🔄 卡若AI 同步 2026-02-22 14:24 | 更新:金仓、卡木、运营中枢工作台 | 排除 >20MB: 8 个

This commit is contained in:
2026-02-22 14:24:13 +08:00
parent 31794af87f
commit d96bd60d5b
7 changed files with 346 additions and 43 deletions

View File

@@ -0,0 +1,220 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
腾讯云 TAT + 宝塔 APIkr宝塔 Node 全量启动修复
1. 修复 site.db 中所有 Node 项目的 project_scriptMODULE_NOT_FOUND 根因node /path 改为 cd /path && npm run start
2. 停止全部 Node、清端口、批量启动
3. 验证直到全部运行或超时
凭证00_账号与API索引.md 腾讯云 SecretId/SecretKey宝塔 API 密钥 qcWubCdlfFjS2b2DMT1lzPFaDfmv1cBT
"""
import base64
import json
import os
import re
import sys
import time
KR_INSTANCE_ID = "ins-aw0tnqjo"
REGION = "ap-guangzhou"
BT_API_KEY = "qcWubCdlfFjS2b2DMT1lzPFaDfmv1cBT"
SHELL_SCRIPT = r'''#!/bin/bash
echo "=== kr宝塔 Node 全量启动修复 ==="
python3 - << 'PYMAIN'
import hashlib, json, os, re, sqlite3, subprocess, time, urllib.request, urllib.parse, ssl
ssl._create_default_https_context = ssl._create_unverified_context
PANEL, K = "https://127.0.0.1:9988", "qcWubCdlfFjS2b2DMT1lzPFaDfmv1cBT"
def sign():
t = int(time.time())
s = str(t) + hashlib.md5(K.encode()).hexdigest()
return {"request_time": t, "request_token": hashlib.md5(s.encode()).hexdigest()}
def post(p, d=None):
pl = sign()
if d: pl.update(d)
r = urllib.request.Request(PANEL + p, data=urllib.parse.urlencode(pl).encode())
with urllib.request.urlopen(r, timeout=25) as resp:
return json.loads(resp.read().decode())
def pids(port):
try:
o = subprocess.check_output("ss -tlnp 2>/dev/null | grep ':%s ' || true" % port, shell=True, universal_newlines=True)
return {int(x) for x in re.findall(r"pid=(\d+)", o)}
except: return set()
def ports(it):
cfg = it.get("project_config") or {}
if isinstance(cfg, str):
try: cfg = json.loads(cfg)
except: cfg = {}
ps = []
if cfg.get("port"): ps.append(int(cfg["port"]))
for m in re.findall(r"-p\s*(\d+)", str(cfg.get("project_script", ""))): ps.append(int(m))
return sorted(set(ps))
# 【1】修复 site.db 启动命令
print("\n【1】修复 site.db 启动命令")
db = "/www/server/panel/data/db/site.db"
fixed = 0
if os.path.isfile(db):
conn = sqlite3.connect(db)
c = conn.cursor()
c.execute("SELECT id, name, path, project_config FROM sites WHERE project_type='Node'")
for row in c.fetchall():
sid, name, path, cfg_str = row[0], row[1], row[2], row[3] or "{}"
path = (path or "").strip()
try: cfg = json.loads(cfg_str) if cfg_str else {}
except: cfg = {}
# 从 project_config 取 path 作为项目根目录
proj_path = cfg.get("path") or cfg.get("project_path") or path
if not proj_path or not os.path.isdir(proj_path):
print(" 跳过 %s (路径不存在: %s)" % (name, proj_path))
continue
old_script = str(cfg.get("project_script") or cfg.get("run_cmd") or "").strip()
cmd = "cd %s && (pnpm start 2>/dev/null || npm run start)" % proj_path
if "cd " not in old_script or proj_path not in old_script:
cfg["project_script"] = cmd
cfg["run_cmd"] = cmd
cfg["path"] = proj_path
c.execute("UPDATE sites SET path=?, project_config=? WHERE id=?", (proj_path, json.dumps(cfg, ensure_ascii=False), sid))
fixed += 1
print(" 已修复: %s -> %s" % (name, proj_path))
conn.commit()
conn.close()
print(" 共修复 %d 个项目" % fixed)
# 【2】停止全部 Node
print("\n【2】停止 Node 项目")
r0 = post("/project/nodejs/get_project_list")
items = r0.get("data") or r0.get("list") or []
for it in items:
name = it.get("name")
if name:
try:
for port in ports(it):
for pid in pids(port): subprocess.call("kill -9 %s 2>/dev/null" % pid, shell=True)
pf = "/www/server/nodejs/vhost/pids/%s.pid" % name
if os.path.exists(pf): open(pf, "w").write("0")
post("/project/nodejs/stop_project", {"project_name": name})
print(" 停: %s" % name)
