""" person_swap/preprocess.py — 素材合规化与预处理 (ffmpeg) 职责: 1. probe_video — 探测源视频(时长/分辨率/编码/fps) 2. check_image — 校验人物参考图(可解码/尺寸/非动图) 3. prep_source_video — 源视频合规化: 转码 h264 + 竖屏 9:16 裁切 + 限长截断 4. prep_reference_image — 参考图归一化(缩放至模型常用尺寸上限) 5. extract_preview_frame — 抽帧预览(供前端展示) ⚠️ 本机 ~/.local/bin/ffprobe 是软链到 ffmpeg 的假 ffprobe(v6.0 单二进制), _find_bin 会用 `-version` 验证真实身份; 无真 ffprobe 时走 ffmpeg stderr 解析兜底。 输出目录约定: agent-local/runtime/person_swap/uploads/{task_id}/ """ import json import os import re import shutil import subprocess from pathlib import Path # ── ffmpeg / ffprobe 路径解析 ── def _find_bin(name: str) -> str: """查找真实二进制: 候选目录 + PATH, 且 -version 首行必须含目标名""" cands = [ os.environ.get("FFMPEG_DIR", ""), str(Path.home() / ".local" / "bin"), "/opt/homebrew/bin", "/usr/local/bin", "/usr/bin", ] seen = [] for d in cands: if d: p = Path(d) / name if p.exists(): seen.append(str(p)) w = shutil.which(name) if w: seen.append(w) for cand in seen: try: r = subprocess.run([cand, "-version"], capture_output=True, text=True, timeout=10) if r.returncode == 0 and r.stdout.splitlines() and name in r.stdout.splitlines()[0]: return cand except Exception: continue return name # 找不到就交给系统 PATH, 调用时会自然报错 # ⚠️ 统一用 lib.ffmpeg 的解析：它会**自动查找真正的 ffprobe** # （本机 ~/.local/bin/ffprobe 是 ffmpeg 的符号链接，假货的 -version 首行不含 "ffprobe"， # 一旦没找到真 ffprobe，_find_bin 会兜底返回 "ffprobe" → 仍会踩到假货） try: os.sys.path.insert(0, str(Path(__file__).resolve().parent.parent)) from lib.ffmpeg import FFMPEG, FFPROBE # noqa: E402 except Exception: FFMPEG = _find_bin("ffmpeg") FFPROBE = _find_bin("ffprobe") # ── 常用约束 (可被 person_swap 配置覆盖) ── DEFAULT_MAX_DURATION = 10 # 单条源视频最长秒数(超长提示分段) DEFAULT_ORIENTATION = "9:16" # 输出竖屏 DEFAULT_IMAGE_MAX = 2048 # 参考图长边上限 _SCRIPT_ROOT = Path(__file__).resolve().parent.parent if str(_SCRIPT_ROOT) not in os.sys.path: os.sys.path.insert(0, str(_SCRIPT_ROOT)) from lib.logger import get_logger from lib.config import load_config logger = get_logger("person_swap") def _cfg() -> dict: cfg = load_config() return cfg.get("person_swap", {}) or {} def _run(cmd: list, timeout: int = 120) -> subprocess.CompletedProcess: logger.debug("执行: %s", " ".join(str(c) for c in cmd)) return subprocess.run(cmd, capture_output=True, text=True, timeout=timeout) # ════════════════════════════════════════════════════════════ # 探测与校验 # ════════════════════════════════════════════════════════════ def _parse_ffmpeg_info(path: str) -> dict: """ffmpeg -i stderr 解析兜底 (无真 ffprobe 时使用)""" r = subprocess.run([FFMPEG, "-hide_banner", "-i", path], capture_output=True, text=True, timeout=30) text = r.stderr dur_s = 0.0 m = re.search(r"Duration:\s*(\d+):(\d+):([\d.]+)", text) if m: h, mi, s = int(m.group(1)), int(m.group(2)), float(m.group(3)) dur_s = h * 3600 + mi * 60 + s # 分辨率组首位 1-9: 排除 0x31637661(avc1 十六进制) 这类误匹配 m = re.search(r"Stream #0[^\n]*?Video:\s*([\w-]+)[^\n]*?([1-9]\d{1,4})x([1-9]\d{1,4})", text) if not m: raise ValueError(f"无法识别媒体流: {text.strip()[-200:]}") codec, w, h = m.group(1), int(m.group(2)), int(m.group(3)) fps = "" mf = re.search(r"(\d+(?:\.\d+)?)\s*fps", text) if mf: fps = mf.group(1) return {"duration_sec": round(dur_s, 2), "width": w, "height": h, "codec": codec, "fps": fps} def probe_video(path: str) -> dict: """探测视频元信息, 失败抛 ValueError; 优先 ffprobe JSON, 兜底 ffmpeg 解析""" if FFPROBE != "ffprobe": r = _run([FFPROBE, "-v", "error", "-print_format", "json", "-show_format", "-show_streams", path]) if r.returncode == 0: try: d = json.loads(r.stdout) streams = d.get("streams", []) v = next((s for s in streams if s.get("codec_type") == "video"), None) if not v: raise ValueError("视频无视频流") fmt = d.get("format", {}) dur = float(fmt.get("duration", 0) or 0) if dur <= 0: dur = float(v.get("duration", 0) or 0) return { "duration_sec": round(dur, 2), "width": int(v.get("width", 0)), "height": int(v.get("height", 