Video & Animation

AI-assisted video editing workflows for cutting, structuring, and augmenting real footage. Covers the full pipeline from raw capture through FFmpeg, Remotion, ElevenLabs, fal.ai, and final polish in Descript or CapCut. Use when the user wants to edit video, cut footage, create vlogs, or build video content.

Script-to-finished-Reel pipeline with two human gates. Writes the script with effect cues, then after the user shoots: picks the best takes, cuts dead air and filler, adds effects, captions and on-screen text, renders a preview, takes notes, renders the final and writes the post caption. Use when the user types /reel, has a video idea, drops raw footage, says "edit video ini", "bikin reel", gives notes on a preview, or shares a reference edit to copy the style of.

AI 短片/动画全流程总导演工作流（Claude Code 当导演 + ChatGPT 图像 API（gpt-image-2）画首尾帧 + Seedance 生成视频 + ffmpeg/Python 后期），从一个想法到成片，目标是花最少的钱、最低的抽卡率。只要用户想用 AI 做视频、短片、动画、MV、广告片、剧情片段，或提到 Seedance、即梦/豆包视频、火山方舟、第三方/中转 Seedance key、.env.local、ChatGPT/gpt-image 生图、Codex 生图、首尾帧、分镜/storyboard、镜头语言、视频提示词、抽卡、AI 视频后期（拼接、转场、调色、放大、卡点），就使用这个 skill——即使用户只说"我有个想法想拍成片子"而没提任何工具。Also use for AI video production, text/image-to-video pipelines, keyframe-based video generation, shot lists and video prompt review.

Create concise project videos from selected project sources, real captures or work-story packets, including an editable composition, MP4, poster and captions when speech is used. Use for project demos, build stories and short explainers; use the existing publishing workflow for posting or scheduling.

一镜到底白模预演工作流。用户给一张参考图或一句描述，技能直接产出 3D 白模运镜预演视频；与用户逐版用大白话修改确认后，再按白模实测的相机运动写详细视频提示词；最终交付「白模视频（运动参考）+ 提示词」，一起交给 Seedance 2.5 等视频模型出成片。适用：一镜到底、伪一镜到底、环绕、前推、升空跟随、悬浮失重、FPV 穿梭等长运镜，要求丝滑不折返、不穿模、主体不出画。内置环境探测（有 Blender 走高精度白模，无则引导便携版或降级 Three.js 网页白模）、四项自动校验（单调 / 出画 / 穿模 / 穿顶）、大白话修改指令翻译，用户几乎零门槛。

Make "code-rendered" animated videos where a program draws every frame (Canvas 2D + procedural Web Audio), rendered headless to MP4 in 16:9 / 9:16 / 4:5 from one timeline. Use for animated explainers, technical walkthroughs, diagrams in motion, and procedural sound or music, where the output should be editable code and not generated pixels.

给 AI 短剧出分镜：三层结构——段（一次视频生成，≤15 秒）→ 分镜（段内 2–5 秒的剪切，认领剧本节拍） → 分镜图（每切一张关键帧：主分镜图钉 0.00 秒，子分镜图钉各自切点）。 每段自带一条 MiniMax H3 视频提示词（官方口径默认英文、逐镜换行，promptLang 可切中文）：对齐指令和 [Shot k] 切点时刻由分镜结构推导、逐字对账，台词逐字进 <d> 块（写法规范已内化为 references/h3-prompt.md，不依赖外部 skill）。 产出 storyboard.json + Markdown + 单页评审报告（分镜节奏带 / 分集分镜表 / 生成批次单 / 配音对齐单，含导出 JSON）。不出图——交付的是每格的画面提示词和它该挂哪些参考图。 18 道质量门全部由脚本确定性检查（第 18 道 shot-recipe 可选：挂上 shot-recipes 卡库才查，不挂就明说跳过）； export 一键导出 H3 投产包（每段提示词 + 按 Picture 序的分镜图清单）。零依赖、零 API key，用当前会话额度。 Use when asked to 分镜、出分镜、镜头表、切镜、storyboard for AI short drama。

