Home
Resources
Agent Skills
Video styles
Terminal — the dense technical explainer video style

Terminal — the dense technical explainer video style

A video-style skill that directs a short vertical video as a dense technical explainer in the manner of Fireship's videos, rewritten from a frame-by-frame study of them, in two modes: the board, where one diagram owns a section and changes on every spoken clause with real clips inside its boxes, and the brief, where a literal picture (a real screen, a stack of captures, a cut-out object, a big word) lands on every line with a cut every one to two seconds. It fixes the rule of one built picture per idea, the material, a motion language, three layers, the elements and their verbs, cutting on the transcript's word timings, captions, how to make every picture from material the client owns with no memes and nothing AI-generated, and a review checklist. It reproduces the moves, never anyone's footage or material.

Studied from Fireship

Updated 14 October 2026 · By Sadeq

When to use it

Direct and build a short vertical video in the terminal style: a dense technical explainer on a black field in two modes. The board: one diagram stays on screen for a whole section and changes on every spoken clause (a layer added, a box swapped, a group pushed aside for the next), with real clips set inside its boxes. The brief: a real screen, a stack of captures or a cut-out object for every line, a big word landing word by word, cuts every one to two seconds. Type is white and monospaced and a short sound marks each hit. Use when someone asks for a video "in the terminal style", "like a Fireship video", a "how it works" or "in 100 seconds" explainer, a tech news brief, or any piece whose script is a list of concrete facts. Do not use it for a mood piece, a founder story, or anything that should feel warm or slow.

Complete file

SKILL.md

---
name: terminal
description: >
  Direct and build a short vertical video in the terminal style: a dense
  technical explainer on a black field in two modes. The board: one
  diagram stays on screen for a whole section and changes on every
  spoken clause (a layer added, a box swapped, a group pushed aside for
  the next), with real clips set inside its boxes. The brief: a real
  screen, a stack of captures or a cut-out object for every line, a big
  word landing word by word, cuts every one to two seconds. Type is white
  and monospaced and a short sound marks each hit. Use when someone asks
  for a video "in the terminal style", "like a Fireship video", a "how it
  works" or "in 100 seconds" explainer, a tech news brief, or any piece
  whose script is a list of concrete facts. Do not use it for a mood
  piece, a founder story, or anything that should feel warm or slow.
license: MIT
metadata:
  author: daena-digital
  version: "2.1"
  collection: video-styles
---

# Terminal — the dense technical explainer style

A terminal video looks like a fast reader's screen: black, white
monospaced type, and every sentence pointed at. It runs in two modes. In
**the board** — the explainer — one diagram owns the screen for a whole
section and changes on every clause: hardware, then the system on it,
then two boxes on that, then one box swapped for a better one, while a
real clip plays inside a box to make the point. In **the brief** — the
news update — a real screen, a stack of captures or a cut-out object
arrives on almost every clause, a big word lands word by word, and a
small sound marks each hit. Nothing decorates. Every picture is the
literal subject of the words over it.

Studied frame by frame (every fifth of a second, and every half second)
from Fireship's own YouTube videos — see Reference. This skill reproduces
the **moves** — the field, the board, the literal picture per line, the
density, the cutting — never his footage, his memes, his screenshots or
his words.

## When to pick terminal over another style

- The script is **a sequence of facts**: what it is, what it does, what
  changed, how the pieces fit. Each sentence can be pointed at.
- The assets are **screens**: the client's own product, site, dashboard,
  code, documentation, terminal output. The style needs one real screen
  for roughly every two sentences.
- The audience is **technical or impatient**, and the piece can run at
  180 words a minute or faster.
- There is **no footage** and no appetite for generated imagery; every
  picture is a capture of something that exists.

Pick something else if the script is about feeling rather than fact, if
it has a place and photographs (grain and map), if it needs to feel
crafted and warm (paper studio), or if a person on camera is the point.

## Before building

Run the concept gate in the `video-craft` skill first: when there is a
reference clip, its essence in three sentences that name none of its
objects; the message line; three concepts; and a storyboard in which every
frame says why it means its line, approved before anything renders. Every
picture passes its two tests — only this subject could have produced it,
and one sentence says why it means its line — and every cut follows the
seam rules there: the same axis, direction and speed on both sides, the cut
landing mid-motion, one main direction for the film, an object carrying the
eye across. Those rules are adapted from HeyGen's open-source HyperFrames
skills (Apache License 2.0); the full credit is in `video-craft`.

