app.py

import os
import re
import json
import uuid
import shutil
import pathlib
import gradio as gr
from PIL import Image

from generator import generate_marketing_assets
from tts_subtitles import synthesize_tts_openai, transcribe_to_srt_openai
from video import (
    compose_video_with_subtitles,
    make_background_image,
    find_japanese_font_path,
    rasterize_svg_to_png,
)
from doc_ingest import summarize_documents_and_urls

DEFAULT_MODEL = os.getenv("OPENAI_LLM_MODEL", "gpt-4o-mini")
DEFAULT_TTS_MODEL = os.getenv("OPENAI_TTS_MODEL", "tts-1")
DEFAULT_WHISPER_MODEL = os.getenv("OPENAI_WHISPER_MODEL", "whisper-1")

VOICE_CHOICES = ["alloy", "verse", "breeze", "bright", "calm"]

OUTPUT_BASE = os.getenv("OUTPUT_DIR", "/tmp/outputs")
pathlib.Path(OUTPUT_BASE).mkdir(parents=True, exist_ok=True)

EXAMPLE = {
    "product_name": "FLDapp HbA1c測定",
    "value_props": "5秒で測定/アプリ自動記録/医療機関連携",
    "target": "30-50代の健康意識が高いビジネスパーソン",
    "tone": "信頼性とスピード感",
    "lang": "ja",
    "seconds": 20,
    "voice": "alloy",
}

def _save_upload(x, out_dir, name_hint):
    if not x:
        return None
    src = x if isinstance(x, str) else x.get("path")
    if not src or not os.path.exists(src):
        return None
    ext = os.path.splitext(src)[1].lower()
    dst = os.path.join(out_dir, f"{name_hint}{ext}")
    if ext == ".svg":
        png_dst = dst.replace(".svg", ".png")
        rasterize_svg_to_png(src, png_dst, width=1024)
        return png_dst
    shutil.copy2(src, dst)
    return dst

def _save_upload_multi(xs, out_dir, name_prefix):
    paths = []
    if not xs:
        return paths
    if isinstance(xs, list):
        for i, item in enumerate(xs, 1):
            p = _save_upload(item, out_dir, f"{name_prefix}_{i}")
            if p:
                paths.append(p)
    else:
        p = _save_upload(xs, out_dir, f"{name_prefix}_1")
        if p:
            paths.append(p)
    return paths

def _parse_urls(url_text: str) -> list[str]:
    if not url_text:
        return []
    tokens = re.split(r"[\s,;]+", url_text.strip())
    urls = []
    for u in tokens:
        u = u.strip()
        if u.startswith("http://") or u.startswith("https://"):
            urls.append(u)
    # 重複除去
    return list(dict.fromkeys(urls))

def run_pipeline(
    product_name, value_props, target, tone, lang, seconds, voice,
    # 参照資料/URL
    doc_files, url_texts, use_docs,
    # 合成素材
    logo_file, bg_image_file, bg_video_file, bgm_file, font_file,
    logo_position, logo_width_px,
    # ★ 新規: トグル類
    show_subtitles, show_title_overlay, title_position, title_fontsize, bg_text_draw
):
    if not os.getenv("OPENAI_API_KEY"):
        raise gr.Error("OPENAI_API_KEY が設定されていません（SpacesのSecretsに追加）。")

    run_id = uuid.uuid4().hex[:8]
    out_dir = os.path.join(OUTPUT_BASE, run_id)
    os.makedirs(out_dir, exist_ok=True)

    # アップロード保存
    doc_paths = _save_upload_multi(doc_files, out_dir, "doc")
    logo_path = _save_upload(logo_file, out_dir, "logo")
    bg_img_user = _save_upload(bg_image_file, out_dir, "bg_user")
    bg_video_user = _save_upload(bg_video_file, out_dir, "bg_user_video")
    bgm_path = _save_upload(bgm_file, out_dir, "bgm")
    font_path = _save_upload(font_file, out_dir, "custom_font")

    # URL整形
    urls = _parse_urls(url_texts)

    # フォント
    if font_path and os.path.exists(font_path):
        chosen_font = font_path
        font_dir = os.path.dirname(font_path)
    else:
        chosen_font = find_japanese_font_path()
        font_dir = os.path.dirname(chosen_font) if chosen_font else None

