The prompt library for editing with AI

# Project: motion graphics in Remotion

This folder is a Remotion project. I am a video editor, not a developer. Explain changes in plain words.

## Setup facts
- Preview: `npm run dev` (Remotion Studio). I keep it open in another terminal.
- Render: `npx remotion render <CompositionId> out/<name>.mp4`
- Remotion Agent Skills are installed (`npx remotion skills add`). Use them for anything Remotion specific before guessing an API.
- Brand rules live in `brand/DESIGN.md`. Read it before you design anything.

## Video rules
- Default composition: 1920x1080, 30 fps. Vertical versions: 1080x1920, same fps.
- Never change fps or dimensions of an existing composition unless I ask.
- Express all timing in frames, and name it: `const IN = 12; const HOLD = 60;`

# Project: FFmpeg editing folder

I edit videos here with FFmpeg through you. Treat my footage like a client's footage.

## Folders
- `raw/` original files. READ ONLY. Never write, move, rename or delete anything in here.
- `work/` temporary files. You can create and delete files here.
- `out/` finished exports only.
- `logs/` one text file per job with the exact commands you ran.

## Rules
- Run `ffprobe` on every input before you build a command. Report resolution, fps, codec, duration and audio channels.
- Keep the source fps and resolution unless I ask for something else.
- Prefer stream copy (`-c copy`) for simple trims on keyframes. Re-encode when cuts must be frame accurate and say so.

# Project: editing through an MCP server

You control my editing app through an MCP server. The app is open on my screen and I can see every change.

## Safety first
- I always work on a duplicate project or a duplicated timeline. At the start of each session, ask me to confirm the name of the timeline or comp you may edit. Touch nothing else.
- Read before you write. Start with the read only tools (list projects, list timelines or comps, list clips, get markers) and summarize what you see.
- Never delete media, bins, timelines or comps. If something should go, move it to a bin or folder called `_to_delete` and tell me.
- If a tool can run raw scripts (ExtendScript, Python, JavaScript), show me the script and explain it in plain words before running it.
- Never start a render or export unless I ask for it in this session.

## How to work
1. Restate the task as a numbered plan with the exact tools you will call.
2. Do the first step only, then tell me what changed and where to look (timeline name, timecode, track).

---
name: cut-silences
description: Find pauses and dead air in a talking head video with ffmpeg silencedetect, propose a cut list, and export a tightened version. Use when the user asks to cut silences, tighten a take, or make a jump cut edit.
argument-hint: <video file> [min pause in seconds, default 0.6]
---

# Cut silences

Input: $ARGUMENTS

1. Run `ffprobe` on the file and report duration, fps and audio channels.
2. Detect silence on the first audio stream:
   `ffmpeg -hide_banner -i INPUT -map 0:a:0 -af silencedetect=noise=-35dB:d=MIN_PAUSE -f null - 2> work/silence.txt`
   If no audio stream exists, stop and tell the user.

---
name: captions
description: Transcribe a video locally with Whisper, build clean SRT captions with readable line lengths, and optionally burn them in. Use when the user asks for captions, subtitles, an SRT or burned in text.
argument-hint: <video file> [burn] [language]
---

# Captions

Input: $ARGUMENTS

1. Extract audio: `ffmpeg -i INPUT -vn -ac 1 -ar 16000 work/audio.wav`
2. Transcribe with word timestamps (ask which Whisper the user has installed if unsure):
   `whisper work/audio.wav --model small --word_timestamps True --output_format json --output_dir work`
3. Build captions from the word timings:

---
name: shorts
description: Find the strongest self contained moments in a long video from its transcript and export vertical 9:16 clips for review. Use when the user asks for shorts, reels, TikToks or clips from a long video.
argument-hint: <video file> [number of clips, default 5] [max seconds, default 45]
---

# Shorts from a long video

Input: $ARGUMENTS

1. Transcribe with word timestamps (see the captions skill) into `work/transcript.json`. Read only the transcript, do not try to watch the video.
2. Find candidate moments that:
   - make sense with no context from earlier in the video
   - open with a strong first sentence (a claim, a question, a number or a surprise) inside the first 2 seconds

---
name: qa-render
description: Run technical checks on a finished video file (specs, loudness, black frames, silence, duration) and report pass or fail. Use before delivering or uploading any export.
argument-hint: <video file> [target: youtube | broadcast | social]
disable-model-invocation: true
---

# QA a render

Input: $ARGUMENTS

Run each check, then print a table with check, result, pass or fail.

