何はなくても自動化だ！YouTube自動化収益獲得チャレンジ中！中編

def build_video_from_csv(csv_path) -> Optional[str]: title = None; desc = None; tags = None rows = [] with open(csv_path, "r", encoding="utf-8") as f: for line in f: s = line.strip() if s.startswith("#title,"): title = s[7:]; continue if s.startswith("#description,"): desc = s[13:]; continue if s.startswith("#tags,"): tags = s[6:]; continue rows.append(line) if not rows: return None df = pd.read_csv(io.StringIO("".join(rows))).fillna("") voice_segments: List[AudioSegment] = [] frames: List[ImageClip] = [] bgm_path = CFG.get("bgm_path","").strip() bgm_seg = AudioSegment.from_file(bgm_path) + int(CFG.get("bgm_gain_db",-12)) if bgm_path and os.path.exists(bgm_path) else None tmp_dir = os.path.join(".tmp_kamishibai", os.path.splitext(os.path.basename(csv_path))[0]) os.makedirs(tmp_dir, exist_ok=True) for idx, row in df.iterrows(): text = str(row.get("text","")).strip() if not text: continue image = str(row.get("image","")).strip() or None dur_str = str(row.get("duration","")).strip() se = str(row.get("se","")).strip() or None vpath = tts(text, idx, tmp_dir) if vpath and os.path.exists(vpath): vseg = effects.normalize(AudioSegment.from_file(vpath)) else: vseg = AudioSegment.silent(duration=int(1000*float(CFG.get("min_scene_sec",2.5)))) if se and os.path.exists(se): se_seg = AudioSegment.from_file(se) vseg = se_seg.overlay(vseg) if len(se_seg) >= len(vseg) else vseg.overlay(se_seg) min_sec = float(CFG.get("min_scene_sec",2.5)) max_sec = float(CFG.get("max_scene_sec",8.0)) tail = float(CFG.get("tail_silence_sec",0.15)) want = len(vseg)/1000.0 + tail if dur_str: try: want = float(dur_str) except: pass want = max(min_sec, min(max_sec, want)) add_ms = int(max(0, want*1000 - len(vseg))) if add_ms > 0: vseg = vseg + AudioSegment.silent(duration=add_ms) voice_segments.append(vseg) bg = make_bg(image) panel = render_panel(text, is_title=False) frame = Image.alpha_composite(bg.convert("RGBA"), panel).convert("RGB") fp = os.path.join(tmp_dir, f"frame_{idx}.jpg") frame.save(fp, quality=95) clip = ImageClip(fp).set_duration(want) # 線形ズーム（ケンバーンズ風） zs = float(CFG.get("zoom_start",1.02)) ze = float(CFG.get("zoom_end",1.08)) clip = clip.resize(lambda t: zs + (ze - zs) * (t / clip.duration)) frames.append(clip) if not frames: return None video = concatenate_videoclips(frames, method="compose") if voice_segments: voice = AudioSegment.silent(duration=0) for seg in voice_segments: voice += seg if bgm_seg: loops = max(1, int(math.ceil(len(voice)/len(bgm_seg)))) if len(bgm_seg) > 0 else 1 long_bgm = AudioSegment.silent(duration=0) for _ in range(loops): long_bgm += bgm_seg long_bgm = long_bgm[:len(voice)] duck_db = int(CFG.get("ducking_db",-8)) mix = (long_bgm - max(0, duck_db)).overlay(voice) else: mix = voice audio_path = os.path.join(tmp_dir, "mix_audio.mp3") mix.export(audio_path, format="mp3") video = video.set_audio(AudioFileClip(audio_path)) base = os.path.splitext(os.path.basename(csv_path))[0] outname = f"{datetime.now().strftime('%Y%m%d_%H%M')}_{base}.mp4" outpath = os.path.join(CFG["queue_dir"], outname) video.write_videofile(outpath, fps=int(CFG["fps"]), codec="libx264", audio_codec="aac", threads=4, preset="medium", bitrate="4000k") # メタ追記 title2 = (title or CFG.get("default_title","今日の紙芝居 #Shorts")).strip() desc2 = (desc or CFG.get("default_description","自動生成の紙芝居ショートです。#Shorts")).strip() tags2 = (tags or CFG.get("default_tags","#Shorts")).strip() append_captions(outpath, title2, desc2, tags2) shutil.move(csv_path, os.path.join(CFG["done_dir"], os.path.basename(csv_path))) return outpath def main(): targets = sorted(glob.glob("stories/queue/*.csv")) if not targets: print("stories/queue が空です。CSVを置いてください。") return for p in targets: try: print(f"==> 生成: {p}") out = build_video_from_csv(p) if out: print(f"出力: {out}") except Exception as e: print(f"失敗: {p} : {e}") if __name__ == "__main__": main()

kamishibai_maker.py