app.py

import gradio as gr 
import numpy as np
import random
import torch
import os
from diffusers import DiffusionPipeline
from openai import OpenAI
from PIL import Image, ImageDraw

# 建立 OpenAI client（讀環境變數）
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "stabilityai/sdxl-turbo"

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
pipe = pipe.to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024


# 1️⃣ 中文 → 英文繪圖 prompt
def generate_prompt(chinese_text):
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {
                "role": "system", 
                "content": (
                    "你是一位專業的海報設計師和健康新知專家。你的任務是將使用者提供的健康主題，轉換為一個專為 SDXL 模型優化的提示詞（Prompt），以生成一張視覺上引人入勝且資訊清晰的健康新知海報。"
                    "請確保提示詞包含以下元素："
                    "1.  **海報風格**：必須明確指定為「專業的健康海報設計（professional health poster design）」、「資訊圖表（infographic）」或類似的風格。"
                    "2.  **核心主題**：將使用者提供的健康主題轉換為具體的視覺元素。例如，如果主題是「每日喝水的重要性」，請將其轉化為「一個水杯中充滿水的插畫（an illustration of a water glass full of water）」。"
                    "3.  **視覺細節**：加入能提升海報專業感的細節，例如「色彩鮮明（vibrant colors）」、「簡潔的設計（minimalist design）」、「高解析度（high resolution）」、「清晰的排版（clear typography）」。"
                    "4.  **畫面構圖**：利用提示詞引導構圖，例如「中央特寫（center focus）」或「清晰的背景（clean background）」。"
                    "5.  **負面提示詞**：自動加入常見的負面提示詞，以避免生成低品質、不專業的圖像。例如：`blurry, low quality, deformed, messy, amateur, text, watermark`。"
                    "**格式要求**："
                    "* 只輸出一個完整的提示詞，不要有任何解釋或額外文字。"
                    "* 請將正向提示詞與負面提示詞分開，中間用逗號隔開。"
                    "* 所有提示詞都使用英文。"
                    "**範例**："
                    "如果使用者輸入：`每日喝水的重要性`"
                    "請輸出：`a professional health poster design about daily hydration, an illustration of a water glass full of water, clean and vibrant colors, minimalist style, high resolution, clear typography --no blurry, low quality, deformed, messy, amateur, text, watermark`"
                )
            },
            {"role": "user", "content": chinese_text}
        ],
        max_tokens=150,
        temperature=0.7,
    )
    return response.choices[0].message.content.strip()


# 2️⃣ 繪圖主函式
def infer(
    prompt,
    negative_prompt,
    seed,
    randomize_seed,
    width,
    height,
    guidance_scale,
    num_inference_steps,
    white_width,
    white_height,
    white_alpha,      # 透明度
    corner_radius,    # 圓角
    progress=gr.Progress(track_tqdm=True),
):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    generator = torch.Generator().manual_seed(seed)

    # 🔹 GPT 產生完整英文 prompt
    final_prompt = generate_prompt(prompt)

    # 強化的負面 prompt
    strong_negative = (
        "text, words, letters, typography, logo, watermark, signature, "
        "messy details, clutter, objects in the center, elements overlapping the central blank area"
    )
    if negative_prompt:
        strong_negative += ", " + negative_prompt

    # 生成圖像
    image = pipe(
        prompt=final_prompt,
        negative_prompt=strong_negative,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        width=width,
        height=height,
        generator=generator,
    ).images[0].convert("RGBA")

    # 🔹 建立透明圖層，畫半透明白底
    overlay = Image.new("RGBA", (width, height), (0, 0, 0, 0))
    draw = ImageDraw.Draw(overlay)

    center_x, center_y = width // 2, height // 2
    x0 = center_x - white_width // 2
    y0 = center_y - white_height // 2
    x1 = center_x + white_width // 2
    y1 = center_y + white_height // 2

    draw.rounded_rectangle(
        [x0, y0, x1, y1],
        radius=corner_radius,
        fill=(255, 255, 255, white_alpha)  # 白色 + 透明度
    )

    # 疊合到原圖
    image = Image.alpha_composite(image, overlay).convert("RGB")

    return image, seed, final_prompt


# 範例
examples = [
    "多吃水果促進健康",
    "運動對小朋友的好處",
    "保持醫院清潔的重要性",
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

# 3️⃣ Gradio UI
with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(" # 健康布告欄產生器 🩺")

        with gr.Row():
            prompt = gr.Text(
                label="輸入中文主題",
                max_lines=2,
                placeholder="輸入布告欄主題，例如：多吃水果促進健康",
            )
            run_button = gr.Button("產生圖像", variant="primary")

        result = gr.Image(label="Result", show_label=False)
        prompt_box = gr.Textbox(label="英文繪圖 Prompt（AI 產生）", interactive=False)

        with gr.Accordion("進階設定", open=False):
            negative_prompt = gr.Text(label="Negative prompt", placeholder="額外要避免的內容")

            seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
            randomize_seed = gr.Checkbox(label="隨機種子", value=True)

            with gr.Row():
                width = gr.Slider(label="寬度", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=384)
                height = gr.Slider(label="高度", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=384)

            with gr.Row():
                guidance_scale = gr.Slider(label="Guidance scale", minimum=0.0, maximum=15.0, step=0.5, value=8)
                num_inference_steps = gr.Slider(label="步數", minimum=1, maximum=50, step=1, value=25)

            with gr.Row():
                white_width = gr.Slider(label="白底寬度", minimum=50, maximum=800, step=10, value=300)
                white_height = gr.Slider(label="白底高度", minimum=50, maximum=800, step=10, value=300)

            with gr.Row():
                white_alpha = gr.Slider(label="白底透明度 (0透明-255不透明)", minimum=0, maximum=255, step=5, value=210)
                corner_radius = gr.Slider(label="圓角大小", minimum=0, maximum=400, step=5, value=50)

        gr.Examples(examples=examples, inputs=[prompt])

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn=infer,
        inputs=[
            prompt,
            negative_prompt,
            seed,
            randomize_seed,
            width,
            height,
            guidance_scale,
            num_inference_steps,
            white_width,
            white_height,
            white_alpha,
            corner_radius,
        ],
        outputs=[result, seed, prompt_box],
    )

if __name__ == "__main__":
    demo.launch()