CinePrompt Skill

AI video prompt builder for cinematographers. Translates natural language shot descriptions into structured prompts optimized for AI video generators.

What It Does

CinePrompt turns vague ideas ("cinematic sunset over mountains") into precise cinematography prompts with lens, movement, lighting, color science, sound design, and 133 total fields. Three workflow modes, 24 generation models, 8 model optimizers.

Live: https://cineprompt.io Guides: https://cineprompt.io/guides (18+ articles, daily additions) Models: https://cineprompt.io/models

CLI Usage

# Install
npm install -g cineprompt

# Auth (Pro subscription required for API)
cineprompt auth cp_your_key_here

# Build a share link from state JSON
cineprompt build '{"mode":"single","subjectType":"character","fields":{...}}'

# Build from file
cineprompt build --file shot.json

# Pipe JSON
cat shot.json | cineprompt build

# List all 133 fields
cineprompt fields

# Show values for a specific field
cineprompt fields mood
cineprompt fields movement_type

Building State JSON

The agent constructs a state object and passes it to the CLI. The CLI creates a share link on cineprompt.io where the user can view, tweak, and copy the prompt.

State Structure

{
  "mode": "single",
  "complexity": "complex",
  "subjectType": "character",
  "fields": {
    "media_type": ["cinematic"],
    "mood": ["contemplative"],
    "genre": ["drama"],
    "setting": "interior",
    "location_type": ["apartment"],
    "custom_location": "A cluttered one-bedroom with peeling wallpaper",
    "env_time": "night",
    "char_label": "A retired boxer",
    "subject_description": "Weathered face, broken nose, calloused hands",
    "expression": "quietly resigned",
    "wardrobe": "Stained white undershirt, suspenders hanging at sides",
    "action_primary": "sitting alone at a kitchen table",
    "shot_type": "medium close-up",
    "framing": ["positioned left-third of frame"],
    "focal_length": "85mm",
    "dof": "shallow depth of field, bokeh",
    "movement_type": ["static, locked-off"],
    "lighting_type": ["practical lights"],
    "key_light": "Single bare bulb overhead, slightly swinging",
    "film_stock": ["Kodak Vision3 500T 5219"],
    "color_grade": ["desaturated"],
    "sfx_environment": ["room tone"],
    "ambient": "Refrigerator hum, distant sirens",
    "props": "Half-empty whiskey bottle, old photograph face-down"
  }
}

Key Parameters

Field Types

Button fields (93) — accept arrays of predefined values. Use cineprompt fields \x3Cname> to see valid options.

"media_type": ["cinematic"],
"mood": ["nostalgic", "contemplative"],
"shot_type": "extreme close-up"

Text fields (40) — accept free-form strings.

"char_label": "A young street musician",
"subject_description": "Dark curly hair, paint-stained fingers",
"dialogue": "I never said goodbye",
"ambient": "Rain on a tin roof, distant thunder"

Modes

Single Shot — one shot, full cinematography control.

Multi-Shot — sequence of shots with global settings + per-shot overrides. Supports recurring characters, transitions between shots.

{
  "mode": "multi_shot",
  "complexity": "complex",
  "fields": {
    "media_type": ["cinematic"],
    "mood": ["tense"]
  },
  "shots": [
    {
      "subjectType": "character",
      "fields": {
        "shot_type": "establishing shot",
        "char_label": "Detective",
        "action_primary": "approaching the building"
      }
    },
    {
      "subjectType": "character",
      "fields": {
        "shot_type": "close-up",
        "char_label": "Detective",
        "expression": "steeling herself",
        "action_primary": "reaching for the door handle"
      }
    }
  ]
}

Frame → Motion — dual-prompt output for img2vid workflows. Build the frame (image prompt), then direct the motion (video prompt). The FM tab uses direct-edit motion text with quick-insert chips:

• Camera chips: Slow push in, Slow pull out, Orbit, Dolly, Crane up/down, Handheld, Tracking, Locked off
• Pacing chips: Slow motion, Real-time, Time-lapse, Hyperlapse
• Transition chips: Whip pan, Steadicam, Rack focus, Reveal, Morph, Dissolve
• Direction chips: Slow Build, One at a Time, Breathe, Anchor, Physics, Chaos, Match

Key Fields Reference

Core (always relevant)

media_type, mood, genre, setting, location_type, custom_location, env_time, weather

Subject — Character

char_label, subject_description, expression, body_language, age_range, build, hair_style, hair_color, skin_tones, wardrobe, action_primary, props

Subject — Creature

creature_label, creature_description, creature_category, creature_size, creature_body, creature_skin, creature_expression

Subject — Object

obj_description, obj_material, obj_condition, obj_scale

Subject — Vehicle

veh_description, veh_type, veh_era, veh_condition

Subject — Landscape

land_scale, land_season

Subject — Abstract

abs_description, abs_quality, abs_movement, abstract_environment

Camera & Lens

shot_type, framing, focal_length, camera_body, lens_brand, lens_filter, dof, movement_type, pacing

Lighting

lighting_type, lighting_style, key_light, fill_light

Color & Look

color_grade, color_science, film_stock, palette_colors

Environment

location, env_bg, env_mg, env_fg

Sound

dialogue, dialogue_character, dialogue_language, delivery_style, delivery_style_custom, voiceover_text, music, music_genre, music_mood, ambient, sfx_environment, sfx_interior, sfx_dramatic, sfx_mechanical, beat_1, beat_2, beat_3

Style

animation_style, documentary_style, commercial_type, music_video_style, social_media_style, format

Scene-to-Prompt

CinePrompt also accepts natural language descriptions via the Scene-to-Prompt feature. Users type a shot description and an LLM auto-populates all fields. The agent can use this as an alternative to manually constructing state JSON — just direct users to the text box at the top of the page.

Generate

CinePrompt includes built-in generation with BYOK (bring your own key) across 24 models:

• Text-to-video (9): Kling O3 Pro, Sora 2 Pro, Veo 3.1, WAN 2.6, Seedance 1.5 Pro, LTX 2.3, Grok Imagine (Fal + Venice)
• Image-to-video (5): Kling O3 Pro, Sora 2 Pro, Veo 3.1, WAN 2.6, LTX 2.3, Grok Imagine
• Reference-to-video (1): Kling O3 Pro R2V (character elements)
• Image gen (4): Nano Banana Pro, NB2, Chroma, Grok Imagine
• Image edit (4): NBP Edit, NB2 Edit, Grok Imagine Edit, Qwen Edit
• Providers: Fal.ai + Venice.ai

Subject Library

Persistent character/element system. Users save subjects with frontal + reference images and field state. Subjects auto-inject into prompts and enable R2V (reference-to-video) generation with Kling Elements.

Internal Scripts

cineprompt-x-post

Daily cron (8:55 AM) that reads today's guide article, finds a trending AI video post on X, and writes a quote-tweet mini-essay for the CinePrompt Discord channel. Output goes to Discord for Tylios to post.

create-share-link.js

Creates share links via Supabase RPC (with API key) or direct insert (with service key). Used internally by the CLI and agent.

Tiers