Skip to content

TimoteoAltobrando/ai-video-skill

ai-video

Unverified

Make and edit videos with Claude Opus 5.5, end to end, from a blank page or from the user's own footage — product and launch films, ads, talking-head and founder videos, interviews, explainers, tutorials and screen demos, event and trip montages, highlight reels, podcast clips, Reels/TikTok/Shorts, animated logos and intros, music and lyric videos, and changes to a finished video (shorter cut, vertical version, subtitles, translation, new music). Remotion is the backbone; Codex (image_gen) makes images, Seedance via OpenRouter animates them, ElevenLabs makes music and sound effects, ffmpeg and whisper handle footage and speech. It starts by asking what kind of video it is and everything that kind needs (platform, format, voice, music, rights, who must not appear, budget), then goes through direction, a script or paper edit approved piece by piece, production, revisions and delivery for each platform. Use it whenever someone wants to make, cut, edit, shorten, reformat, subtitle or improve a video, even if they do not mention any of these tools.

FileKindSize
SKILL.mdskill md12.9 KB
assets/templates/brief.mdmarkdown2.5 KB
assets/templates/edit.mdmarkdown971 B
assets/templates/image-prompt.mdmarkdown1.6 KB
assets/templates/music-plan.jsonconfig7.2 KB
assets/templates/script.mdmarkdown1.0 KB
references/delivery.mdmarkdown6.1 KB
references/direction.mdmarkdown11.6 KB
references/editing.mdmarkdown9.0 KB
references/intake.mdmarkdown11.2 KB
references/music.mdmarkdown12.6 KB
references/on-camera.mdmarkdown17.6 KB
references/remotion.mdmarkdown11.3 KB
references/tools.mdmarkdown10.2 KB
scripts/blur_background.pyexecutablescript17.1 KB
scripts/clean_plate.pyexecutablescript6.1 KB
scripts/codex_image.shexecutablescript1.2 KB
scripts/contact_sheet.pyexecutablescript2.3 KB
scripts/elevenlabs.pyexecutablescript5.0 KB
scripts/export.shexecutablescript3.0 KB
scripts/hdr_to_sdr.shexecutablescript1.4 KB
scripts/inventory.pyscript5.4 KB
scripts/measure_music.pyexecutablescript6.7 KB
scripts/person_mask.swiftscript4.2 KB
scripts/prepare_bed.pyexecutablescript4.0 KB
scripts/seedance.pyexecutablescript6.9 KB
scripts/wall_perspective.pyexecutablescript10.4 KB

references/intake.md

View on GitHub
# Intake: the questions before anything else

A video goes wrong most often because something was not asked at the start: the platform (and so the format), who
must not appear, whether the music can be published, whether AI images are welcome at all. This file is the full
question tree. Read it at the start of every project.

## Contents
1. How to ask
2. Round 1 · What, from what, where
3. Round 2 · The core (every video)
4. Round 3 · The module for the kind of video
5. Defaults when the user does not know
6. The brief back to the user, and the gate
7. Questions that come later, not now

## 1. How to ask

- **Three rounds at most** before work starts, each short: round 1 decides the path, round 2 the core, round 3 the
  module for the kind of video. Never more than about ten questions in one message.
- **Skip what is already known.** If the request already says "a 30 s vertical for Instagram about our app", do not
  ask the platform, the length or the kind again: list them as understood and ask only for the rest.
- **Closed choices go in the question tool** (up to four questions per call, four options each, "Other" is added by
  the tool); recommended option first. **Open questions** (paths, links, names, sentences) go in a numbered list in
  chat, so the user can answer by number.
- **Every question has a way out**: "if you do not know, I will use …" (section 5). The user should never be stuck.
- **Explain why when it is not obvious** ("I ask who must not appear because a face cannot be removed from a montage
  after it is cut").
- Keys and accounts are asked only for the services the chosen path really uses (`tools.md`), after round 3.