    # 参照資料/URL を要約（任意）
    extra_context = None
    if use_docs and (doc_paths or urls):
        try:
            extra_context = summarize_documents_and_urls(
                doc_paths, urls, model=DEFAULT_MODEL, lang=lang, target_char_len=2000, work_dir=out_dir
            )
        except Exception:
            extra_context = None

    try:
        # 1) LLMでコピー/台本生成（資料要約を反映）
        assets = generate_marketing_assets(
            product_name=product_name,
            value_props=value_props,
            target=target,
            tone=tone,
            lang=lang,
            seconds=int(seconds),
            model=DEFAULT_MODEL,
            extra_context=extra_context,
        )
        headline = assets["headline"]
        video_script = assets["video_script"]
        thumbnail_text = assets["thumbnail_text"]
        ogp = assets["ogp"]

        # 2) TTS
        audio_path = os.path.join(out_dir, "narration.mp3")
        synthesize_tts_openai(
            text=video_script,
            out_path=audio_path,
            voice=voice,
            model=DEFAULT_TTS_MODEL,
            format="mp3",
        )
        if not os.path.exists(audio_path):
            raise gr.Error("音声生成に失敗しました。")

        # 3) 字幕（SRT）
        srt_path = os.path.join(out_dir, "captions.srt")
        srt_text = transcribe_to_srt_openai(audio_path=audio_path, model=DEFAULT_WHISPER_MODEL)
        with open(srt_path, "w", encoding="utf-8") as f:
            f.write(srt_text)

        # 4) 背景
        if bg_img_user and os.path.exists(bg_img_user):
            bg_path = bg_img_user
        else:
            bg_path = os.path.join(out_dir, "bg.png")
            make_background_image(
                out_path=bg_path,
                title=headline,
                subtitle=thumbnail_text,
                lang=lang,
                font_path=chosen_font,
                draw_text=bool(bg_text_draw),   # ★ 追加
            )

        # 5) 動画合成（字幕/タイトルのON/OFFに対応）
        video_out = os.path.join(out_dir, "short.mp4")
        compose_video_with_subtitles(
            image_path=bg_path if not bg_video_user else None,
            audio_path=audio_path,
            srt_path=srt_path,
            out_path=video_out,
            width=1080,
            height=1920,
            fps=30,
            bg_video_path=bg_video_user,
            logo_path=logo_path,
            logo_pos=logo_position,
            logo_width=int(logo_width_px),
            bgm_path=bgm_path,
            fonts_dir=font_dir,
            include_subtitles=bool(show_subtitles),                         # ★ 追加
            title_text=headline if bool(show_title_overlay) else None,      # ★ 追加
            title_pos=title_position,
            title_fontfile=chosen_font,
            title_fontsize=int(title_fontsize),
            title_box=True,
        )
        if not os.path.exists(video_out):
            raise gr.Error("動画合成に失敗しました。")

        # 6) OGP画像
        ogp_img_path = os.path.join(out_dir, "ogp.png")
        base = Image.open(bg_path).convert("RGBA").resize((1200, 630))
        if logo_path and os.path.exists(logo_path):
            try:
                logo = Image.open(logo_path).convert("RGBA")
                w = int(1200 * (logo_width_px / 1080))
                h = int(logo.height * (w / logo.width))
                logo = logo.resize((w, h), Image.LANCZOS)
                x = 1200 - w - 32 if "right" in logo_position else 32
                y = 32 if "top" in logo_position else 630 - h - 32
                base.alpha_composite(logo, (x, y))
            except Exception:
                pass
        base.convert("RGB").save(ogp_img_path)

        return (
            headline,
            video_script,
            thumbnail_text,
            json.dumps(ogp, ensure_ascii=False, indent=2),
            audio_path,
            srt_path,
            video_out,
            ogp_img_path,
        )
    except gr.Error:
        raise
    except Exception as e:
        raise gr.Error(f"処理中にエラー: {type(e).__name__}: {e}")

with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("""
    # 🎬 マルチモーダル要素→素材ジェネレーター
    参照資料（PDF/TXT/DOCX/URL）を要約してコピーに反映し、
    **字幕のON/OFF** と **動画内タイトル（ヘッドライン）テロップのON/OFF** を切替可能。
    ロゴ/背景/BGM/フォントを使い、日本語フォントで字幕焼き込みの縦型ショート動画を生成します。
    """)

    with gr.Row():
        with gr.Column():
            # 元データ
            product_name = gr.Textbox(label="製品名")
            value_props = gr.Textbox(label="訴求点（カンマ区切り推奨）")
            target = gr.Textbox(label="ターゲット")
            tone = gr.Textbox(label="トーン", value="信頼性とスピード感")
            lang = gr.Dropdown(["ja", "en"], value="ja", label="言語")
            seconds = gr.Slider(8, 60, value=20, step=1, label="動画尺（秒・台本目安）")
            voice = gr.Dropdown(VOICE_CHOICES, value="alloy", label="ナレーション音声（OpenAI TTS）")