except Exception as e: print("%s: %s" % (name, str(e)[:40]))
time.sleep(0.5)
time.sleep(3)
# 【3】批量启动最多3轮
print("\n【3】批量启动 Node 项目")
for round_num in range(3):
r1 = post("/project/nodejs/get_project_list")
items = r1.get("data") or r1.get("list") or []
to_start = [it for it in items if it.get("name") and not it.get("run")]
if not to_start:
print(" 全部已运行")
break
print("%d轮: 待启动 %d" % (round_num + 1, len(to_start)))
for it in to_start:
name = it.get("name")
if not name: continue
try:
for port in ports(it):
for pid in pids(port): subprocess.call("kill -9 %s 2>/dev/null" % pid, shell=True)
pf = "/www/server/nodejs/vhost/pids/%s.pid" % name
if os.path.exists(pf): open(pf, "w").write("0")
post("/project/nodejs/stop_project", {"project_name": name})
time.sleep(0.5)
r = post("/project/nodejs/start_project", {"project_name": name})
ok = r.get("status") is True or "成功" in str(r.get("msg", ""))
print(" %s: %s" % (name, "OK" if ok else "FAIL"))
except Exception as e: print(" %s: ERR %s" % (name, str(e)[:30]))
time.sleep(1.5)
time.sleep(8)
# 【4】最终状态
print("\n【4】最终状态")
r2 = post("/project/nodejs/get_project_list")
items2 = r2.get("data") or r2.get("list") or []
run_count = sum(1 for x in items2 if x.get("run"))
total = len(items2)
print(" 运行 %d / %d" % (run_count, total))
for it in items2:
print(" %s: %s" % (it.get("name"), "运行中" if it.get("run") else "未启动"))
PYMAIN
echo ""
echo "=== 完成 ==="
'''
def _read_creds():
d = os.path.dirname(os.path.abspath(__file__))
for _ in range(6):
root = d
if os.path.isfile(os.path.join(root, "运营中枢", "工作台", "00_账号与API索引.md")):
with open(os.path.join(root, "运营中枢", "工作台", "00_账号与API索引.md"), "r", encoding="utf-8") as f:
text = f.read()
sid = skey = None
in_tx = False
for line in text.splitlines():
if "### 腾讯云" in line or "腾讯云" in line and "Secret" in line: in_tx = True
if in_tx and "### " in line and "腾讯云" not in line: break
m = re.search(r"SecretId[^|]*\|\s*`([^`]+)`", line, re.I)
if m and "AKID" in str(m.group(1)): sid = m.group(1).strip()
m = re.search(r"SecretKey\s*\|\s*`([^`]+)`", line, re.I)
if m: skey = m.group(1).strip()
return sid or os.environ.get("TENCENTCLOUD_SECRET_ID"), skey or os.environ.get("TENCENTCLOUD_SECRET_KEY")
d = os.path.dirname(d)
return None, None
def main():
sid, skey = _read_creds()
if not sid or not skey:
print("❌ 未配置腾讯云 SecretId/SecretKey"); return 1
try:
from tencentcloud.common import credential
from tencentcloud.tat.v20201028 import tat_client, models
except ImportError:
print("请安装: pip install tencentcloud-sdk-python-tat"); return 1
cred = credential.Credential(sid, skey)
client = tat_client.TatClient(cred, REGION)
req = models.RunCommandRequest()
req.Content = base64.b64encode(SHELL_SCRIPT.encode("utf-8")).decode()
req.InstanceIds = [KR_INSTANCE_ID]
req.CommandType = "SHELL"
req.Timeout = 300
req.CommandName = "kr宝塔_Node全量启动修复"
resp = client.RunCommand(req)
inv_id = resp.InvocationId
print("✅ TAT 已下发 InvocationId:", inv_id)
print(" 步骤: 修复 site.db 启动命令 → 停止 Node → 批量启动(3轮) → 验证")
print(" 等待 150s 后查询结果...")
time.sleep(150)
try:
req2 = models.DescribeInvocationTasksRequest()
f = models.Filter()
f.Name, f.Values = "invocation-id", [inv_id]
req2.Filters = [f]
r2 = client.DescribeInvocationTasks(req2)
for t in (r2.InvocationTaskSet or []):
st = getattr(t, "TaskStatus", "")
print("\n状态:", st)
tr = getattr(t, "TaskResult", None)
if tr:
j = json.loads(tr) if isinstance(tr, str) else {}
out = j.get("Output", "")
if out:
try:
out = base64.b64decode(out).decode("utf-8", errors="replace")
except:
pass
print(out[:7000])
except Exception as e:
print("查询失败:", e)
return 0
if __name__ == "__main__":
sys.exit(main())