0)), "codec": v.get("codec_name", ""), "fps": v.get("avg_frame_rate", ""), "size_bytes": int(fmt.get("size", 0) or 0), } except Exception: pass # 解析失败落到 ffmpeg 兜底 info = _parse_ffmpeg_info(path) info["size_bytes"] = os.path.getsize(path) return info def check_image(path: str) -> dict: """校验人物参考图: 可解码 + 尺寸, 返回 {width,height,ok,error}""" try: info = _parse_ffmpeg_info(path) w, h = info["width"], info["height"] except Exception as e: return {"ok": False, "error": f"图片无法解码: {e}"} if w < 256 or h < 256: return {"ok": False, "error": f"参考图过小({w}x{h}), 至少 256px"} return {"ok": True, "width": w, "height": h, "codec": info["codec"], "error": ""} # ════════════════════════════════════════════════════════════ # 预处理(生成合规中间文件) # ════════════════════════════════════════════════════════════ def prep_source_video(src_path: str, out_path: str, max_duration: int = 0, orientation: str = "") -> dict: """源视频合规化 → h264 + 竖屏裁切 + 限长 返回 {out_path, duration_sec, width, height, trimmed: bool} 超长时截取前 max_duration 秒(并返回 trimmed=True) """ info = probe_video(src_path) max_duration = max_duration or int(_cfg().get("max_duration", DEFAULT_MAX_DURATION)) orientation = orientation or _cfg().get("orientation", DEFAULT_ORIENTATION) dur = info["duration_sec"] duration_arg = [] if dur > max_duration: duration_arg = ["-t", str(max_duration)] logger.info(" 源视频 %.1fs 超限, 截取前 %ds", dur, max_duration) # 裁切到竖屏: 保持画面内容, 按 9:16 裁掉左右 (16:9 横屏 → 切中间竖条) vf = [] if orientation == "9:16": vf.append("scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920") else: vf.append("scale=1920:1080:force_original_aspect_ratio=increase,crop=1920:1080") cmd = [FFMPEG, "-y", "-i", src_path, *duration_arg, "-vf", ",".join(vf), "-c:v", "libx264", "-preset", "veryfast", "-crf", "20", "-c:a", "aac", "-b:a", "128k", "-movflags", "+faststart", out_path] r = _run(cmd, timeout=300) if r.returncode != 0: raise ValueError(f"视频预处理失败: {r.stderr.strip()[-300:]}") out_info = probe_video(out_path) return {**out_info, "out_path": out_path, "trimmed": dur > max_duration} def prep_reference_image(src_path: str, out_path: str, max_side: int = 0) -> dict: """参考图归一化: 转 jpg + 长边上限, 返回 {out_path, width, height}""" chk = check_image(src_path) if not chk["ok"]: raise ValueError(chk["error"]) max_side = max_side or int(_cfg().get("image_max_side", DEFAULT_IMAGE_MAX)) vf = f"scale='min({max_side},iw)':'min({max_side},ih)':force_original_aspect_ratio=decrease" out = Path(out_path) if out.suffix.lower() not in (".jpg", ".jpeg"): out = out.with_suffix(".jpg") cmd = [FFMPEG, "-y", "-i", src_path, "-vf", vf, "-q:v", "2", str(out)] r = _run(cmd, timeout=120) if r.returncode != 0: raise ValueError(f"参考图处理失败: {r.stderr.strip()[-300:]}") return {"out_path": str(out), "width": chk["width"], "height": chk["height"]} def extract_preview_frame(video_path: str, out_path: str) -> str: """抽取视频中帧作为预览图(jpg)""" out = Path(out_path) if out.suffix.lower() not in (".jpg", ".jpeg"): out = out.with_suffix(".jpg") cmd = [FFMPEG, "-y", "-i", video_path, "-frames:v", "1", "-q:v", "3", str(out)] r = _run(cmd, timeout=60) if r.returncode != 0: raise ValueError(f"抽帧失败: {r.stderr.strip()[-200:]}") return str(out) # ── CLI 自测 ── if __name__ == "__main__": import argparse ap = argparse.ArgumentParser(description="person_swap 预处理自测") ap.add_argument("video", help="源视频路径") ap.add_argument("--image", default="", help="参考图路径(可选, 测试图片处理)") args = ap.parse_args() print(f"FFMPEG={FFMPEG} FFPROBE={FFPROBE}") info = probe_video(args.video) print("源视频:", info) out = "/tmp/ps_prep_test.mp4" res = prep_source_video(args.video, out) print("预处理:", res["out_path"], f"{res['width']}x{res['height']}", f"{res['duration_sec']}s", "trimmed" if res["trimmed"] else "") if args.image: print("参考图检查:", check_image(args.image)) r = prep_reference_image(args.image, "/tmp/ps_ref_test.jpg") print("参考图处理:", r["out_path"])