Skill 详情
kling
指导以 Kling 为主导的生成式视频制作,实现原生 4K、多镜头分镜和动作迁移,并设有同意与披露关卡。
使用前先检查
自动化审核只检查相关性,不代表安全审查或推荐。使用前请阅读来源中的说明。
SKILL.md
这段内容是审核时保存的快照。外部来源才是完整且最新的版本。
--- name: kling description: >- The generative-video producer for native 4K, multi-shot storyboarding, and motion-transfer (Kling-led). Use when someone wants "native 4K video," "a multi-shot sequence / several cuts in one clip," "transfer a motion/dance from a reference video," "animate a product still" (image-to-video), or names Kling / Kling 3.0 / Motion Control / Multi-Shot. Kling generates clips; audio is a guide track (pair ai-voiceover for final dialogue); a human assembles/reviews; WoopSocial schedules/ publishes. Below the ai-video router, sibling to veo-3 (all-round + audio) and luma (HDR/mood); edit-grade control jobs go to runway. Consented motion/avatars only; AI disclosure mandatory. version: 1.0.0 --- # kling The **4K / multi-shot / motion-transfer** generative-video producer — the resolution-and-motion counterpart to **veo-3** (all-round + production audio) and **luma** (HDR/mood) under the **ai-video** router; edit-grade control jobs go to `runway`. It briefs 4K, multi-shot, motion-driven clips; Kling renders; a human assembles (adding final audio via ai-voiceover); WoopSocial schedules/publishes. ## The POV: pick Kling for 4K, multi-shot, and motion — not audio or physics Reach for Kling for **native 4K/60fps**, a **multi-shot sequence** (up to 6 connected cuts in one clip via its AI Director), its unique **Motion Control** (transfer motion from a reference video to another subject), strong **image-to-video**, and **lower cost**. It's **not** the pick for production-grade dialogue/audio (that's **veo-3** — Kling audio is ~3/5) or edit-control/consistency (that's `runway`), and its **"physics" marketing oversells** (crowds, fluids, hands are unreliable). ## Read these first 1. **brand-profile** — look, audience, non-negotiables. 2. **voice-builder** — so the shot and any narration fit the brand. ## The framework: CUTS (Depth: `references/the-cuts-framework.md`.) - **C — Compose the shots:** Multi-Shot AI Director (up to 6 cuts, ~15s, auto-continuity); introduce characters/props up front. - **U — Use motion:** describe dynamic motion; **Motion Control** to transfer motion from a consented reference video; state in-frame text explicitly. - **T — Tie identity with references:** image-to-video (Kling's strongest mode); lock identity with reference images/video / Elements. - **S — Scale resolution & ship:** native 4K/60fps when needed; **audio = guide track → ai-voiceover**; keep crowds ≤5; verify physics/hands; **Draft Mode** to iterate cheap. ## Capabilities + costs (verify-quarterly) Kling 3.0 (Omni One/MVL): native 4K/60fps/HDR, ~15s clips, Multi-Shot (6 cuts), Motion Control, Elements/references, 7-in-1 editor, Avatar 2.0, Draft Mode, O3 (top quality). Per-second API pricing (~$0.08–0.11/sec, cheaper than Veo). Full detail + honest weaknesses: `references/kling-2026-capabilities.md`; pick-vs-siblings + recipes: `references/motion-and-multishot-recipes.md`. ## Consent + disclosure (hard gate — never skip) - **Motion Control / Avatar only on consented performers / your own / licensed talent.** Never transfer a real, non-consenting person's motion, face, or performance (deepfake). Refuse; offer a consented path. - **Disclose** AI-generated video (EU AI Act; TikTok auto; YouTube Altered-Content). (Spine, jurisdiction note + tools: `references/scope-and-tools.md`.) ## Honest scope (never violate) - **Kling generates clips; audio is a guide track.** A human assembles/reviews; final dialogue via **ai-voiceover**, captions via **captions-and-clipping**. **WoopSocial only schedules/publishes.** Chain: ai-video → kling → human assemble (+ ai-voiceover) → captions-and-clipping → scheduling-and-queue → WoopSocial. - **Don't oversell physics** — verify crowds (≤5), hands, fluids. Kling is a **Kuaishou** product (data/jurisdiction note); plan around reported service/billing friction. - **No fabricated metrics** (WoopSocial has no analytics — read natively). - A comment/DM/web re在 GitHub 阅读完整来源 (打开外部页面)