View File

@@ -14,9 +14,9 @@ from pathlib import Path
OLLAMA_URL = "http://localhost:11434"
DEFAULT_CTA = "关注我,每天学一招私域干货"
CLIP_COUNT = 8
MIN_DURATION = 60 # 1 分钟
MAX_DURATION = 180 # 3 分钟
CLIP_COUNT = 15
MIN_DURATION = 60 # 最少 1 分钟
MAX_DURATION = 300 # 最多 5 分钟
def parse_srt_segments(srt_path: str) -> list:
@@ -42,22 +42,23 @@ def parse_srt_segments(srt_path: str) -> list:
def fallback_highlights(transcript_path: str, clip_count: int) -> list:
"""规则备用:按时长均匀切分,每段 60-180 秒"""
"""规则备用:每段 60-300 秒1-5 分钟)"""
segments = parse_srt_segments(transcript_path)
if not segments:
return []
total = segments[-1]["end_sec"] if segments else 0
interval = max(120, total / clip_count) # 每段约 2 分钟
seg_dur = min(300, max(60, total / clip_count)) # 每段 1-5 分钟
result = []
for i in range(clip_count):
start_sec = int(interval * i + 30)
end_sec = min(int(start_sec + 120), int(total - 5)) # 约 2 分钟
if end_sec <= start_sec + 30:
start_sec = int(i * seg_dur)
end_sec = min(int(start_sec + seg_dur), int(total - 5))
if end_sec <= start_sec + 59: # 不足 1 分钟跳过
continue
# 找该时间段内的字幕
texts = [s["text"] for s in segments if s["end_sec"] >= start_sec and s["start_sec"] <= end_sec]
excerpt = (texts[0][:50] + "..." if texts and len(texts[0]) > 50 else (texts[0] if texts else ""))
hook = (excerpt[:15] + "..." if len(excerpt) > 15 else excerpt) or f"精彩片段{i+1}"
joined = " ".join(texts)[:80] if texts else ""
excerpt = (joined + "..." if len(joined) > 50 else joined) or f"精彩片段{i+1}"
hook = (excerpt[:18] + "..." if len(excerpt) > 18 else excerpt) or f"精彩片段{i+1}"
h, m, s = start_sec // 3600, (start_sec % 3600) // 60, start_sec % 60
eh, em, es = end_sec // 3600, (end_sec % 3600) // 60, end_sec % 60
result.append({
@@ -85,27 +86,66 @@ def srt_to_timestamped_text(srt_path: str) -> str:
return "\n".join(lines)
def _sec_to_hhmmss(sec: float) -> str:
"""秒数转为 HH:MM:SS"""
s = int(sec)
h, m = s // 3600, (s % 3600) // 60
ss = s % 60
return f"{h:02d}:{m:02d}:{ss:02d}"
def _parse_time_to_sec(t: str) -> float:
"""解析 HH:MM:SS、HH:MM:SS.mmm、HH:MM:SS,mmm 为秒"""
t = str(t).strip().replace(",", ".")
parts = re.split(r"[:.]", t)
if len(parts) >= 3:
try:
return int(parts[0]) * 3600 + int(parts[1]) * 60 + float(parts[2])
except (ValueError, TypeError):
pass
# 纯数字视为秒
m = re.match(r"^(\d+(?:\.\d+)?)\s*$", t)
if m:
return float(m.group(1))
return 0
def _filter_short_clips(data: list) -> list:
"""过滤掉时长 < 60 秒的切片"""
result = []
for item in data:
if not isinstance(item, dict):
continue
st = item.get("start_time") or item.get("start") or "00:00:00"
et = item.get("end_time") or item.get("end") or "00:01:00"
dur = _parse_time_to_sec(et) - _parse_time_to_sec(st)
if dur >= 60:
result.append(item)
else:
print(f" 过滤短片段: {item.get('title','?')} (仅{dur:.0f}秒)", file=sys.stderr)
return result
def _build_prompt(transcript: str, clip_count: int) -> str:
"""构建高光识别 prompt完整观点+干货,1-3分钟,全中文)"""
txt = transcript[:18000] if len(transcript) > 18000 else transcript
return f"""你是资深短视频策划师。请从视频文字稿中识别 {clip_count}**完整的核心观点/干货片段**
"""构建高光识别 prompt1-5分钟,主题与内容必须一致,全中文)"""
txt = transcript[:25000] if len(transcript) > 25000 else transcript
return f"""你是资深短视频策划师。请从视频文字稿中识别尽可能多的**完整干货片段**,目标 {clip_count}以上
【时长要求】每个片段 **60-300 秒1-5 分钟)**,少于 60 秒的不要输出。
【主题一致】title、hook_3sec、transcript_excerpt 必须**完全对应**该时间段内的实际内容,禁止泛泛而谈。
- title从该段时间内的核心观点提炼15字内
- hook_3sec从该段第一句或核心金句提炼15字内
- transcript_excerpt该段内容的中文摘要50字内
【切片原则】
- 每个片段必须是**完整的一个话题/观点**,有头有尾,逻辑闭环,不能截断
- 时长 **60-180 秒1-3 分钟)**,尽量接近 2 分钟,确保内容完整
- 优先选:金句、完整故事、可操作方法论、反常识观点、情绪高点、成体系讲解
- 每段必须是完整话题,有头有尾
- 优先:金句、故事、方法论、反常识、情绪高点
- 相邻片段间隔至少 60 秒
- 从文字稿时间戳精确提取 start_time、end_time
【输出字段】所有内容**必须使用简体中文**,若原文是英文请翻译后填写
- title: 核心观点标题15字内用于文件名
- start_time: "HH:MM:SS"(从文字稿时间戳精确提取)
- end_time: "HH:MM:SS"
- hook_3sec: 封面 Hook 文案15字内吸引点击
- cta_ending: "{DEFAULT_CTA}"
- transcript_excerpt: 本片段核心内容摘要50字内中文
- reason: 推荐理由(中文)
【输出字段】全部**简体中文**,英文原文需翻译
- title、start_time、end_time、hook_3sec、cta_ending"{DEFAULT_CTA}"、transcript_excerpt、reason
【强制】title、hook_3sec、transcript_excerpt、reason 必须全部简体中文,禁止英文。
只输出 JSON 数组,不要 ``` 或其他文字。
视频文字稿:
@@ -233,8 +273,24 @@ def main():
data = fallback_highlights(str(transcript_path), args.clips)
if not isinstance(data, list):
data = [data]
# 强制中文:若 title/hook 含英文,翻译为中文
print(" 确保导出名与封面为中文...")
# 过滤短于 1 分钟的切片
data = _filter_short_clips(data)
# 统一 start_time/end_time 为 HH:MM:SS兼容 Ollama 返回秒数)
for item in data:
if not isinstance(item, dict):
continue
st = item.get("start_time") or item.get("start")
if isinstance(st, (int, float)):
item["start_time"] = _sec_to_hhmmss(st)
et = item.get("end_time") or item.get("end")
if isinstance(et, (int, float)):
item["end_time"] = _sec_to_hhmmss(et)
# 若 AI 返回的片段全被过滤,用规则备用
if not data and transcript_path.exists():
print(" AI 片段时长无效改用规则切分1-5 分钟)", file=sys.stderr)
data = fallback_highlights(str(transcript_path), args.clips)
# 强制中文
print(" 确保导出名与封面为简体中文...")
data = _ensure_chinese_highlights(data)
out_path = Path(args.output)
out_path.parent.mkdir(parents=True, exist_ok=True)

View File

@@ -594,8 +594,14 @@ def enhance_clip(clip_path, output_path, highlight_info, temp_dir, transcript_pa
print(f" ⊘ 跳过字幕烧录(检测到原片已有字幕/图片)")
else:
start_time = highlight_info.get('start_time', '00:00:00')
parts = start_time.split(':')
start_sec = int(parts[0]) * 3600 + int(parts[1]) * 60 + float(parts[2])
try:
if isinstance(start_time, (int, float)):
start_sec = float(start_time)
else:
parts = str(start_time).strip().split(':')
start_sec = int(parts[0]) * 3600 + int(parts[1]) * 60 + float(parts[2])
except (IndexError, ValueError):
start_sec = 0
end_sec = start_sec + duration
subtitles = parse_srt_for_clip(transcript_path, start_sec, end_sec)
for sub in subtitles:
@@ -606,7 +612,8 @@ def enhance_clip(clip_path, output_path, highlight_info, temp_dir, transcript_pa
img_path = os.path.join(temp_dir, f'sub_{i:04d}.png')
create_subtitle_image(sub['text'], width, height, img_path)
sub_images.append({'path': img_path, 'start': sub['start'], 'end': sub['end']})
print(f" ✓ 字幕图片 ({len(sub_images)}张)")
if sub_images:
print(f" ✓ 字幕图片 ({len(sub_images)}张)")
# 4. 检测静音
silences = detect_silence(clip_path, SILENCE_THRESHOLD, SILENCE_MIN_DURATION)

View File

@@ -93,6 +93,8 @@ def main():
parser.add_argument("--skip-highlights", action="store_true", help="跳过高光识别(已有 highlights.json")
parser.add_argument("--skip-clips", action="store_true", help="跳过切片(已有 clips/,仅重新增强)")
parser.add_argument("--language", "-l", default="zh", choices=["zh", "en"], help="转录语言(纳瓦尔访谈等英文内容用 en")
parser.add_argument("--skip-subs", action="store_true", help="跳过字幕烧录(原片已有字幕时用)")
parser.add_argument("--force-burn-subs", action="store_true", help="强制烧录字幕(忽略检测)")
args = parser.parse_args()
video_path = Path(args.video).resolve()
@@ -205,19 +207,19 @@ def main():
# 4. 增强(封面+字幕+加速soul_enhancePillow无需 drawtext
enhanced_dir.mkdir(parents=True, exist_ok=True)
ok = run(
[
sys.executable,
str(SCRIPT_DIR / "soul_enhance.py"),
"--clips", str(clips_dir),
"--highlights", str(highlights_path),
"--transcript", str(transcript_path),
"--output", str(enhanced_dir),
],
"增强处理(封面+字幕+加速)",
timeout=900,
check=False,
)
enhance_cmd = [
sys.executable,
str(SCRIPT_DIR / "soul_enhance.py"),
"--clips", str(clips_dir),
"--highlights", str(highlights_path),
"--transcript", str(transcript_path),
"--output", str(enhanced_dir),
]
if getattr(args, "skip_subs", False):
enhance_cmd.append("--skip-subs")
if getattr(args, "force_burn_subs", False):
enhance_cmd.append("--force-burn-subs")
ok = run(enhance_cmd, "增强处理(封面+字幕+加速)", timeout=900, check=False)
import shutil
enhanced_count = len(list(enhanced_dir.glob("*.mp4")))
if enhanced_count == 0 and clips_list:

View File

@@ -62,6 +62,22 @@ bash "/Users/karuo/Movies/soul视频/soul 派对 106场 20260221_output/热点
---
## [卡若复盘] Soul 106 单目录输出+竖屏居中2026-02-22
**🎯 目标·结果**
目标:成片统一输出到单目录、竖屏完全居中、挤掉右侧空白、有封面+字幕、同名覆盖。结果:已调整脚本与增强逻辑。
**📌 修改内容**
1. **单目录输出**:临时目录 `.tmp_slice` 只做中间处理,最终成片全部落在 `output` 根目录,不再分子目录。
2. **同名覆盖**soul_enhance 不再预先清空 mp4直接写入覆盖ffmpeg 使用 `-y` 覆盖。
3. **竖屏裁剪**`crop=1080:1920:0:(ih-1920)/2`,取左侧 1080 列并垂直居中,挤掉右侧空白,内容偏左显示(符合「往右挤掉右侧空白」)。
4. **封面与字幕**:沿用 soul_enhance 的 Pillow 封面 + 烧录字幕逻辑。
**📁 输出路径**
`/Users/karuo/Movies/soul视频/soul 派对 106场 20260221_output/*.mp4`(如 soul106_01_xxx_enhanced.mp4
---
## [卡若复盘] 文档与字幕简体中文优化2026-02
**🎯 目标·结果·达成率**

View File

@@ -89,3 +89,4 @@
| 2026-02-22 12:42:56 | 🔄 卡若AI 同步 2026-02-22 12:42 | 更新:金仓、卡木、运营中枢工作台 | 排除 >20MB: 8 个 |
| 2026-02-22 13:08:21 | 🔄 卡若AI 同步 2026-02-22 13:08 | 更新:卡木、总索引与入口、运营中枢工作台 | 排除 >20MB: 8 个 |
| 2026-02-22 13:45:50 | 🔄 卡若AI 同步 2026-02-22 13:45 | 更新:金仓、卡木、运营中枢工作台 | 排除 >20MB: 8 个 |
| 2026-02-22 13:57:34 | 🔄 卡若AI 同步 2026-02-22 13:57 | 更新:金仓、水溪整理归档、卡木、运营中枢工作台 | 排除 >20MB: 8 个 |

View File

@@ -92,3 +92,4 @@
| 2026-02-22 12:42:56 | 成功 | 成功 | 🔄 卡若AI 同步 2026-02-22 12:42 | 更新:金仓、卡木、运营中枢工作台 | 排除 >20MB: 8 个 | [仓库](http://open.quwanzhi.com:3000/fnvtk/karuo-ai) [百科](http://open.quwanzhi.com:3000/fnvtk/karuo-ai/wiki) |
| 2026-02-22 13:08:21 | 成功 | 成功 | 🔄 卡若AI 同步 2026-02-22 13:08 | 更新:卡木、总索引与入口、运营中枢工作台 | 排除 >20MB: 8 个 | [仓库](http://open.quwanzhi.com:3000/fnvtk/karuo-ai) [百科](http://open.quwanzhi.com:3000/fnvtk/karuo-ai/wiki) |
| 2026-02-22 13:45:50 | 成功 | 成功 | 🔄 卡若AI 同步 2026-02-22 13:45 | 更新:金仓、卡木、运营中枢工作台 | 排除 >20MB: 8 个 | [仓库](http://open.quwanzhi.com:3000/fnvtk/karuo-ai) [百科](http://open.quwanzhi.com:3000/fnvtk/karuo-ai/wiki) |
| 2026-02-22 13:57:34 | 成功 | 成功 | 🔄 卡若AI 同步 2026-02-22 13:57 | 更新:金仓、水溪整理归档、卡木、运营中枢工作台 | 排除 >20MB: 8 个 | [仓库](http://open.quwanzhi.com:3000/fnvtk/karuo-ai) [百科](http://open.quwanzhi.com:3000/fnvtk/karuo-ai/wiki) |