In the brief the cuts are hard, so the push before a cut and the push after
it run the same way at the same speed. In the board there is no cut inside a
section: the main direction is the camera's travel across the board in the
order the steps happen, and it comes back only when the diagram itself
loops. A line drawing from one box to the next, or a flying document
landing in its slot, is the carrier.

## One built picture per idea

Every clause of the script gets a picture that **acts the idea out**,
and it moves for as long as the idea is spoken.

- **Something changes on every clause**: a box added, a layer swapped, a
  capture landing, a word typed. Nothing on screen is still for more
  than a second and a half.
- **An explanation is a board, not a slideshow.** A "how it works"
  section keeps one diagram for its whole length and changes it, so the
  viewer watches the system being assembled rather than reading slides
  about it.
- **A capture is raw material.** It is cropped to the field the sentence
  is about, set inside a diagram box, stacked, or flown in as a
  document — never the whole explanation with a highlight on it.
- **Open on an object.** The first seconds show a real object that is
  the subject's name or its literal meaning (a shipping container for
  "container", a form for "enquiry"), cut out from a real photograph and
  placed on the field, before the board begins.

## The two modes

| | The board | The brief |
|---|---|---|
| For | How it works, what it is, why it is built that way | What changed, what happened, a list of facts |
| The frame | Black with a faint field of dots; the diagram in the middle third, the label column at its left | A picture full-bleed, pushing in |
| Change | One piece of the diagram per clause: added, removed, swapped, moved | One picture per clause: a capture, a stack, a document, a cut-out |
| Cuts | None inside a section; one between sections | Every 1.5–2.5 seconds, on the word |
| Proof | A real clip or capture placed inside a box | A capture with the read rows highlighted, a headline stack |

A short can use both: a brief to say what changed, a board to say how
it works.

## The world

One material, obeyed everywhere. Decide these before a single frame:

| Property | Choice |
|---|---|
| Surface | True black, or the brand's darkest neutral if it is within a few steps of black, with a faint field of small dots (stars at 10–20% white) that drifts slowly. No gradient, no wash. In the brief, a picture fills the frame edge to edge and black is never the composition; in the board, the black is the board's table and the diagram fills its middle |
| Board | Boxes with one-pixel white borders (or filled in a flat colour for the thing being built), monospaced labels inside, a small icon or logo-free picture where the box names a thing. Layers stack from the bottom up; a bracket and a label in the left column name each layer |
| Cutout | A real object or person from a licensed photograph, cut out with a hard edge, placed on the field at full colour. Never a meme, never a public figure, never a generated picture |
| Type | A monospaced face for everything: lines, labels, counters, captions. White at full strength for the current line, 60% for anything that has already been said. The brand display face may appear once, on the title card, and nowhere else |
| Accent | One colour, the brand's, used for the one highlighted run in a document, the active box in a diagram, the underline on the headline that matters. Never as body text |
| Screens | Real captures at 2x, cropped to the region the sentence is about and blown up to fill the frame's width, in a thin rounded frame with a one-pixel border at 20% white. A capture is never shrunk to fit; when the sentence is about one field, that field is the whole picture |
| Footage | Real footage for a literal human beat (a person reading, two people at a table), muted, full-bleed, under a scrim only where type sits on it. Never generated, never a clip whose subject is somebody else's brand |
| Stack | A second capture lands over the first, offset 24 px down and right, the earlier one dimmed to 60%. Two or three deep at most; the newest is always on top |
| Highlight | A translucent block of the accent over the exact words or rows the voice is reading, on a document or a screen, with everything outside the block dimmed to 50% |
| Headline stack | Two to four real headlines, each a white card with the source name small, stacked with a slight offset, the newest on top |
| Sound | A short library sound on each landing: a click for a label, a soft thud for a screen, a tick for each diagram box, a whoosh for a stack. One family of sounds for the whole video, mixed well under the voice |

## Motion language

One ease per job, and the same ease every time that job comes up.

