All skills
simota avatar

/sketch

@16a3f28
by shingo imotasimota/agent-skills85 stars
15

Generating AI image-generation code using the Gemini API. Handles text-to-image generation, image editing, and prompt optimization. Use when image generation code is needed.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/sketch

This session only. Nothing lands on disk.

referencecinematic-prompting.md

≈4.1k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Cinematic Prompting

Purpose

The difference between a flat AI-generated image and one that feels deliberate is cinematographic literacy. Camera, lens, lighting, depth of field, and composition vocabulary lets the prompt direct the image like a director, not a tourist with autoflash. This reference catalogs the vocabulary, the prompt patterns, and the verification criteria.

Scope Boundary

  • IN scope: shot-type vocabulary, camera / lens specification, lighting setups, depth of field, film stock and color science, composition rules, prompt-construction patterns, intent verification.
  • OUT of scope: text-to-image API integration (generate), seed-locked variation (batch), style-transfer / character consistency (style), upscaling / inpainting (upscale), provenance / disclosure (provenance), content policy (policy).

Core Concepts

Why Cinematic Vocabulary Helps

Modern image models (Gemini 3.1 Flash Image, Gemini 3 Pro Image, FLUX, SDXL Turbo) have ingested decades of cinema and photography metadata. Calling a "Rembrandt-lit close-up shot on Kodak Portra 400 with shallow depth of field" gives the model a specific historical reference vector vastly stronger than "a portrait that looks nice".

Generic prompt Cinematic prompt
"A nice portrait of a woman" "Medium close-up portrait, 85mm lens at f/1.8, Rembrandt lighting from camera-left, shallow depth of field, shot on Kodak Portra 400, golden hour"
"A landscape photo" "Wide-angle landscape, 24mm lens at f/11, deep focus, Ansel Adams Zone System, low-angle morning light, large-format Velvia 50 emulation"
"An action shot" "Mid-action tracking shot, 35mm anamorphic, slight motion blur, hard side-light, low-angle, color-graded Day-for-Night"

Shot Types

Shot Frame Use
Extreme wide (EWS) Subject tiny in vast environment Establish setting; isolation
Wide (WS / LS) Full body + environment Context
Medium wide (MWS) Knees up Action with context
Medium (MS) Waist up Conversation default
Medium close-up (MCU) Chest up Interview
Close-up (CU) Shoulders up Emotion
Big close-up (BCU) Face fills frame Intensity
Extreme close-up (ECU / XCU) Eyes / detail Drama / detail emphasis
Macro Subject magnified Texture / object

Combine with angle: low-angle (heroic), high-angle (vulnerable), Dutch tilt (unease), bird's-eye (observational), POV (immersion), worm's-eye, OTS (over-the-shoulder).

Camera Format

Format Vocabulary Effect
35mm full-frame "35mm full-frame" Modern photographic default
Medium format "Hasselblad medium format" / "Pentax 645" Higher resolution, characteristic 6×6 / 6×7 ratio
Large format "4x5 large format" / "8x10 view camera" Sharp, Ansel Adams aesthetic
Super 35 "Super 35 cinema" Common cinema sensor
Anamorphic "anamorphic 2.39:1" / "Panavision anamorphic" Wide cinematic, oval bokeh, blue lens flares
16mm "16mm film grain" Indie / documentary feel
8mm / Super 8 "Super 8 home-movie aesthetic" Vintage / nostalgic
iPhone / smartphone "shot on iPhone 15 Pro" Modern casual realism
Polaroid "Polaroid 600 instant film" Square, soft, vintage
Drone / aerial "drone shot, DJI Mavic 3" High-angle scale

Lens Vocabulary

Lens Field of view Use
Ultra-wide 14mm Very wide; distortion Architecture, dramatic perspective
Wide 24mm Wide environment Landscape, environmental portrait
Wide-normal 35mm Slight wide Documentary, street
Normal 50mm Eye-mimic Reportage, casual portrait
Short tele 85mm Compressed Portrait classic
Tele 100-135mm Compressed bg Beauty / fashion portrait
Long tele 200-400mm Heavy compression Wildlife, sports, voyeuristic
Macro 100mm 1:1 magnification Detail / product
Tilt-shift Selective focus plane Architectural correction; miniature effect
Fisheye Strong distortion Skate / surreal

Aperture controls depth of field:

f-stop DOF Effect
f/1.4 - f/2 Razor thin Subject isolated; bokeh dominant
f/2.8 - f/4 Shallow Portrait standard; some bg context
f/5.6 - f/8 Medium Both subject and bg readable
f/11 - f/16 Deep Landscape; everything sharp
f/22+ Maximal Diffraction softness; Zone System

Lighting Vocabulary

Setup Description Mood
Rembrandt Triangle of light on shadow-side cheek Classical portrait
Butterfly / Paramount Light from above-front; nose shadow under nose Beauty / glamour
Loop Slight side; small shadow off nose Standard portrait
Split Half face lit, half shadow Drama / noir
Broad Light hits the side closer to camera Wider face
Short Light hits the side further from camera Slimmer face
Hard light Direct sun, point source High contrast, sharp shadows
Soft light Diffused, large source Wrap-around, gentle shadows
Backlight / rim Behind subject Halo, separation
Silhouette Subject dark against bright bg Iconography
Chiaroscuro High contrast light/shadow Dramatic, Caravaggio
Practical In-frame light source (lamp, neon) Naturalistic
Motivated Off-frame source consistent with scene Realistic
High-key Bright, low-contrast Commercial / fashion
Low-key Dark, high-contrast Noir / horror
Three-point Key + fill + backlight Standard professional
Available / natural No artificial Documentary
Golden hour First / last hour of sun Warm, long shadows
Blue hour Twilight before sunrise / after sunset Ethereal, cyan-magenta
Overcast Diffuse cloud cover Even, soft
Magic hour Specific minutes of sunset Cinematic mood