把文本、文档、网页或结构化资料做成带字幕、配图与动效的解说视频（口播/讲解/科普短视频，竖屏 3:4 或横屏 16:9）：结构化写稿 → TTS 配音 → 配图 → Hyperframes 渲染 MP4。用户想把文章、报告、网页、笔记、公众号/博客长文、论文、会议纪要、周报做成视频、口播视频、讲解视频、科普短视频时使用；不适用于剪辑已有视频、纯配音、导出 PPT。

Record a captioned, optionally voice-narrated screen recording of a running web app with Playwright — smooth cursor, keyboard and click sounds, zoom and pan, 2×–4K sharp — and deliver it where it is needed — a plain .mp4, inline in a GitHub PR or issue, on a Linear issue, in Slack, or a customer-facing tutorial. Use when asked for a demo video, screen recording, walkthrough of a UI or change, narrated tour, "UI/UX walkthrough", PR or ticket video, release or bug-repro clip, or a customer onboarding / how-to video (subtitle captions, Gmail email scene).

Make a silent, text-led product film where a photoreal Blender set carries a visual metaphor and a 2D block-grid ending spells out the headline and call to action. Use when someone wants a launch video, product explainer or short brand film rendered from code with Blender (bpy), Python and ffmpeg, especially on a machine with no GPU.

Produce a narrated walkthrough video of a web app - a screen recording with AI voiceover, subtitles, a visible cursor and highlights on the control being discussed - by running the app, scripting the story, recording it with Playwright and assembling an MP4 with ffmpeg. Use this whenever the user wants a demo video, feature walkthrough, product tour, how-to or explainer video, release or onboarding clip, or a "video to show my boss / CTO / client / team" of something in their app, and also when they ask to redo, shorten, slow down or re-voice a video made this way. Use it even if they only say "record the flow" or "make a video of this feature" without mentioning voiceover or subtitles.

Make a narrated developer explainer video (Remotion + ElevenLabs) about a line, block, file, or architecture of a codebase, for onboarding yourself or a teammate.

Plan, edit and review supplied footage as a coherent film using local tools. Use for source inventories, interview or corporate edits, narration-led shotboards, multilingual captions, rough cuts, production panels, edit revisions and delivery checks. Keeps source evidence, integer-frame timing and render provenance connected. Uses existing FFmpeg and optional SV Edits or SV Motion workflows. Does not generate missing factual footage or act as a general website builder.

Produce a finished, professional motion-graphics video (MP4 with an original synthesized music track, motion cut to the beat) entirely on this Mac — no ffmpeg, no stock footage, no downloads. Use this whenever the user wants a promo, launch, "what's new", release-notes, feature, App Store preview-style, social (Reels/TikTok/Shorts), showreel or explainer video for an app, product or update — including phrasings like "làm video", "make a video", "motion graphics", "animated promo", "video giới thiệu tính năng", "clip cho bản cập nhật", or "turn these release notes into a video". Also use it to revise a video made with this skill earlier (slower pacing, different music mood, new aspect ratio…

Mon Monteur Vidéo — monteur vidéo de bout en bout pour vidéos promo, annonces d'événement, présentations de site ou d'application, montages de témoignages et déclinaisons réseaux sociaux (16:9, 9:16, WhatsApp). Choisit le bon moteur (HyperFrames, Remotion, Premiere Pro, Pillow+ffmpeg), cale image et son sur une seule source de temps, fabrique musique, bruitages, voix et sous-titres, puis contrôle le rendu image par image avant de livrer. À utiliser pour « monte une vidéo », « fais une pub », « teaser », « version TikTok », « coupe les silences », « ajoute des sous-titres », « cale sur la musique ».

Make a vertical music Short (about 30 s) where one stick figure stands or walks alone in a crowded hand-painted 3D street - Blender blockout, image-model repaint projected back on the set, a code-animated crowd of 100-200 stick people (planted-foot walk, umbrellas, phones, lamp light, cast shadows, rain), a time-lapse ending and a finishing pass cut to the song. Use on top of the painted-3d-short skill when people must walk through the painted world.