| Property | Choice |
|---|---|
| Cut | Hard, on the first word of the clause the new picture illustrates, at most three frames early and never late. In the brief the pace is 1.5–2.5 seconds a cut on a 9:16 short. In the board there is no cut inside a section |
| Board change | One piece per clause, on its word: a box pops in from 90% over four frames, a layer slides up from below over six, a swapped box fades out as its replacement pops in on the same spot. Everything else on the board holds still so the change is the only motion |
| Make room | When the voice moves to a new subject, the group already on screen scales to about 50% and slides to one side over 10–12 frames, and the new subject enters on the freed side. The old group stays, small, until the comparison is made |
| Slot clip | A short real clip plays inside a board box, cropped to the box, to show what the box means (slow, broken, busy). It starts on its word and loops once at most |
| Object entrance | A cutout scales up from 0 with a short overshoot over six frames, or swings in from off frame on a slight rotation that settles |
| Document fly-in | A capture enters on the black from about 40% scale with a 3D tilt of 15–25 degrees that settles flat over 8–12 frames, landing with a thud |
| Kinetic word | In the brief, one big word or short phrase in a heavy face lands word by word over the picture, in the accent, two to three frames a word |
| Glitch | In the brief, a section change may be a 3–4 frame glitch: the picture sliced into horizontal bands that shift sideways with a colour split. Twice a video at most |
| Screen entrance | Pops to full size from 96% over four frames with a short ease out, and a thud. No spring, no overshoot |
| Highlight | Snaps on in one frame, exactly when the voice reaches the first highlighted word; the dim outside it follows over three frames |
| Diagram | Each box pops in on its word with a tick, each line draws over six to ten frames after both of its boxes exist. A diagram is never shown finished; it is always built |
| Headline stack | Each card slides in from the right over six frames and lands with a whoosh, the previous cards shifting down and back by a few pixels |
| Labels and lines | Type on at three to four frames a word, or cut in whole if the line is under three words. No letter-by-letter typing; the pace does not allow it |
| Punch-in | On a point that matters, the picture scales from 1.00 to 1.15 over four frames and holds. Once a scene at most |
| Flash | A two-frame flash of white or the accent on the title cut and on a punch-in, decaying over four frames. Twice a video at most |
| Exit | Hard cut. Nothing leaves on its own; the next picture replaces it |
| Camera | Every picture moves: a slow push of 1.00 to about 1.06 over its scene on a capture or a clip, in addition to any punch-in. A still picture reads as a slide |
| Hold | In the brief, a picture with nothing new entering for more than two seconds should have been cut, or had a second picture stacked on it. In the board, a board with nothing changing for more than a second and a half is a slide |
| First frame | The opening frame is the subject's object already placed on the field with the title, or the title in the display face over the first screen at 40%. Frame 0 is the poster |

Breaking a rule once is a choice the viewer feels. Breaking it twice is a
different style.

## Three layers, never fewer

Every scene has all three. A scene with one layer is a slide.

1. **Background.** In the brief, the picture full-bleed: the capture or
   the clip filling the frame edge to edge and pushing in slowly. In the
   board, the black with its drifting field of dots and the label column.
   A small picture alone in the middle of bare black is the one thing
   this layer must never be.
2. **Midground.** In the brief, what sits on the picture: the stacked
   capture, the document, the big word. In the board, the board itself
   and the clips inside its boxes.
3. **Foreground.** The reader's marks: the highlight block, the arrow
   to the thing being named, the label that says what the picture is,
   the counter. These are what make the picture read as *pointed at*
   rather than *shown*.

Two focal points at most: the picture and the highlight on it, or the
box being built and the label naming it.

## The elements and their verbs

Each thing on the screen has exactly one way of arriving.

| Element | What it is | Its verb |
|---|---|---|
| Board | One diagram that owns a section | Changes one piece per clause and never cuts away |
| Slot | A board box holding a real clip or capture | The clip starts on its word inside the box |
| Cutout | A real object or person cut out of a licensed photograph | Scales in with a short overshoot, or swings in and settles |
| Screen | A real capture, cropped to the subject, filling the width | Pops in on its word with a thud, then pushes in slowly |
| Clip | Real footage of a literal human beat, muted, full-bleed | Cuts in already moving and pushes for the scene |
| Stack | A second capture over the first, offset | Lands on its word with a thud, the one under it dims |
| Document | A capture of text or code, with one highlighted run | Pops in, then the highlight snaps on when the voice reaches the run |
| Diagram | Boxes, lines, labels, built in the order the voice names them | Boxes pop on their words, lines draw after both ends exist |
| Headline stack | Two to four real headlines with sources | Cards slide in from the right and stack |
| Label | A short monospaced caption naming what is on screen | Types on at a few frames a word |
| Line | The one spoken phrase that matters, set large in monospace | Cuts in whole on its first word, the marked run in the accent |
| Kinetic word | One big word in a heavy face over a picture | Lands word by word in the accent |
| Flying document | A capture of a page on the black | Flies in tilted and settles flat |
| Arrow | A straight white line with a head, from a label to a thing | Draws over six frames, head last |
| Punch-in | A scale on the current picture | Scales in four frames and holds |

## Cutting on the voice

The transcript's word timings are the edit decision list. Build the
video from them, not from a bar count.

