeai as genai
        genai.configure(api_key=api_key)
        model = genai.GenerativeModel('gemini-2.5-flash')
        
        prompt = (
             f"Please translate the Chinese phrase '{query}' into English, "
             f"then expand it into a detailed image generation prompt for a funny, absurd Taiwanese good morning meme image.\n\n"
             f"Creative direction:\n"
             f"The image should feel like a chaotic Taiwanese elder-style good morning card, "
             f"but upgraded into a ridiculous internet meme that group chat members would want to roast.\n\n"
             f"Requirements:\n"
             f"1. The main subject must clearly be related to: '{query_en}'.\n"
             f"2. Make the scene visually absurd, silly, flamboyant, and overly confident in a stupid way.\n"
             f"3. Combine Taiwanese elder-style good morning card aesthetics with meme humor: "
             f"dramatic sunrise, glowing clouds, lotus flowers, sparkles, rainbow decorations, "
             f"oversaturated colors, retro 1990s clip-art, fake inspirational poster style, and chaotic composition.\n"
             f"4. The image should look harmless, funny, awkwardly cheerful, and easy to roast in a group chat.\n"
             f"5. Add one ridiculous visual gag related to the subject, such as the subject acting too serious, "
             f"posing heroically, failing confidently, or doing something meaningless with great determination.\n"
             f"6. Do not include real public figures, political symbols, hate symbols, violence, sexual content, horror, or offensive stereotypes.\n"
             f"7. Do not generate readable text inside the image, because Traditional Chinese text will be added later by the program.\n\n"
             f"Only return the final English image prompt. Do not include explanations, labels, or quotation marks."
        )
        response = model.generate_content(prompt)
        expanded = response.text.strip().replace('"', '').replace('「', '').replace('」', '')
        logger.info(f"Gemini 擴寫的英文生圖提示詞: {expanded}")
        return expanded
    except Exception as e:
        logger.error(f"Gemini 提示詞擴寫錯誤: {e}")
        return default_expanded

def draw_rainbow_text(draw, text, x_center, y_start, font, stroke_width=6, stroke_fill="#000000"):
    if not font:
        return y_start
        
    total_w = 0
    char_widths = []
    for char in text:
        w = draw.textlength(char, font=font)
        char_widths.append(w)
        total_w += w
        
    start_x = x_center - (total_w / 2)
    current_x = start_x
    
    for idx, char in enumerate(text):
        color = RAINBOW_COLORS[idx % len(RAINBOW_COLORS)]
        
        draw.text(
            (current_x, y_start),
            char,
            font=font,
            fill=color,
            stroke_width=stroke_width,
            stroke_fill=stroke_fill
        )
        current_x += char_widths[idx]
        
    bbox = font.getbbox("早")
    char_h = bbox[3] - bbox[1] if bbox else 50
    return y_start + char_h + 10

def overlay_text_on_image(image_path, greeting_text, output_path):
    try:
        img = Image.open(image_path).convert("RGB")
        if img.size != (DEFAULT_WIDTH, DEFAULT_HEIGHT):
            img = img.resize((DEFAULT_WIDTH, DEFAULT_HEIGHT), Image.Resampling.LANCZOS)
            
        draw = ImageDraw.Draw(img)
        
        header_font = get_chinese_font(120, "kaiu")
        body_font = get_chinese_font(52, "kaiu")
        footer_font = get_chinese_font(32, "kaiu")
        
        header_text = f"🌸 {random.choice(PRESET_HEADERS)} 🌸"
        header_y = 60
        if header_font:
            draw.text(
                (DEFAULT_WIDTH / 2, header_y),
                header_text,
                font=header_font,
                fill="#FFFF00",
                anchor="ma",
                stroke_width=8,
                stroke_fill="#000000"
            )
            
        body_y_start = 740
        if body_font:
            lines = wrap_text(greeting_text, body_font, 920, draw)
            current_y = body_y_start
            for line in lines:
                current_y = draw_rainbow_text(
                    draw,
                    line,
                    x_center=DEFAULT_WIDTH / 2,
                    y_start=current_y,
                    font=body_font,
                    stroke_width=6,
                    stroke_fill="#000000"
                )
                
        footer_text = "👍 認同請分享，祝您平安喜樂！ 👍"
        footer_y = 930
        if footer_font:
            draw.text(
                (DEFAULT_WIDTH / 2, footer_y),
                footer_text,
                font=footer_font,
                fill="#33FF33",
                anchor="ma",
                stroke_width=5,
                stroke_fill="#000000"
            )
            
        img.save(output_path, "JPEG", quality=90)
        logger.info(f"早安圖加工疊加完成，存檔至: {output_path}")
        return True
    except Exception as e:
        logger.error(f"Pillow 文字疊加失敗: {e}")
        try:
            shutil.copy(image_path, output_path)
            return True
        except Exception:
            return False

def generate_good_morning_image(user_input, api_key=None, custom_greeting=None):
    resolved_api_key = api_key or GEMINI_API_KEY

    query = user_input.strip()
    if query.startswith("@create image"):
        query = query.replace("@create image", "", 1).strip()
    if not query:
        query = "蓮花"
        
    logger.info(f"開始處理早安圖主題任務: {query}")
    
    if custom_greeting:
        greeting_text = custom_greeting
    else:
        greeting_text = generate_text_with_gemini(resolved_api_key, query)
        
    expanded_prompt = expand_prompt_with_gemini(resolved_api_key, query)
    
    logger.info("呼叫 Gradio Space 端點 (stabilityai/stable-diffusion-3.5-large)...")
    try:
        from gradio_client import Client
        token = HF_TOKEN
        client = Client("stabilityai/stable-diffusion-3.5-large", token=token)
        
        result = client.predict(
            prompt=expanded_prompt,
            negative_prompt="blurry, low quality, bad anatomy, deformed, ugly, disfigured, poor lighting, out of frame, cropped",
            seed=0,
            randomize_seed=True,
            width=DEFAULT_WIDTH,
            height=DEFAULT_HEIGHT,
            guidance_scale=4.5,
            num_inference_steps=40,
            api_name="/infer",
        )
        
        if result and isinstance(result, tuple) and len(result) > 0:
            temp_image_path = result[0]
        elif isinstance(result, str):
            temp_image_path = result
        else:
            raise ValueError(f"無法識別的 Gradio API 返回格式: {result}")
            
        logger.info(f"SD3.5 基礎底圖繪製成功，暫存於: {temp_image_path}")
        
        data_dir = os.path.abspath("data")
        os.makedirs(data_dir, exist_ok=True)
        
        image_id = str(uuid.uuid4())
        final_image_name = f"{image_id}.jpg"
        final_image_path = os.path.join(data_dir, final_image_name)
        
        success = overlay_text_on_image(temp_image_path, greeting_text, final_image_path)
        if not success:
            raise RuntimeError("Pillow 後製疊加文字發生崩潰失敗")
            
        return {
            "success": True,
            "id": image_id,
            "query": query,
            "greeting_text": greeting_text,
            "expanded_prompt": expanded_prompt,
            "image_path": final_image_path,
            "filename": final_image_name
        }
        
    except Exception as e:
        logger.error(f"整條生圖流水線執行失敗: {e}")
        return {
            "success": False,
            "error": str(e)
        }