            # 参照資料/URL
            doc_files = gr.File(label="参照資料（PDF/TXT/DOCX 複数可）", file_count="multiple", file_types=["file"])
            url_texts = gr.Textbox(label="参照URL（改行/スペース/カンマ区切りで複数可）", lines=3, placeholder="https://example.com/page1\nhttps://example.com/page2")
            use_docs = gr.Checkbox(value=True, label="資料/URLを要約してコピーに反映する（推奨）")

            # 合成素材
            logo_file = gr.File(label="ロゴ画像（PNG/JPG/SVG対応）", file_types=["image"])
            bg_image_file = gr.File(label="背景画像（任意・PNG/JPG）", file_types=["image"])
            bg_video_file = gr.File(label="背景動画（任意・mp4/mov など）", file_types=["video"])
            bgm_file = gr.File(label="BGM（任意・mp3/wav）", file_types=["audio"])
            font_file = gr.File(label="カスタムフォント（任意・TTF/OTF）", file_types=["file"])
            logo_position = gr.Dropdown(["top-left", "top-right", "bottom-left", "bottom-right"], value="top-right", label="ロゴ位置")
            logo_width_px = gr.Slider(64, 800, value=240, step=1, label="ロゴ幅（px）")

            # ★ 新規トグル
            show_subtitles = gr.Checkbox(value=True, label="字幕を焼き込む")
            show_title_overlay = gr.Checkbox(value=False, label="動画内にタイトル（ヘッドライン）テロップを表示")
            title_position = gr.Dropdown(["top", "center", "bottom"], value="top", label="タイトル位置")
            title_fontsize = gr.Slider(24, 120, value=64, step=1, label="タイトルフォントサイズ")
            bg_text_draw = gr.Checkbox(value=True, label="内製背景にテキストを描く（headline/sub）")

            run_btn = gr.Button("生成する", variant="primary")

        with gr.Column():
            headline = gr.Textbox(label="見出し/ヘッドライン")
            video_script = gr.Textbox(label="短尺動画 用 台本", lines=12)
            thumbnail_text = gr.Textbox(label="サムネ文言")
            ogp_json = gr.Code(label="OGP JSON（title/description/alt）", language="json")
            audio = gr.File(label="音声（mp3）")
            srt = gr.File(label="字幕（srt）")
            video = gr.File(label="動画（mp4）")
            ogp_img = gr.File(label="OGP画像（png）")

    # Examples は非ファイル入力のみ
    gr.Examples(
        examples=[
            [
                EXAMPLE["product_name"], EXAMPLE["value_props"], EXAMPLE["target"], EXAMPLE["tone"], EXAMPLE["lang"],
                EXAMPLE["seconds"], EXAMPLE["voice"],
                "", True,                   # url_texts, use_docs
                "top-right", 240,           # logo_position, logo_width_px
                True, False, "top", 64,     # show_subtitles, show_title_overlay, title_position, title_fontsize
                True                        # bg_text_draw
            ]
        ],
        inputs=[
            product_name, value_props, target, tone, lang, seconds, voice,
            url_texts, use_docs,
            logo_position, logo_width_px,
            show_subtitles, show_title_overlay, title_position, title_fontsize, bg_text_draw
        ],
        label="クイックフィル（資料/URL/ファイル類は任意でアップロードしてください）",
        cache_examples=False,
    )

    run_btn.click(
        fn=run_pipeline,
        inputs=[
            product_name, value_props, target, tone, lang, seconds, voice,
            doc_files, url_texts, use_docs,
            logo_file, bg_image_file, bg_video_file, bgm_file, font_file, logo_position, logo_width_px,
            show_subtitles, show_title_overlay, title_position, title_fontsize, bg_text_draw
        ],
        outputs=[headline, video_script, thumbnail_text, ogp_json, audio, srt, video, ogp_img],
        api_name="generate",
    )

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