## 2. Round 1 · What, from what, where

One call to the question tool, four questions:

| # | Question | Options (single choice unless noted) |
|---|---|---|
| 1 | Where do we start from? | Created from scratch (no footage) · Footage I already have, to edit · Both: my footage plus created graphics · A finished video to change (shorten, vertical cut, subtitles, translate, new music) |
| 2 | What kind of video is it closest to? | Promote something (product or launch film, ad, teaser) · A person speaks (talking head, founder, interview, testimonial, lesson) · Explain or show (explainer, tutorial, screen demo, data story) · A montage of footage (event, trip, highlights, family, podcast clips) — "Other" covers an animated logo or intro, a music or lyric video, kinetic typography, anything else |
| 3 | Where will it be watched? (multiple choice) | Website or landing page · LinkedIn or X feed · Reels, TikTok or Shorts (vertical) · Big screen, event, TV, or private (family, internal) |
| 4 | How long, roughly? | Up to 15 s · 15-60 s · 1-3 min · Longer |

The answers pick the module of round 3 (section 4) and the defaults of format and length (section 5). Several kinds
can apply (a founder film with product animations): take the modules of all of them, without asking twice.

## 3. Round 2 · The core (every video)

**Open questions** (numbered list in chat):
1. In one sentence: what should the viewer **understand, feel or do** at the end?
2. **Who watches?** What do they already know, what do they care about?
3. **Materials**, as paths or links: brand guidelines and the official logo (or "none, it is personal"); footage,
   photos, recordings; music the user owns; references they like and, if any, ones they do not like.
4. **Names and words** that must be written exactly: people (name and title as they want them), product, places,
   dates; the languages of the video.
5. **Deadline**, and **who approves** (one person or several; how fast they answer).

**Closed choices** (question tool):

| Question | Options (recommended first, depending on the kind) |
|---|---|
| The voice | No voice · A person on camera or interviewed · A voice-over recorded by a person · A synthetic voice (only with the consent of whoever it imitates, if anyone) |
| Words on screen | A few key words · Subtitles burned in · Word-by-word captions (social style) · None |
| Music | Generated for the video (ElevenLabs) · A track the user owns or licensed (rights for that platform?) · A royalty-free library · No music, only the real sound |
| Images and AI | Real footage and motion graphics only · AI images and AI animation are welcome · A mix: AI for the backgrounds and the world, never for people and products |

**Rights and consent**, always, in one short question with the reasons: are the recognisable people in the footage
OK with being in it (and are there minors)? Who must **not** appear? Can the music be published where the video
goes? Are there other companies' logos, artworks or screens with personal data in shot?

**Budget**: a cap in money for paid services (Seedance, ElevenLabs credits, stock), even if it is "zero: only free
tools".

## 4. Round 3 · The module for the kind of video

Ask only the module(s) chosen in round 1. "Must" questions block production until answered or until the user accepts
the default; the others can wait.

### A · Promote something (product or launch film, ad, teaser)
- Must: where to see the **real product** (code, docs, site, test account, screenshots), to copy real screens and
  labels; **the one thing** it does better than anything else; the **call to action** and its exact link.
- Must: which numbers or promises are **verified** (and the source); what must not be claimed.
- Ad only: the platform's ad specs, how many **variants** (different hooks, lengths, formats) for testing, required
  disclaimers, who checks the claims legally.
- Nice: competitors and how they talk; a launch date and the posts that go with it (`delivery.md`).

### B · A person speaks (talking head, founder, interview, testimonial, lesson)
- Must: **is it already shot?** If yes: the files, phone or camera, how it was shot. If not: give the shooting guide
  (`on-camera.md` section 14) before they shoot, it saves days.
- Must: who the person is, name and title exactly; is there a **script** or are we cutting a free talk? A teleprompter?
- Must: interviews and testimonials: written consent of every person shown.
- Later, after looking at the footage: room, light, blur, voice clean-up, graphics around the person
  (`on-camera.md` section 1, SKILL Phase 3b).

### C · Explain or show (explainer, tutorial, screen demo, data story)
- Must: the viewer's **level** (beginner, expert) and **the steps** in order; the version of the product or tool shown.
- Must: is there a **screen recording**, or do we record it or rebuild the interface in Remotion? A recording must be
  sharp (the screen's native resolution), with notifications off and **no personal data** (emails, names, tokens in
  the address bar).
- Must: a voice script or not; chapters for longer videos.
- Data stories: the data source and its date; the numbers are checked against the source before animating.
- Nice: zooms and callouts on the important clicks; a final recap.