AR One Click Reel (سينما الورق). يحوّل أي موضوع (تاريخ، تاريخ إسلامي، رياضة، سياسة، قصة قصيرة، علوم، اقتصاد، تقنية، فكرة غير تقليدية) إلى فيديو عربي سينمائي بتقنية الورق المقصوص والستوب موشن، على مراحل معتمدة واحدة تلو الأخرى، من البحث والسكربت إلى الصور والصوت والتحريك والمونتاج. Use when the user asks for a paper-cut / paper-craft / stop-motion video, "فيديو ورق مقصوص", "ستوب موشن ورقي", "AR One Click Reel", "سينما الورق", "paper cinema", or wants an Arabic story or explainer told in a handcrafted paper world.

Make a coded 3D "infinite zoom" explainer video (MP4) where one continuous camera move travels across many orders of magnitude, with a live measurement on screen and familiar objects shown beside the subject for scale. Everything is drawn by code (Three.js, headless Chrome on the GPU, ffmpeg), with no video-generation models. Use this whenever the user wants a 3D explanation or scale video, a "zoom into X down to atoms", a "powers of ten" or "halved N times" style clip, a size-comparison animation, or shares a reference video of a continuous zoom with a changing number and asks for something like it; also for "make a 3D explainer about how small / big / thin / sharp / deep X really is", and…

Make simple 2D cartoon characters and short talking sketch videos, drawn entirely in code on this machine. Use when asked to design or build a cartoon character, draw a character sheet, make a character talk or say a line, animate a dialogue, or make a short comedy sketch, skit, explainer or animatic with one or more characters. Produces PNG character sheets and MP4 videos with lip-synced speech. Not for photoreal, 3D or AI-generated video.

把一段很长的网球练球视频（对墙练、发球机、对拉都行）按击球声自动剪成一个个完整回合，去掉捡球、走动、停顿。当用户说「帮我把练球视频剪成一段段」「把打球的部分剪出来」「去掉捡球」「按回合切网球视频」，或给出一个网球视频目录要求剪辑时使用。也适用于乒乓球、羽毛球、壁球等击球声清楚的练习视频。

视频声音全流程技能：从多个免费音源检索/下载音效与 BGM 素材，离线合成可循环背景音乐，清理人声并做侧链闪避混音与响度归一。用于音效选材、背景音乐、配音处理、混音、响度规范等需求。

Make animated explainer videos and vertical Shorts entirely from code, with low-poly 3D scenes (three.js) and motion graphics, reviewed in a live browser editor before any render. Use this whenever the user wants a "3D short", "3D video", "animated video with code", "faceless explainer", "video made by Claude Code", a Short/Reel from a script or voiceover, or says things like "make a video about X" without footage to edit. Covers the full workflow - script, shot plan, voice timing, style rules, live preview, review notes, chunked render, YouTube packaging. For editing an existing recording (talking head or screen share) use youtube-video-editor instead.

Turn web UI into stretchy morph UI, where transitions squash, stretch, overshoot and settle on springs instead of easing curves. When invoked, first audit the user's existing code for components that could become stretchy (tabs, toggles, buttons, modals, dropdowns, accordions, carousels, toasts, sidebars, date pickers, add to cart and more), then list what can change and ask the user to confirm or add their own requests before editing anything. Includes a spring engine, 24 working reference components, a scanner, a browser checker, and a method for inventing new stretchy components. Use whenever the user wants animated or alive-feeling UI, micro-interactions, spring, elastic, bouncy, gooey…

Route any video task to the right tool BEFORE running it — cut/trim/join/concat/remux/extract-audio, transcode/scale/compress, silence/dead-air removal, programmatic assembly with text/overlays/transitions, color grade / HLG->SDR tone-map / denoise / stabilize / upscale, 3D titles/motion-graphics, or AI/text-based editing. Picks among FFmpeg, auto-editor, MoviePy, DaVinci Resolve (Python API), Blender headless, and Descript MCP, and only pays DaVinci Resolve's ~15x cost when corrective/creative refinement actually earns it. Use whenever deciding HOW to process video on this Windows laptop, especially iPhone HEVC / 10-bit HLG footage (which greys out on a naive SDR H.264 export). Trigger…