1. Specs with ffprobe: container, codec, resolution, fps, duration, audio codec, channels, sample rate. Compare with the delivery spec in CLAUDE.md if there is one.

# First motion graphic with Claude Code + Remotion

## Step 1. Set up (once, in a terminal)
You need Node.js and Claude Code installed. Then:

    npx create-video --yes --blank my-video
    cd my-video
    npx remotion skills add
    npm run dev

Leave that terminal open. Remotion Studio opens in your browser.
Optional: Remotion also ships an official Claude Code plugin, see remotion.dev/docs/ai/claude-code-plugin

## Step 2. Start Claude Code in a second terminal

# Claude Code + DaVinci Resolve: first session

## Step 1. Pick a server that fits your edition
- Resolve Studio: most MCP servers work, because external scripting is a Studio feature.
- Resolve Free: pick a server that says it works on Free. Since Resolve 21.1, Python scripting is Studio only, so many Free workarounds stopped working. Check each tool's notes on agentcut.dev/davinci-resolve-mcp/

## Step 2. Install the server
Follow that server's README exactly. It will give you a command to run the server.

## Step 3. Register it with Claude Code
Either run the command from the README, which usually looks like:

    claude mcp add resolve -- <the command from the README>

---
name: lower-third
description: Build or update a reusable lower third component in this Remotion project and render it with transparency for use in an NLE. Use when the user asks for a lower third, name title, name strap or chyron.
argument-hint: "Name" "Title" [seconds on screen, default 5]
---

# Lower third

Input: $ARGUMENTS

1. Read `brand/DESIGN.md` for fonts, colors, easing and safe margins. If it does not exist, ask the user for a font and one accent color.
2. If `src/components/LowerThird.tsx` exists, reuse it. Otherwise create it with props: name, title, accentColor, align (left or right), durationInFrames.
3. Motion: accent bar wipes in over 10 frames, name slides 24 px and fades in over 12 frames starting at frame 4, title follows 4 frames later. Hold. Exit is the reverse, 2 frames faster than the entrance. Ease out on entrances, ease in on exits.
4. Position: inside the graphics safe area, bottom left by default, never closer than 5% to any edge.

# Claude Motion: 30 second explainer

Claude Motion (beta, Team and Enterprise plans) builds animations as code from text, charts, shapes and images and exports MP4. It does not generate footage or people. Start it from Output > Motion in the message box, or from a Motion template in the Artifacts tab.

## Prompt (attach your script, data or slides first)

Make a 30 second animated explainer from the attached file.
Audience: [who watches, for example new customers or my YouTube viewers].
Where it plays: [YouTube video intro / all hands screen / Instagram feed].
Format: [16:9 1920x1080 or 9:16 1080x1920], 30 fps.
Structure:
1. 0 to 4 s: one line hook with the main number or claim from the file.
2. 4 to 22 s: three beats, one idea each, max 8 words on screen per beat.
3. 22 to 30 s: one line takeaway and a still frame I can hold under a voiceover.

On the timeline "[TIMELINE NAME]" only (it is a duplicate, the original is safe):

1. Find every pause longer than [0.6] seconds on the dialogue track [A1].
2. Do NOT cut yet. Add a red marker at the start of each pause, named "PAUSE <length>s".
3. Report: number of pauses, total time they add up to, and the 5 longest with timecodes.

After I review the markers and say go:
4. Remove each marked pause, keeping 4 frames of handle on both sides so words are not clipped.
5. Ripple the gaps closed on all tracks together so B roll and music stay in sync with dialogue. If a cut would break a clip on another track, skip it and mark it "SKIPPED: overlap".
6. Tell me the old and new duration and how many cuts you made.

Create captions for the timeline "[TIMELINE NAME]".