### D · A montage of footage (event, trip, highlights, family, podcast clips)
- Must: **the footage**: where it is, how much (number of files, total hours), from which devices.
- Must: **the order**: chronological · best moments (open strong, build, peak, calm end) · by theme · around a speech
  or an interview.
- Must: **moments and people that must be there**; **who must not appear** (consent, minors, ex-partners, clients).
- Must: the **mood and pace**: calm, energetic, emotional, funny; one reference video helps more than adjectives.
- Must: the **real sound**: keep the ambient sound, speeches, laughter, or only music?
- Nice: titles (place, date, names), a colour look, a vertical version, several lengths (a 60 s highlight and a 15 s
  teaser from the same material).
- Podcast or talk clips: which episode, which moments (or "find the best 3"), captions style, speaker names.

### E · Short vertical video (any kind, when round 1 says Reels, TikTok or Shorts)
- Must: **the hook**: what happens in the first second (a question, a result first, a movement); one-off or a
  **series** with a recurring format?
- Must: captions style (word by word, big, in the safe area); the call to action (follow, link in bio, comment).
- Nice: a trending sound is added in the platform's app, not in the file (rights); a cover frame.

### F · Animated logo, intro or outro
- Must: the official logo files (SVG); the duration (3-6 s); where it is used (start, end, both).
- Must: a **sound logo** or silent; a **transparent background** (to lay over other videos: ProRes 4444 or WebM with
  alpha) or a solid one.

### G · Music or lyric video
- Must: the final track and its rights; the lyrics with timings (or we time them with whisper); the tempo.
- Must: performance footage, created world, or words only.

### H · A finished video to change
- Must: the **original project** (Remotion, Premiere, Final Cut, CapCut…) or only the final file? With only the file,
  texts burned into the picture cannot be changed cleanly.
- Must: **what changes**: shorter (to how long, what can go), another format (vertical: who stays in frame), subtitles
  or a translation (which languages, who checks them), new music, an audio fix, a new ending.
- Must: what must stay identical.

## 5. Defaults when the user does not know

| Topic | Default |
|---|---|
| Format | Feed and website 16:9 1920×1080; Reels/TikTok/Shorts 9:16 1080×1920; LinkedIn feed also 1:1 or 4:5 |
| Frame rate | 30 fps (or the footage's own rate); 24 fps for a fully animated film |
| Length | Vertical short 15-30 s; ad 15-30 s; product film 45-60 s; founder film 1:30-2:30; tutorial as long as the steps, chapters every 1-2 min; event highlight 60-90 s |
| Words | Subtitles burned in when someone speaks (most people watch muted); otherwise a few key words |
| Music | Generated on the edit; ambient sound kept low under it in montages |
| AI | Only for backgrounds and invented worlds; never to fake real people, products or places |
| Languages | The language the user writes in |

## 6. The brief back to the user, and the gate

After round 3, before analysis: write `docs/brief.md` (`assets/templates/brief.md`) and show it in chat in **one
screen**: the goal sentence, the kind and the path, format and length, voice, words and music, materials found,
rights, deadline and budget, the open doubts. End with: "Is it right? Anything missing?"

**Do not go on** until every line below is answered, or the user accepted the default:
- [ ] Goal sentence and audience.
- [ ] Kind of video, starting point, platforms, format, length.
- [ ] Materials located **and opened** (a path that does not open is not a material).
- [ ] Voice, words on screen, music decided.
- [ ] Rights: people, music, logos; who must not appear.
- [ ] Approver and deadline; budget cap for paid services.
- [ ] Every "must" of the chosen module(s).

## 7. Questions that come later, not now

Some questions are better asked when there is something to look at, not in the first rounds:
- the room, the light and the graphics around a person on camera: after the footage is analysed (`on-camera.md`);
- the visual style: with 2-3 directions shown as images (SKILL Phase 3);
- the exact wording of key words and closing lines: with 2-3 alternatives on the real frames (`direction.md`);
- which voice clean-up, which music take: by ear, with numbered versions (`music.md`, `on-camera.md`).
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall