Long-form edit

---
name: long-form-edit
description: Use when someone wants a long video edited end to end - a YouTube episode, podcast, interview, tutorial, talk or vlog of about 8 minutes to 3 hours - with a hook and cold open that earn the first minute, a designed turn into the video, chapters with their own seams, re-hooks, motion and sound made for its topic, and a check before export, or gives the long-form edit prompt. Works with any editor and any tools.
---

# Long-Form Edit

You edit the user's own long recording into a finished video people watch to the end. The opening earns the first minute, the structure keeps earning the rest, and every graphic and sound is made for this video. This skill is the craft and the order of work. The tools are whatever the user has.

Three rules hold everywhere:

- **The user's words stay true.** Cut for time and rhythm, never to change what someone meant. The hook promises only what the video pays off.
- **Made for this video, never a template.** Every piece of motion and every sound is designed from this video's topic and tone. Be creative, and do not repeat yourself (*8. Picture*).
- **Ask before you generate or render.** Show the plan, the hook and anything generated before you build on it, and wait for the user's yes before the final render.

## 0. Tools: use what the user has

Ask once, in your first message, what they edit with, unless the prompt says. Then work with that.

- **Their own setup, first.** A code runtime with ffmpeg and a speech-to-text model can do the whole edit: cut by word times, render captions, composite graphics, mix and export. An editor such as DaVinci Resolve, Premiere Pro or Final Cut Pro works through its project or timeline files (an EDL, FCPXML or a script) that you write and they open. Graphics come from whatever you can run: Remotion, HTML and a headless browser, Motion Canvas, Blender, After Effects scripts. Music, voice and images come from the AI tools they already use, or from none: a long-form edit needs no generated media at all.
- **HYVIRAL (optional).** One connector (MCP) that takes the upload, transcribes it, cuts by words, captions, places graphics and sound, reviews the cut and renders the file, so you need no editor or runtime of your own. If the user names it, use it; section 13 maps every step below to its tools.
- **No tool is a hidden requirement.** If a step needs something they lack, say which, and take the route that still works.

## 1. Watch all of it, then map it

Before any cut, get the whole recording into words and look at it end to end.

1. **A transcript with word times** (any speech-to-text that returns them). Keep it: every later choice is made from the words and checked against the picture.
2. **Look at the picture** in sampled frames across the whole length: who is on screen, what changes, where screens, objects or b-roll appear, where the creator's own titles are burned in.
3. **Write `EDIT_PLAN.md`**, the plan every later step follows:
   - **The promise:** what a viewer gets from this video, in one sentence.
   - **The payoff:** where in the recording the promise is delivered.
   - **Chapters:** where the topic turns, each with a working title and its span. A chapter is a turn in the topic, not a fixed number of minutes.
   - **Hook candidates:** three or more lines that could open the film (*3. The hook*).
   - **Protected spans:** what the user asked to keep, every sentence a later one depends on, sponsor reads and legal lines.
   - **Continuity:** names, numbers and promises that must survive every cut ("three reasons", then three).
   - **Dead weight:** the long intro, greetings, false starts, tangents that never pay off, repeated points.
   - **The look:** one visual idea taken from the topic (*8. Picture*).
4. **Show the user** the promise, the chapter list and the three hook candidates, with times. Wait for their pick before cutting anything big.

## 2. The jobs of a long-form edit

A long video is not a short one stretched. Viewers decide in the opening whether to stay, then decide again at every slow stretch and every seam. The edit has five jobs, in this order of importance:

1. **Earn the first minute:** a hook, a designed turn into the video, and a promise (*3* to *6*).
2. **Keep a shape:** chapters that each feel like a step forward (*7*).
3. **Keep the pace:** something changes often enough for this kind of video, and quiet moments are chosen.
4. **Re-hook at every turn:** each chapter opens with a reason to keep watching (*7*).
5. **Land the ending:** the payoff, then out, with no trailing goodbye (*11*).

## 3. The hook: choosing the moment

The hook is a real moment from later in the recording, played first. On a video over about ten minutes it is a cold open of 5 to 15 seconds, up to 30 on an episode over half an hour. On a shorter video it can be the first 5 to 10 seconds of the video itself, tightened.

**What makes a moment hook-worthy**, in order of weight:

- **It stands alone.** A stranger with no context understands it.
- **It opens a question the video answers:** a result without its how, a claim without its proof, a story stopped at its turn.
- **It has energy:** stakes, a contradiction, an exact number, a confession, real emotion or laughter.
- **It does not give away the payoff.** Tease the ending; never show it.
- **It plays well:** clean sound, the speaker's energy up, nothing in the frame to explain.

**Kinds of hook**, and the videos each suits:

| Hook | What it is | Suits |
|---|---|---|
| Result first | The outcome, shown or said, before how it happened | Tutorials, builds, challenges, case studies |
| The contradiction | The line that goes against what the audience assumes | Commentary, opinion, education |
| The stakes | What went wrong, what was at risk, what it took | Stories, business, interviews |
| The confession | The speaker admitting a mistake or a surprise | Podcasts, vlogs, founder talks |
| Already moving | A scene in the middle of the action, no setup | Vlogs, travel, documentary |
| The question | A sharp question the viewer answers in their head | Explainers, debates, Q&A |
| The montage | Three short moments that promise a journey | Long episodes with several high points, only when the user wants it |

Shortlist three, from different kinds where you can. For each, give the time, the exact words, the kind, and one line on why it works. Say which you would pick and why, then let the user choose.

## 4. Cutting the hook

- **In on the claim.** Start on the first word that matters: no "so", "um", "I think" or breath before it. A visual hook starts mid-action.
- **Out on the open question.** End before the answer, or on the reaction that follows the line. Leave a short beat of room after the last word, about a third of a second, so it does not feel clipped.
- **Tighten inside it.** Take out fillers and long breaths, and keep it sounding like the person.
- **Frame 0 does work.** Never black, never a logo alone: the first frame is something happening, because it also sells the thumbnail moment.
- **Say it on screen, too.** Within the first second, a short hook line (seven words or fewer, the stakes or the question) as kinetic type or a bold caption. Many people start the video muted.
- **One camera move,** a punch-in on the key word. More than that turns the hook into noise.
- **Sound:** the hook plays mostly dry, or over a low bed. A rising swell into the turn is a good option.

## 5. The turn: from the hook into the video

The turn is the most designed seam in the film, and it is used once. Pick it from the tone of the video. It needs a cause on screen (a movement, a word, a sound), and its sound lands on the cut.

| Turn | How it plays | Its sound | Suits |
|---|---|---|---|
| Hit to title | The hook's last word lands, the title card slams in on the beat for one to two seconds, then the video | A hit or a bass drop, with the music starting | Confident, fast-cut channels |
| Whip | A fast blurred pan or slide carries the hook out and the video in | A whoosh peaking on the cut | Energetic talk, entertainment |
| Dip and breath | The picture dips to black for about half a second | A low boom, or a beat of silence | Serious, documentary, emotional |
| Rewind | The hook scrubs backwards at speed to the start, with a label such as "Earlier that day" | A tape rewind | Story-driven videos, vlogs |
| Freeze and label | The last frame of the hook freezes, a label or stamp names the problem, then a cut | A record scratch or a stamp hit | Comedy, commentary |
| Match cut | A gesture, word or shape at the end of the hook matches the first shot of the video | A soft whoosh, or nothing | Interviews with b-roll, travel |
| Title under the first line | The video's first sentence starts under the title card, sound before picture | The bed fades in under the voice | Podcasts, calm interviews |
| Zoom through | The camera pushes into a screen or an object in the hook and comes out in the video | A rush of air | Tutorials, screen-heavy videos |

- **Keep it short.** A title or brand sting holds one to two seconds; a long logo intro loses people.
- **Change the music at the turn.** The bed starts, drops or switches here, so the ear hears the film begin.
- **Do not reuse the turn at chapter seams.** It may come back once, at the very end, as a bookend.

## 6. The first minute after the turn

The next 30 to 60 seconds confirm the viewer's choice to stay.

- **The promise,** in one sentence: what they will get. Use the speaker's own words when they said it; otherwise a short title card.
- **A reason to stay to the end:** the stakes, or a tease of the best later moment ("and the third one is the one nobody expects").
- **Why this person,** in one line, when the video needs credibility.
- **A roadmap,** optional: the chapters on screen as a quick list.
- **Cut** the greetings, the "before we start", the "don't forget to subscribe" and the recap of the last episode. If the speaker asks for subscribers, move that line later or shorten it. Never cut or move a sponsor read without asking: it may be under contract.
- **Pace this minute the densest** of the film: tight cuts, the first graphics, the clearest sound.

## 7. Structure: chapters, pace and re-hooks

**Chapters**

- A chapter starts where the topic turns. Give each a title that promises something ("Why the first launch failed"), not a number ("Part 2").
- Give every chapter seam the same treatment, so the film reads as one: one kind of transition, one design of chapter card, one sound. The card is a signature piece, designed once for this video and repeated by design.
- On YouTube, list the chapters in the description: the first at 0:00, at least three, each at least ten seconds long.

**Pace**

- In talk, something changes every 10 to 20 seconds: a cut, a reframe, a graphic, a b-roll shot, a sound. Faster in the first minutes and in lists; slower for a story's emotional turn.
- Cut on sentences. Remove fillers, false starts and dead air, but keep the beat after a punchline and the breath before a big line. A film with every pause removed sounds like a machine.
- Cut tangents that never pay off. Keep the ones that carry the speaker's personality.
- Repetition: when a point is made twice, keep the better take.

**Re-hooks**

- **Open every chapter with a reason to watch it:** its best line as a two-to-four-second flash-forward, a question card it answers, or the chapter's stakes in one line.
- **Plan a change around the middle,** where long videos sag: a new visual mode, a different location or shot, a reveal, a moment of humour.
- **Before a long explanation,** say on screen why it matters.
- **Keep the promise visible.** If the video promised "five mistakes", count them on screen as they come.

## 8. Picture: made for this video, never repeated

**One look, from the topic.** Before you design anything, pick one visual idea that comes from what the video is about, and let every piece follow it. A video about investing might live in a trading terminal; a cooking story in a recipe notebook; a gaming topic in a HUD; a history talk in archive paper and stamps; a science explainer in a lab notebook. Use the creator's colours, fonts and on-screen habits, and extend them rather than replace them.

**Make the pieces yourself, custom for this video.** With a code runtime, design and render the motion graphics from scratch for this film: chapter cards, data and chart cutaways, quote cards, names and titles, maps, timelines, lists that build as they are said, screens in a device, a whole-frame world for one section. Every piece takes its words and numbers from the speaker, enters in time with them, and carries the sentence it sits on. Without a runtime, use the editor's own titles and effects, styled to the same look.

**Be creative, and vary.** A strong long-form edit surprises a little, chapter after chapter:

- **Choose fresh for each video.** Do not carry a set of moves from the last one.
- **Vary inside the film.** A kind of move (a punch-in, a pop-up, a stamp) appears a few times per chapter at most. Only the signature pieces (the chapter card, the names) repeat, and they repeat by design.
- **Rotate within a kind.** Three quote cards can share a design and still differ in layout and entrance.
- **Leave room.** One idea on screen at a time, and calm stretches between moves, so each one lands.

**A menu to start from,** not a limit. Mix, match and invent:

- *Camera:* a punch-in on a claim, a slow push through a long stretch, a reframe to hide a jump cut, a pull-back reveal, picture-in-picture when a screen takes over.
- *Type:* kinetic words as they are spoken, a word stamped on, a checklist that ticks, a comment or question that the next line answers, a headline wall.
- *Data:* bars or lines that grow as each number is said, count-ups, a before and after.
- *Treatments for a different moment:* black and white for a flashback or a "before", a different grade for a memory, a speed ramp or slow replay of a punchline, a freeze-frame with a label.
- *Voice treatments:* a phone or radio band for a quoted call, a small room for a memory. Make them with audio filters on that line only, and keep the captions in sync.
- *B-roll and inserts:* the user's own footage first, then screen recordings, photos, and only then generated or stock shots, each with the rights to use it.

**Avoid what looks generated:** everything sliding up and fading the same way, three cards in a row, floating particles in an empty frame, a title that keeps moving after it lands, the same whoosh on every move, a logo parked in a corner for the whole film.

**Always:** nothing covers a face; text stays clear of the captions and of the platform's buttons; a line someone must read stays on screen long enough to read twice.

## 9. Motion that feels made

- **Everything eases in and out.** Arrivals slow into place, exits speed away, camera moves ease at both ends. A hard pop is a deliberate hit on a beat, not a default.
- **Overlap.** The next element starts before the last one finishes; words a couple of frames apart.
- **What has landed holds still.** A held shot may drift a few percent so it never looks frozen; text and cards do not bob or pulse.
- **Transitions have a cause:** a cut on a movement, a card growing into the next frame, a word rushing past.
- **Inside a scene, hard cuts.** Save designed transitions for the turn and the chapter seams.

## 10. Sound and music

- **The voice leads.** While someone speaks, music sits well under the voice, about 15 dB as a starting point, and effects stay below the voice too. Measure it rather than guess.
- **A bed per chapter or section,** not one track looping for an hour. The music changes at the turn and at seams, fades in and out, and ends on its own ending.
- **Silence is a tool.** Drop the music before a big line or a reveal, and bring it back after.
- **Effects with a cause:** on graphic entrances, transitions and reveals, made for the look (paper sounds for a paper world, clicks for a UI). About two or three a minute as a starting point, never on top of words. Rotate takes of a repeated sound.
- **Clean the voice lightly:** reduce hum and noise, level it, and do not over-process.
- **Loudness:** YouTube plays loud uploads back at about -14 LUFS, so mix near it, with true peaks at or under -1 dBTP.
- **Rights:** every track and effect must be cleared for where the video goes. On YouTube an uncleared track draws a copyright claim.

## 11. Captions, text and the ending

**Captions**

- On 16:9 for YouTube, upload a caption file (SRT) for the CC button. Burn captions into the picture only when the creator's style uses them, then large, phrase by phrase, readable on a phone held sideways.
- For a vertical cut of the same video, burned captions are a must.
- Check names, numbers and jargon in the captions by hand.

**The ending**

- Land the payoff, then get out: cut "so yeah, that's about it" and the long goodbye.
- The call to action is one short line, after the payoff, never before it.
- On YouTube, leave room for end screens: they show in the last 5 to 20 seconds.
- A bookend works well: echo the hook's line or picture, now answered.

## 12. Check it before export

1. **Watch the first minute three times:** at full speed with sound, muted, and at phone size. Would a stranger stay?
2. **Walk the seams:** the turn, every chapter seam and the ending, each with a few seconds either side.
3. **Answer these** for the whole film:
   - Does the hook make a promise the video keeps?
   - Does something change often enough, without constant motion?
   - Is any move, graphic style or sound used so often that it stops being noticed?
   - Can every text be read on a phone, clear of faces, captions and buttons?
   - Is the voice always on top, and does the loudness land near the target?
   - Do names, numbers and the "three things" survive every cut?
4. **Fix what fails,** check those windows again, then render.
5. **Report:** the file, the length before and after, the hook you chose and why, the chapter list with times (ready to paste into a YouTube description), what each chapter carries, the tools you used, and anything the user should check by eye.

## 13. If you use HYVIRAL (optional)

HYVIRAL is one way to run every step above through a single connector, `https://api.hyviral.com/mcp`, added as a custom connector in the user's AI app. Its own skill, <https://hyviral.com/skill/SKILL.md>, has the full tool reference; for this kind of edit read its `playbooks/edit-existing.md` and `references/long-form.md`. Some connector steps (the first watch, a review, generated media and the export) return a quote and wait for the user's yes: show them what it is, and call again only after they agree.

| Step in this skill | With HYVIRAL |
|---|---|
| The source | `upload_request` gives the user an upload link; the connector never fetches a URL, a YouTube link included |
| 1. Watch and map | `video_ingest` once, then `video_project` `describe` for the parts and `read` for the words with their ids |
| A safe copy | `fork_project` once; work on the copy |
| 3 to 5. Hook and turn | `video_cut` `splitClip` at the hook's first and last word, then `reorderClip` to the front; `video_overlays` `setClipTransition` on the turn (`dip-to-black`, `whip-pan`, `zoom-punch`, `fade`, `wipe`, `slide`, `glitch`); `addText` for the hook line |
| 7. Tighten | `video_cut` with filler removal, scoped away from protected spans |
| 7. Chapter seams | `splitClip` at each chapter's first word, then one transition and one card per seam, set last |
| 8. Camera and treatments | `video_look`: zooms and punch-ins, `colorGrade` such as `bw` for a flashback, speed changes |
| 8. Custom pieces | Render them in your runtime and place them with `video_overlays` `addOverlay`; the `hyviral-studio` skill (<https://hyviral.com/skill-studio/SKILL.md>) has a kit for them |
| 10. Music and effects | `video_audio`: a bed per chapter with `beds`, and `addSound` for each effect |
| 11. Captions | `video_captions` before the graphics |
| 12. Check | `video_preview` for frames and a link to watch, `video_review` on the open, the seams and the busy stretches |
| Export | `video_export` once, then follow its job until it is delivered |

Everything else in this skill (the hook, the turn, the look, the variety, the check) is the same whichever tools do the work.

# The prompt

First give your agent the long-form-edit skill (paste the skill text and ask it to save it as a skill). Then replace everything in [BRACKETS] below and paste it into a new chat. Anything you leave in brackets is your agent's call.

**Tools:** use whatever you already edit with: DaVinci Resolve, Premiere Pro, Final Cut Pro, or an agent that can run code with ffmpeg. The skill works the same with all of them. If you'd rather have one connection do the cutting, captions, graphics, sound and render, HYVIRAL is one option: add `https://api.hyviral.com/mcp` as a custom connector in your AI app (steps: https://hyviral.com/docs/mcp/connect).

**Have the file ready.** Give your agent the original video file. If it's only on YouTube, download it from YouTube Studio: Content, then the video's ⋮ menu, then Download.

---

Use the long-form-edit skill to edit my video into a finished long-form video.

- **The video:** [VIDEO - what it is and how long, e.g. "a 2-hour podcast episode, two hosts, 16:9"]
- **Where it goes:** [PLATFORM - e.g. "YouTube, 16:9", or "YouTube, plus a 9:16 cut of the best minute"]
- **What I want back:** [LENGTH - e.g. "the whole episode, tightened", "a 12-minute cut", or "about 25 minutes"]
- **Who it's for:** [AUDIENCE - e.g. "people starting their first small business"]
- **Must keep:** [KEEP - a line, a name or a moment that must not be cut, or "nothing in particular"]
- **Names on screen:** [NAMES - e.g. "Ana Ruiz, host; Tom Lee, guest", or "none"]
- **Music:** [MUSIC - a mood in a word or two, my own tracks, or none]
- **Look:** [LOOK - e.g. "warm and clean", my brand colors as hex codes, or "your call, from the topic"]
- **Tools:** [TOOLS - e.g. "DaVinci Resolve", "ffmpeg and Remotion", "the HYVIRAL connector", or "whatever you can run"]
- **Avoid:** [AVOID - e.g. no stock footage, don't cut the sponsor read, no swearing in the captions]

Work in this order:

1. Get the whole video into words and look at it end to end. Then show me the promise, the chapter list with times, and three hook candidates with the exact words, and say which you'd pick. Wait for my pick.
2. Build the opening: the hook, the turn into the video, and the first minute. Show it to me before you go on.
3. Edit the rest: tighten it, set the chapters and their seams, re-hook each chapter, and design motion graphics and sound effects made for this video's topic. Be creative and don't repeat yourself.
4. Show me anything you want to generate before you generate it, and wait for my yes before the final render.
5. Check it as the skill says, fix what fails, render, and give me the file, the length before and after, the hook you chose and why, and the chapter list ready to paste into the YouTube description.