1. Use the app's own transcription if this tool supports it, otherwise transcribe the dialogue track locally and build an SRT.
2. Caption rules:
   - max [42] characters per line, max 2 lines (vertical video: 1 line, max 28)
   - reading speed under 20 characters per second
   - each caption on screen between 0.8 and 6 seconds
   - never split a name, a number and its unit, or an adjective and its noun across lines
3. Correct these names and terms exactly as written: [Name One, Brand Two, Product Three].
4. Import the captions as a caption or subtitle track, not as burned in graphics.
5. Apply the style: [font], [size], white text, [outline or box], bottom center, inside the title safe area.
6. List the 5 captions you are least sure about with timecodes so I can check them.

Source: [FILE OR TIMELINE]. Goal: [5] vertical clips, each [20 to 45] seconds.

1. Transcribe the source with word timestamps.
2. Find moments that work with no context, start with a strong first line in the first 2 seconds, and end on a finished thought.
3. Give me a table: start, end, length, first line, last line, one sentence on why it works. Give me [8] options so I can choose.
4. Wait for my picks.
5. For each pick: cut frame accurate, reframe to 1080x1920 keeping the speaker's face in the upper third, add one line captions kept out of the bottom 20% of the frame.
6. Name files `short_<number>_<first three words>.mp4` and give me a list with a suggested title for each.
Do not promise views. Just tell me why each clip might hold attention.

Set up renders for the timeline "[TIMELINE NAME]". Do not start rendering until I confirm.

Deliverables:
1. Master: [ProRes 422 HQ], source resolution and fps, stereo audio. Name: [PROJECT]_master_[DATE].mov
2. YouTube: H.264, 1920x1080, source fps, high bitrate. Name: [PROJECT]_youtube.mp4
3. Social square: 1080x1080, center framed. Name: [PROJECT]_1x1.mp4
4. Vertical: 1080x1920, use the "Vertical" timeline if it exists, otherwise ask me. Name: [PROJECT]_9x16.mp4
5. Audio only: WAV 48 kHz 24 bit. Name: [PROJECT]_mix.wav

Steps:
- Add each one to the render queue with the exact settings above and the output folder [PATH].
- Show me a table of the queued jobs with codec, resolution, fps and file name.
- After I say go, render, then report each file's size and duration and flag any that do not match the timeline duration.

In the project "[PROJECT NAME]" media pool or bin "[IMPORT BIN]":

1. Read the metadata of every clip: file name, date, camera or device, resolution, fps, duration, has audio.
2. Propose a bin structure first, for example:
   01_VIDEO / <shoot day> / <camera>
   02_AUDIO / <recorder>
   03_GRAPHICS
   04_MUSIC
   05_STILLS
   99_CHECK (anything you are not sure about)
3. Propose new clip names in the form [DAY]_[CAMERA]_[SCENE or subject]_[NUMBER], for example D1_A_Interview_003. Show me a sample of 10 before and after.
4. Rename the clip names inside the project only. Never rename, move or delete files on disk.
5. Flag clips that do not match the project fps or that have no audio when the others do.
6. Report how many clips went into each bin.

In the open After Effects project, create a new comp "LT_Name_Title", 1920x1080, [29.97] fps, 6 seconds.

Build:
1. A thin accent bar (shape layer, [accent color hex]) at the bottom left, inside the title safe area (at least 96 px from the left and bottom edges).
2. Text layer "NAME" in [font], 54 px, near black or white depending on [light or dark] background.
3. Text layer "Title or role" in the same font, 32 px, 70% opacity, 8 px under the name.
4. Animation in: bar scales X from 0 to 100% over 10 frames, name slides 24 px from the left and fades in over 12 frames starting at frame 4, title follows 4 frames later. Easy ease out on all entrance keyframes.
5. Hold until 1 second before the end, then animate out in reverse over 8 frames.
6. Add a Null or controller layer with Slider and Color controls for accent color, bar width and text offset, and link them with expressions.
7. Put the text layers in Essential Graphics so the name and title can be edited from Premiere.

Before you start, list the layers you will create. After you finish, render one frame at 1 second and at 3 seconds so I can check it.

Comp: "[COMP NAME]". Make it a template that someone who does not know After Effects can use.