1. **Transcribe with word timings** (any aligner; the timings drive every
   cut) and write them down as frames at the output frame rate.
2. **Cut on every clause that names a new thing.** A new picture begins
   when the voice names something it has not shown. Scenes run one to
   two seconds; a 20-second short has twelve to sixteen cuts, not seven.
   A clause with no new thing shares the previous picture and gets a
   punch-in, a stacked capture or a label instead, never a bare hold.
3. **The picture is literal.** When the voice says "the dashboard", the
   dashboard is on screen. When it says "three services", three boxes
   exist. When it says a name, the name is typed. A metaphorical picture
   is a picture from another style.
4. **Something lands every second.** A screen, then a highlight on it,
   then a label, then the next screen. The sound on each landing is what
   lets a viewer keep the pace with the audio off.
5. **Write the beat table before building.** Three columns: frames, the
   words spoken, what lands. If a row's third column is empty, the voice
   is narrating a black screen — cut the sentence or find its picture.

| Frames | Spoken | What lands |
|---|---|---|
| 0–40 | "Daena Digital builds the systems" | Frame 0: title on black, the homepage capture behind at 40%. On "builds" the capture pops to full and the title cuts out |
| 40–82 | "websites, apps, and the integrations that tie them together" | Three boxes tick in on their three words; two lines draw on "tie them together" with the accent on the middle box |
| 82–118 | "and Sima runs the marketing" | Hard cut to a capture of a real campaign report, cropped to the region, a highlight snapping on the row the voice reads |

A board section, for "how a site gets found":

| Frames | Spoken | What the board does |
|---|---|---|
| 0–30 | "Your site is a stack of pages" | Frame 0: a cutout of a real paper stack on the field, the title over it. On "pages" it shrinks left and five page boxes stack up on the right, labelled "PAGES" |
| 30–60 | "a crawler reads each one" | A crawler box pops in above the stack; a line draws from it to each page box in turn, each box filling with its capture as the line reaches it |
| 60–95 | "and decides what it is about" | The crawler box is swapped for a box holding a slow real clip of somebody reading, then four labels pop in beside the stack, one per word the voice names |

## Captions

Word-synced, three to five words a page, in the monospaced face, white
on no bar, the spoken word in the accent. They sit low in the safe zone
and never share a row with a label. In this style the captions are part
of the type system, so their face and weight match the labels exactly.

## Assets without generating anything

The style's honesty is that every picture is a capture of something that
exists. Make the assets from material the client owns or has the right
to show.

- **Screens.** Captures of the client's own product, site, dashboards,
  documentation and code at 2x, cropped to the region the sentence is
  about. Blur or remove any customer data, any third-party logo and any
  account name before placing. A screen the client does not own is not
  shown; it is named in a label.
- **Documents.** The client's own pages, specs, changelogs or code. A
  public standard or documentation page may be captured where its licence
  allows and credited in the provenance.
- **Headlines.** Only real headlines, with the real source name, and only
  where quoting a headline is permitted. The card shows the headline and
  the source; it never reproduces the article's body or image.
- **Diagrams and boards.** Drawn in the composition from the script's
  own nouns. Never traced from somebody else's diagram.
- **Cutouts.** Real objects from licensed stock photographs (or the
  client's own), cut out with any mask tool. No public figures, no
  branded products, nothing generated.
- **Footage.** Real stock footage with a licence that allows use beside
  a brand, for the literal human beats only, each clip credited by source
  and id in the provenance. A clip that shows somebody else's logo or a
  recognisable product is cropped or dropped. Never generated.
- **Sound.** Short effects from a library whose licence permits this use
  (a paid pack, or CC0), one family for the whole video, each credited
  in the provenance. No generated sound effects.
- **Type.** An open-licence monospaced face (JetBrains Mono, IBM Plex
  Mono, Space Mono). Record the licence.
- **Voice and music.** Whatever the client's provenance rules allow,
  credited the same way.

No AI-generated image, clip or sound effect on screen. No memes, reaction
clips, public figures or other people's screenshots: the studied
channel's humour runs on other people's material and this style does
not. Its place is taken by a licensed stock clip set inside a board box
(a person waiting, a door that sticks) that makes the same point. If a picture does not
exist, the answer is a label, a diagram built from the script's nouns, or
a cut sentence, not a synthetic screen.

