agentleFS
raphaelbgrraphaelbgr/ai-video-prompting/skills/prompt-expander/SKILL.md

prompt-expander

Skill0 starsChanged 2 months agoMITView on GitHub

Use when someone hands you a rough video idea, logline or one-liner and wants a full cinematic prompt, when writing or polishing a Gemini Omni Flash / Google Flow prompt, or when planning a conversational edit of an existing clip. Turns "a lonely astronaut finds a flower" into a complete, generation-ready prompt by deciding every dimension a clip can express.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the reference for any axis you are unsure how to fill.

    line 38
  2. Distinctive:

    Use the defaults table for anything unspecified.

    line 50
  3. Distinctive:

    Add a timestamp breakdown only if there are genuinely 2–3 beats.

    line 56
  4. Distinctive:

    State the defaults you used in the Levers list: Never silently bury a consequential choice.

    line 77
  5. Distinctive:

    Default to assuming and proceeding, noting it in Levers.

    line 82
  6. Distinctive:

    Use the surgical rule from the audio reference §B1:

    line 91

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Prompt expander

Turn a rough idea into a complete, generation-ready cinematic prompt — by deliberately deciding every dimension a clip can express, then writing only the relevant choices as natural prose.

Written against Gemini Omni Flash in Google Flow (~10s clips with synced audio; accepts text + reference image/audio/video; supports multi-turn conversational editing). The dimension model transfers to any video generator; the mechanics in §B of the audio reference are Omni-specific.

When to use: someone hands you a logline and wants it "expanded" · you are writing or polishing a Flow prompt · you want to edit an existing clip (use the Editing path, not full expansion).


The dimensions — think through ALL of them

For every prompt, consciously decide each axis, then keep only what serves the idea:

#DimensionReference
1Subject — who/what, 5–8 specific descriptorssubject-action-continuity.md
2Action / physics — one beat, strong verbs, secondary motionsame
3Setting — location, era, weather, depth layerssame
4Camera — shot size, angle, movement, lensvisual-language.md
5Light & colour — setup, time of day, palette, film stocksame
6Style & mood — visual style, tone word, atmospheresame
7Audio — dialogue / SFX / ambient / music, or explicit silenceaudio-and-mechanics.md
8Beat — the single emotional moment the clip deliverssubject-action-continuity.md

Read the reference for any axis you are unsure how to fill. Do not paste the menus into the prompt — choose a few options and write prose.


Workflow

  1. Parse the idea: pull out subject, action, setting, and any stated mood, genre or audio.
  2. Classify: a fresh generation, or an edit of an existing clip? An edit goes to the Editing path below.
  3. Fill the gaps: for each of the 8 dimensions, choose options that fit the mood. Use the defaults table for anything unspecified.
  4. Respect the constraints (audio reference §B5): one beat per ~10s · always specify audio · phrase exclusions positively · do not rely on rendered on-screen text · keep references to a few.
  5. Write the prompt as flowing prose in this order: camera → subject → action → setting → light/style → audio. Add a timestamp breakdown only if there are genuinely 2–3 beats.
  6. Lint the result — veo-prompt-lint — and fix any ERRORs. If that is unavailable, self-check against the Common mistakes list below.
  7. Present three things: the prompt block, a short Levers list of the assumptions you made so they can be tweaked, and 1–2 suggested conversational edits for iterating.

Sensible defaults (when the user is silent)

AxisDefault
Stylephotorealistic, cinematic
Cameraslow push-in, eye-level, 50mm, shallow depth of field
Lightmotivated natural light matched to the setting's time of day
Motion24fps real-time, natural real-world physics
Audiomatching ambient bed + one tied SFX; no dialogue unless implied
Aspect16:9 (9:16 if "shorts / vertical / phone" is implied)
Beatone clear reveal or reaction

State the defaults you used in the Levers list. Never silently bury a consequential choice.

Ask vs. assume

Default to assuming and proceeding, noting it in Levers. Ask only when a choice is both consequential and genuinely ambiguous — almost always one of: realistic vs. animated/stylised, dialogue vs. silent, vertical vs. widescreen. Bundle those questions; do not drip them one at a time.


Editing path (an existing clip)

Use the surgical rule from the audio reference §B1:

  • One change per turn. Isolate a single dimension; verify before the next edit.
  • Freeze-list: name explicitly what stays. Change <X>. Keep the camera, subject, location, timing, and audio identical.
  • Batch changes as a numbered list, ending with the same freeze line.
  • Remember: editing the spoken audio of a clip is currently restricted.

Worked example

Idea: "a lonely astronaut finds a flower on a dead planet"

Expanded prompt:

Slow push-in, low-angle medium shot on a weathered astronaut in a scuffed white EVA suit, cracked gold visor reflecting a dim red sun, kneeling on cracked ochre desert rock. He slowly reaches a trembling gloved hand toward a single impossible blue flower pushing up through the dust; grit drifts in the low gravity. Vast empty rust-coloured plain stretches to a hazy horizon behind him, distant dust devils. Bleak desaturated palette with one vivid blue accent, harsh low sunlight, 35mm film grain, melancholic and awestruck. Ambient: faint wind and the hiss of suit air. SFX: soft servo whir as his hand moves. No dialogue, no music.

Levers: photorealistic (could go painterly) · 16:9 (could be 9:16) · silent (could add a breath or VO) · single push-in (could add a [00:06-00:10] insert close-up on the flower).

Suggested edit: Change the flower from blue to glowing bioluminescent teal. Keep the camera, subject, location, timing, and audio identical.


Common mistakes

  • Pasting whole menu tables into the prompt — pick a few options, write prose.
  • Forgetting audio → the model invents unwanted speech.
  • Cramming a multi-scene story into 10 seconds — one beat.
  • "no people / no text" phrasing → flip it positive ("a deserted street", "a clear sky").
  • "locked tripod" → use "static shot".

The reference menus

FileCovers
visual-language.mdShot sizes, angles, camera movements, lenses and optics, composition, motion feel, lighting setups, time of day, colour temperature, palettes and grading, film stock, visual styles, mood, texture
subject-action-continuity.mdSubject descriptor slots (people, creatures, objects), action verbs and pacing, physics cues, setting layers, continuity and multi-shot sequencing, emotional beats
audio-and-mechanics.mdDialogue, SFX, ambient, music, sync, the silence gotcha — plus Omni Flash mechanics: conversational editing, reference inputs, exclusions, timestamps, constraints

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.