1. List every text layer, color, image and timing value in the comp and tell me which ones you suggest exposing as controls.
2. After I confirm, add a "CONTROLS" null with Slider, Color, Checkbox and Point controls for those values, and connect them with expressions. Use clear names like "Accent color" and "Hold seconds".
3. Make text layers auto size their background boxes with sourceRectAtTime so long words do not break the design.
4. Add the chosen controls to the Essential Graphics panel in a logical order: text first, colors second, timing last.
5. Test with a very short text and a very long text and tell me if anything breaks.
6. Do not export the .mogrt. Tell me when it is ready and I will export it.

Project: "[PROJECT NAME]". New media location: [FOLDER PATH].

1. List every offline or missing clip with its old path, file name, duration and timecode.
2. Search the new folder for matches by file name first, then by duration plus start timecode if the names changed.
3. Show me a table: offline clip, proposed match, how you matched it (name, or duration plus timecode), confidence.
4. Relink only the high confidence matches after I confirm. Leave the rest offline and mark them in a bin "99_UNMATCHED".
5. Do not transcode, copy or move files on disk.
6. Report how many clips you relinked and how many are still offline.

File or timeline: [NAME]. Target: [YouTube about -14 LUFS integrated, true peak -1 dBTP] or [client spec].

1. Measure integrated loudness, loudness range and true peak, and report them.
2. Measure dialogue and music separately if they are on separate tracks.
3. Propose changes before applying them: dialogue level, music ducking under dialogue (how many dB), limiter settings.
4. After I confirm, apply them non destructively (track or clip gain, or a new export, never overwrite the source).
5. Measure again and show before and after numbers side by side.
6. List any spots where dialogue is still hard to hear, with timecodes.

Design an n8n workflow (I run [self hosted n8n with FFmpeg installed / n8n cloud]) that:

1. Triggers when a new video lands in [Google Drive folder / YouTube channel / S3 bucket].
2. Gets a transcript with word timestamps.
3. Picks [3] candidate clips of 20 to 45 seconds that work without context.
4. Cuts them with FFmpeg, reframes to 9:16 and burns one line captions.
5. Saves the clips to [folder] and sends me a message in [Slack or email] with the links and the first line of each clip.
6. Never posts anything publicly. A human approves every clip.

First give me the node list in order, what each node does, and which credentials I need. Point out which steps need self hosted n8n. Only after I approve, write the workflow JSON I can import.

Make a [15] second kinetic typography sequence, [1920x1080 / 1080x1920], [30] fps, from this script:
"[LINE 1] / [LINE 2] / [LINE 3] / [PAYOFF LINE]"

Look:
- One heavy grotesk typeface, all caps for key words, tight tracking (about -2%).
- Two colors only: [near black #111111] background, [off white #F5F3EE] text, plus one accent [#FF4D2E] for the single most important word per line.
- Cut on words, not on lines. 3 to 8 frames per word on fast beats, hold the payoff line at least 1.5 seconds.

Motion vocabulary (use a mix, never all at once):
- words snap in with a fast ease out (about 6 frames) and a 2 to 4 px overshoot at most
- masked reveals from below
- scale punch on the accent word (100% to 108% and back over 8 frames)
- hard cuts between lines, no cross dissolves

Design a lower third system for a corporate interview series.

Look:
- Flat and quiet. No gradients, no glow, no 3D.
- Sans serif, two weights only: name in Semibold, role in Regular at 60% of the name size.
- Optional thin accent line (2 to 4 px) in the brand color [#HEX], to the left of the text or under the name.
- Text sits on the footage with a soft shadow (0 2 px 12 px, 25% black) or on a solid box at 85% opacity if the footage is busy.
- Bottom left, inside the title safe area, aligned to the same left edge as any other graphics.

Motion:
- In: 12 to 15 frames, ease out, small move (16 to 24 px) plus fade. Accent line leads by 3 frames.
- Hold: at least 4 seconds, or long enough to read the role twice.
- Out: 8 to 10 frames, ease in, reverse of the in.

Burn in captions in the bold, word by word style that creators like Alex Hormozi made popular. Description of the look, not a copy of any template:

- 1 to 3 words on screen at a time, never more than one line.
- Very heavy sans serif (Black or ExtraBold weight), ALL CAPS, large: about 8 to 10% of frame height on 1080x1920.
- White text with a thick black outline (6 to 10 px at 1080 wide) and a soft drop shadow.
- The word being spoken is highlighted in one accent color [yellow #FFE600 or green #3BFF6A]. Other words stay white.
- Key words (numbers, money, strong verbs) get a quick scale pop: 100% to 115% and back over 6 frames.
- Position: center of frame or just below center, never in the bottom 20% where the app UI sits.
- Timing: each word appears exactly when it is spoken, about 1 to 2 frames after the word starts. Never early.
- No emojis unless I ask. No animation on filler words.

Source: word level timestamps from Whisper. Output: the captioned video plus the caption file in [ASS / SRT / CapCut text track].

Caption style in the high energy look common on big creator channels like MrBeast. Describe and build it, do not copy any channel's actual assets or fonts:

- 2 to 4 words per caption, one line, centered just below the middle of the frame.
- Chunky rounded sans serif, bold, mixed case or caps.
- White fill, thick dark outline, plus a hard offset shadow (4 to 6 px down and right).
- Rotate color for emphasis words through a small palette: [yellow, cyan, red], max one colored word per caption.
- Every new caption enters with a fast pop: scale 80% to 105% to 100% over 5 frames.
- Big numbers and money amounts get their own caption, larger (130%), with a short shake (2 px, 4 frames).
- Sound effect hits (whoosh, pop) can be marked as [SFX] in a separate list with timecodes, do not add audio yourself.
- Keep everything inside the safe area and away from where the face is.

Show me 3 sample frames (a normal line, an emphasis word, a big number) before rendering the full video.

On the timeline "[TIMELINE]" (a duplicate), build a film emulation look on a single clip first: "[CLIP NAME]". Footage is [camera and log format, for example Sony S-Log3 / S-Gamut3.Cine].

Node tree, each node labeled:
1. "INPUT CST": Color Space Transform from [camera color space and gamma] to DaVinci Wide Gamut / DaVinci Intermediate.
2. "BALANCE": neutral white balance and exposure. Do not push contrast here.
3. "CONTRAST": gentle S curve, lift the blacks very slightly so pure black never happens (film rarely sits at true zero).
4. "SEPARATION": warm the skin and highlights slightly, cool the shadows slightly. Subtle, max a few points.
5. "SAT": reduce saturation in the very brightest highlights (highlight rolloff), keep mid tone saturation.
6. "HALATION" (Studio only, or skip on Free): a faint reddish glow around bright edges.
7. "GRAIN" (Studio only, or skip on Free): fine 35 mm grain, low strength.
8. "OUTPUT CST": DaVinci Wide Gamut to Rec.709 Gamma 2.4 with tone mapping.

Rules: keep each node subtle, I will push it myself. Tell me which nodes need Studio. Save the result as a still and a PowerGrade named "AgentCut Film Base" and do not apply it to other clips until I say so.

Template (fill in each line, then join into one paragraph):

Subject: [who or what, with 2 or 3 concrete visual details]
Action: [one clear action, present tense]
Setting: [where, time of day, weather]
Style: [cinematic live action / documentary / 16 mm film look / clean commercial]
Camera: [shot size: wide, medium, close up] + [angle: eye level, low angle, top down] + [one move: slow dolly in, static locked off, handheld follow, slow orbit]
Lens and focus: [35 mm, shallow depth of field, background softly out of focus]
Lighting and mood: [soft window light, golden hour backlight, cold overcast]
Audio (if the model supports it): [ambient sound only, for example distant traffic and wind]

Example:
A barista in a green apron pours steamed milk into a white cup, forming a leaf pattern. Small busy cafe in the morning, rain on the window behind her. Cinematic live action, natural color. Close up at eye level, slow dolly in toward the cup. 50 mm lens, shallow depth of field, background softly blurred. Soft window light from the left, warm and calm. Ambient cafe sounds, no music.

The image already shows what things look like. The text should only say how things move.

Formula: [camera move]. [subject motion]. [scene motion]. [style word].

Start with only the camera move, generate, then add one element at a time.

Examples:
- Locked camera. The steam rises slowly from the cup. Soft light flickers across the table. Cinematic live action.
- The camera slowly pushes in. The subject turns her head toward the camera and smiles. Hair moves gently in the wind.
- Slow orbit around the product from left to right. Reflections slide across the surface. Clean commercial look.
- Handheld camera follows behind the subject as he walks down the hallway. Lights pass overhead.

Rules:
- Positive phrasing only. Write "Locked camera" or "The camera stays still", never "no camera movement".

[Subject doing one simple everyday action] in [real location, time of day].
Observational documentary style, handheld camera at shoulder height with small natural sway, slightly imperfect framing, the camera reframes once to follow the action.
35 mm lens, natural available light, no stylized lighting, muted realistic colors, light film grain.
The subject does not look at the camera.
Ambient location sound only.

Matching tips for the edit:
- Generate at the same aspect ratio and frame rate as your real footage when the model allows it.
- Grade AI shots with the same node tree or LUT as the real footage, then add the same grain.
- Use them short (2 to 4 seconds) between real shots, not as the hero shot.
- Label AI generated shots in your project and follow your client's and platform's disclosure rules.

Text to video:
A [product, for example matte black wireless earbuds case] on a [surface, for example polished dark stone pedestal], studio product shot. The camera slowly orbits 90 degrees around the product from left to right at the product's height. Soft key light from above left, thin rim light outlining the edges, dark gradient background. Clean commercial look, sharp focus on the product, very shallow depth of field. Slow and smooth, 5 seconds.

Image to video (from your own product photo):
Slow orbit around the product from left to right. Reflections slide across the surface. The background stays dark and still. Clean commercial look.

Variations, one at a time:
- "slow push in from wide to close up on the logo"
- "the camera rises from below the pedestal to eye level"
- "light sweep passes across the product from left to right"

Check every frame for warped logos or text. Generators often distort brand marks. Replace the label in post if needed.

Build a reusable VHS glitch transition, [12] frames long, that I can put on cuts.

Layers, in order:
1. RGB split: offset red left and blue right by up to 8 px at the peak frame, back to 0 at the ends.
2. Horizontal tracking lines: 2 or 3 thin bands that roll down the frame during the transition.
3. Slight horizontal jitter: random 2 to 6 px shifts on a few frames, not every frame.
4. Chroma bleed: soften the color channels slightly so edges smear.
5. Noise: light static that peaks with the RGB split.
6. Optional: a timestamp in the corner in a pixel font, "[PLAY ▶ 00:00:00]".

Curve: everything ramps up for 5 frames, peaks for 2, ramps down for 5.
Make strength a single 0 to 100 control so I can use it subtle (20) or heavy (80).
Keep it to transitions. Do not apply it to the whole video.

# DESIGN.md: Motion graphics brand system

Read this before creating any graphic, title, caption or transition. Fill in the brackets once, then never improvise outside these rules.

## 0. Dials
| Dial | Value | Meaning |
|---|---|---|
| Energy | [4] | 1 calm documentary, 5 clean YouTube, 10 hype trailer |
| Density | [3] | 1 one idea on screen, 10 dashboard |
| Motion | [5] | 1 cuts and fades only, 10 everything animates |

## 1. Canvas
- Masters: 1920x1080 (16:9) and 1080x1920 (9:16), [30] fps. Never mix frame rates in one project.
- Graphics safe area: keep text inside the central 90% of width and height (5% margin each side, following EBU R 95 graphics safe). Action safe: 93%.

# CAPTIONS.md: Caption and subtitle style system

Two modes. Pick one per video and say which in the job.

## Mode A: Subtitles (accessibility, long form, client work)
- Max 42 characters per line, max 2 lines.
- Reading speed: max 20 characters per second for adult viewers.
- Duration: min 0.83 s (20 frames at 24 fps), max 7 s.
- Gap between captions: at least 2 frames.
- Position: bottom center inside title safe. Move to top only when it would cover burned in text or a face.
- Style: [sans serif], white, 2 px black outline or a 60% black box, no animation.
- Line breaks: after punctuation, before conjunctions and prepositions. Never split a name, a number and its unit, or an article and its noun.
- Include [MUSIC] and speaker labels only when the speaker is off screen or unclear.
- Spelling: use the term list below exactly.