Every asset made or chosen for the video — each capture, each crop, each
clip, the voice take, the music, each sound effect — is shown to the
person and approved on its own before it is placed. A changed asset is
shown again. Nothing reaches the render on the maker's say-so.


## Content rules that travel with the style

- No numbers on screen that are not in the script and sourced. A counter
  is a number; a counter with no source is a fabricated number that
  moves.
- A headline stack is evidence, so every card is real and the source is
  named. Three real headlines are better than four with one paraphrased.
- The highlight marks what the voice is reading, never a sentence it is
  not.
- No client logos, awards or testimonials unless they are real and
  permitted.

## Review before export

- [ ] The concept gate in `video-craft` was run, and its review list passes.
Render one frame per second as a contact sheet and check it against this
list. A still that fails here fails in motion too.

- [ ] Frame 0 is the subject's object or the title over the first picture, not an empty black.
- [ ] Every clause changes something; nothing is still for more than a second and a half.
- [ ] In the brief, every frame is filled edge to edge; no scene is a small picture on bare black.
- [ ] In the board, the diagram persists for the section and changes one piece per clause.
- [ ] Every picture moves: a push on every capture and clip, no still frame.
- [ ] Every scene shows three layers.
- [ ] In the brief, twelve or more cuts in 20 seconds; in the board, no cut inside a section.
- [ ] Every picture is the literal subject of the words over it.
- [ ] Every highlight covers exactly the words or rows the voice reads.
- [ ] Every diagram was built in the order the voice named its parts.
- [ ] Every headline card is real and names its source.
- [ ] No customer data, account name, third-party logo or meme is visible.
- [ ] Every sound effect is from the one chosen family and is credited.
- [ ] Every asset in the render was shown to the person and approved.
- [ ] The provenance lists every capture's origin and every sound's licence.
- [ ] The first and last frames read as the same world.

Then watch it once at full speed with the sound off. If the pictures tell
the story without the voice, the cut is right.

## Reference

The first video built in this style is sample 5 on
<https://daenadigital.com/resources/examples/videos>: a 20-second
how-it-works for the studio's own enquiry form. Its first cut was
rejected as "not very rich" (seven scenes, small pictures on bare black);
its second, in the brief mode, was fourteen full-bleed cuts. It is now a
board: one diagram follows the message from the form to a meeting — the
four fields fill their slots, the server box lands with the one line of
code that sends there, a third party is struck out, spam chips are
filtered, the message is saved beside its backup, and real clips play
inside the last boxes — and the camera pulls back on the last word to
show the whole system.

How the style was studied: on 9 October 2026, frames were sampled from
Fireship's "Docker in 100 Seconds" (every half second, 0–40 s) and a
2026 news video (every fifth of a second, 0–30 s) and laid out as
contact sheets. The explainer had two hard cuts in forty seconds: one
board — hardware, operating system, two virtual machines, then the
machines swapped for one engine with two apps — changed on every clause,
with a real clip set inside its boxes, after an opening where the
product's name became a literal object on a photograph. The news video
cut every 0.2–5 seconds: captures flying in tilted, a person cut out over
a capture, a big red word landing word by word, a glitch between
sections. Version 1.0 of this file knew only the second mode.

Knowledge base

What we studied to learn the style

Docker in 100 SecondsStudied frame by frame every 0.5 s, 0-40 s: two hard cuts in forty seconds; one board changes on every clause, a real clip set inside its boxes, the name turned into a literal object first
The most expensive software bug in history...Studied frame by frame every 0.2 s, 0-30 s: the brief mode; captures flying in tilted, a cutout over a capture, a big word landing word by word, a glitch between sections
open-video-edit: the fireship-style-edit and fireship-style-short skillsThe grammar as numbers: 2.2-second cuts long-form and about 1.5 on a short, footage that literally shows what each line says, headline stacks, highlighted docs, diagrams built up, graphics landing on the word at most 0.1 s early
How Fireship became YouTube's favorite programmer (Engineer's Codex)The 100-seconds format, the density, and why the channel never shows a face
Fireship case study (Faceless Directory)Fast-paced explainers built from screen recordings and motion graphics, cut in Premiere Pro
Remotion-Fireship: create viral videos with React code (BrightCoding)A programmatic reproduction of the look: code typed on with a cursor, dark code panels, type that lands on spring curves

Sources

SKILL.md

What we studied to learn the style

What this is based on