Huo15 Comic Lipsync

给视频镜头+对白音频做口型同步（Kling 2.5 Lip Sync）。没有对白的镜头跳过。触发词：对口型、lipsync、口型同步。

Security Scan

The skill mostly does what it claims (uploading video+audio to Kling for lip-sync) but the package metadata omits the required KLING_API_KEY and the code will upload user video/audio to a third‑party API and create local checkpoint/cost files — this mismatch and privacy/billing impact warrant caution.

Purpose & Capability

The skill's stated purpose (Kling lip-sync) matches the code in scripts/lipsync.py which posts video+audio to Kling. However the registry metadata claims no required environment variables while the code requires KLING_API_KEY; that is an incoherence between claimed requirements and actual capability. The bundled shared library also contains an ArkClient that needs ARK_API_KEY if used, but that client is not invoked by lipsync.py.

Instruction Scope

SKILL.md shows the local CLI usage and a Kling API example, which aligns with behaviour. But the runtime instructions/code will: (1) read videos and audio from provided directories, (2) upload raw video+audio files to a third‑party endpoint (https://api.klingai.com), (3) download the processed video, and (4) write .checkpoint.json and .cost.json into the project directory. The SKILL.md / registry did not clearly surface the KLING_API_KEY requirement or the fact that raw media will be uploaded off‑device (privacy/billing implication).

Install Mechanism

No install spec; this is a filesystem-bundled skill (code included in scripts/). Nothing is fetched from remote URLs during install. No unexpected installers or archive extraction were observed.

Credentials

The skill actually requires a KLING_API_KEY environment variable at runtime (scripts/lipsync.py raises if missing), but the registry metadata lists no required env vars. The shared ark_api.py also expects ARK_API_KEY when instantiated (not used by lipsync), which increases surface area if other code paths are exercised. Requesting API keys is proportionate to the stated function, but omitting them from metadata is a notable inconsistency and a usability/security risk.

Persistence & Privilege

The skill does not request persistent/always-on privileges. It writes outputs and bookkeeping files (.checkpoint.json and .cost.json) under the project/out directories provided by the user — expected for this workflow and limited to the project's directory.

What to consider before installing

Key points before installing or running: - This skill will upload your raw video and audio files to a third‑party service (https://api.klingai.com) using a KLING_API_KEY. That is required at runtime even though the skill metadata didn't declare it — provide a key only if you trust Kling and understand billing/privacy. - The code will create/modify .checkpoint.json and .cost.json in the project directory (out_dir parent). Expect local bookkeeping files and cost tracking. - Billing: the skill charges per‑scene lip‑sync; the code uses pricing in config.py and CostGuard to enforce caps. Test on short/dummy clips first to estimate cost. - Review or sandbox first: if the media contain sensitive content, do not upload until you verify Kling's privacy/retention policy. Consider using throwaway API keys or a service account with minimal privileges and monitor usage. - Metadata mismatch: the registry should declare KLING_API_KEY (and any other required creds). Ask the publisher to correct the metadata or inspect the code yourself before use. If you want, I can: (a) point out the exact lines that require KLING_API_KEY and where uploads occur, (b) suggest a minimal local wrapper to dry‑run without uploading, or (c) produce a checklist of privacy/billing questions to ask the Kling provider.

火15 漫剧-对口型 Skill

视频 + 音频 → 口型同步后的视频。

---

输入 / 输出

python scripts/lipsync.py \
  --video-dir output/demo/videos \
  --audio-dir output/demo/audio \
  --out-dir output/demo/lipsync

每个镜头取该镜第一条对白的音频做口型同步；无对白直接复制原视频。

API

POST https://api.kling.com/v1/videos/lip-sync
Headers: Authorization: Bearer {KLING_API_KEY}
Body:
{
  "video_url": "...",
  "audio_url": "...",
  "mode": "kling-v2.5"
}

注意

• 视频最短 3s，如果对白音频 <3s 自动补静默
• 单镜成本 ¥3，48 镜 = ¥144（可通过 --no-lipsync 关闭省钱）