# THUMBNAILS.md: YouTube thumbnail system

## Canvas
- 1280x720 px, 16:9 (YouTube's recommended size). Export JPG or PNG.
- Design for the smallest place it shows: check it at 168x94 px. If you cannot read it there, it fails.
- Keep the bottom right 15% free of important content (the video length badge sits there).

## Layout (pick one per video, keep it for the series)
1. Face + 3 words: face on the right third, looking at the text or the viewer, text on the left two thirds.
2. Before / after: split down the middle, left "before", right "after", arrow or line between.
3. Object hero: one big object or screenshot, slightly rotated, text top left.
- One focal point. The eye should land in under a second.

## Text

# SHORTS.md: Social shorts template system

## Canvas
- 1080x1920, [30] fps, H.264, AAC 48 kHz.
- Length: [20 to 45] s. The first frame must already be interesting.
- Safe zone: keep text and faces out of the top 10%, the bottom 20% and the right 12% (platform UI). Main action lives in the middle 60%.

## Structure
| Part | Time | Rules |
|---|---|---|
| Hook | 0 to 2 s | Strongest line or visual first. Text hook on screen, max 6 words, matches what is said |
| Setup | 2 to 8 s | Why this matters, one sentence |
| Body | 8 s to end minus 4 s | One idea per 3 to 5 seconds. Cut or change framing at least every 3 s |
| Payoff | last 4 s | Answer the hook. End on a finished sentence |

# TITLES.md: Titles and lower thirds system

## Card types
| Type | Content | Position | Hold |
|---|---|---|---|
| Name strap | Name + role | Bottom left, safe area | 4 to 6 s, first appearance only |
| Location | Place, city | Top left | 3 s |
| Date or time jump | "Three months later" | Center | 2.5 s |
| Chapter | Number + title | Center or left column | 3 s |
| Source | "Source: [name]" | Bottom right, small | As long as the claim is on screen |

## Type
- Name: [typeface] Semibold, 44 px at 1080 tall.
- Role: same typeface Regular, 30 px, 75% opacity.

# EXPLAINER.md: Explainer and data animation system

## Principles
- One idea per scene. One chart per scene.
- Every number on screen comes from the source file. If a number is missing, show [placeholder], never invent it.
- Show the source on screen, small, bottom right, whenever a number appears.

## Scene timing
- Each scene: 4 to 8 seconds. Build 1 to 1.5 s, hold at least 2 s, exit 0.5 s.
- Numbers count up over 1 s with ease out, then hold the final value.
- Narration leads, animation follows: the visual appears on the word, not before.

## Charts
- Bar charts start at zero. Always.

# LOOK.md: Color and grade look system

## Pipeline
- Color managed: DaVinci Wide Gamut / DaVinci Intermediate timeline, output Rec.709 Gamma 2.4 for web and TV. [Change if your deliverable is HDR or P3]
- Input transforms per camera, set once per clip on the INPUT node or via project color management. Never grade log footage by eye without a transform.

## Node template (same order on every clip)
1. INPUT CST (if not project managed)
2. EXPOSURE and WB
3. CONTRAST
4. SAT and HUE
5. SECONDARIES (skin, sky): qualifier windows only here
6. LOOK (shared group node or PowerGrade, never edited per clip)
7. OUTPUT CST

# KINETIC.md: Kinetic typography system

## Rhythm
- Lock to music: [BPM]. At 30 fps, one beat = 1800 / BPM frames (120 BPM = 15 frames).
- Words hit on beats or half beats. Lines change on bars (every 4 beats).
- Hold the most important line for at least 2 bars.

## Type
- One typeface family. Two weights max: [Black] for key words, [Medium] for the rest.
- Size by importance, not by length: key word 2x to 3x the size of connecting words.
- Tracking: -2% to -4% on heavy weights, 0 on medium.
- Never more than 6 words on screen at once.

## Motion vocabulary (pick 3 for the whole piece)