Color Science / Film Stock

Stock / Look Aesthetic
Kodak Portra 400 Warm skin tones, slightly muted
Kodak Ektar 100 High saturation, fine grain
Fuji Velvia 50 Saturated greens / blues; landscape
Fuji Pro 400H Cool, pastel, fashion
Cinestill 800T Tungsten-balanced; halation around lights
Ilford HP5 / Tri-X B&W, grainy, photojournalism
Polaroid 600 Square, soft, vintage
Lomography Cross-processed, vignetted
Teal-and-orange Hollywood blockbuster look
Bleach bypass Desaturated, silver retention
Day-for-night Blue-graded daytime to imply night
Technicolor Saturated, 3-strip era
Cool LUT / Warm LUT Generic shifts

Composition Rules

Rule Description Use
Rule of thirds Subject on third-line intersection Default
Centered / symmetrical Subject dead-center Portrait, religious, Wes Anderson
Leading lines Lines guide eye to subject Landscape, architecture
Framing Foreground frames subject Depth, layered
Negative space Empty area emphasizes subject Minimalist
Diagonal Strong diagonal element Energy, tension
Triangle 3-point composition Stability
Repetition / pattern Repeating elements Graphic, design
Foreground-Mid-Background (FMB) 3-layer depth Cinematic
Golden ratio / spiral Fibonacci composition Classic art
Headroom / nose room Space above head, in front of face Portrait
Reflection / mirror Subject + reflection Doubled meaning
Rule of odds 3 / 5 subjects (odd) Visual interest

Style References

Adding director / photographer references invokes their entire visual signature:

Reference What it conveys
Roger Deakins cinematography Subtle realism, controlled silhouettes
Wes Anderson aesthetic Centered, pastel, symmetrical
Stanley Kubrick one-point perspective Dread, vanishing point
Tarkovsky long take Slow, melancholic
Annie Leibovitz portrait Theatrical lighting, environmental
Henri Cartier-Bresson Decisive moment, B&W reportage
Sebastião Salgado Heroic B&W landscape, social documentary
Saul Leiter Soft color, layered street
Daido Moriyama Grainy, high-contrast Tokyo
Yasujirō Ozu tatami shot Low-angle Japanese domestic
Hopper painting Lonely Americana, hard light
Rembrandt painting Dramatic chiaroscuro
Studio Ghibli backgrounds Detailed naturalistic anime
Moebius / Jean Giraud Sci-fi line art, surreal

Use one or two; stacking five name-references muddies the signal.

Prompt Construction Pattern

Effective cinematic prompts follow Subject + Style + Composition + Technical:

[Subject + action + emotion],
[Shot type] [Camera angle],
[Camera / lens / aperture],
[Lighting],
[Film stock / color grade],
[Composition rule + framing detail],
[Style reference if any],
[Aspect ratio / quality]

Example:

Tired barista mid-yawn behind counter, Medium close-up, slight low angle, 35mm full-frame at f/2 with shallow depth of field, Window light from camera-right, warm tungsten practical, Cinestill 800T color grade with halation, Rule-of-thirds with negative space at right, Roger Deakins style, 16:9 cinematic widescreen.

This 60-word prompt outperforms a vague 5-word ask by an order of magnitude.

Verification Criteria

After generation, check intent match:

Criterion Check
Shot type matches EWS prompt produced EWS, not CU
Lens / aperture honored Shallow DoF if requested f/1.8
Lighting setup recognizable Rembrandt triangle visible
Color grade / film stock Specific look present
Composition rule honored Rule of thirds / centered as specified
Style reference recognizable Deakins / Wes Anderson character preserved
Aspect ratio correct 16:9, 2.39:1, 1:1, 4:5 etc.

If 3+ criteria fail, restructure the prompt rather than re-roll seeds.

Common Failure Patterns

Failure Cause Fix
Generic "professional photo" feel Too vague Add lens / lighting / film
Bokeh requested but not present Aperture not specified Add f/1.4 or f/1.8
Wrong camera angle Angle not specified Add "low-angle" / "high-angle"
Style reference ignored Too many references stacked Cut to 1-2
Wrong era / aesthetic Modern keyword conflicts Match film stock to era
Composition broken No composition rule Add "rule of thirds" / "centered"
Color cast off No film stock / LUT Specify Portra / Ektar / etc.
All faces look the same No facial-specific prompt Add age / ethnicity / expression

Genre Templates

Pre-built Subject + Style + Composition combinations:

Genre Template
Documentary portrait "MS, 50mm at f/2.8, available natural light, slight reportage feel, Kodak Tri-X B&W, rule of thirds, Henri Cartier-Bresson tradition"
Editorial fashion "MCU, 85mm at f/1.4, butterfly lighting + softbox fill, Fuji Pro 400H pastel grade, centered, Annie Leibovitz"
Landscape "EWS, 24mm at f/11, deep focus, golden hour, Velvia 50 saturated, leading lines, Ansel Adams Zone System"
Sci-fi atmosphere "WS, 35mm anamorphic 2.39:1, motivated practical lighting, teal-and-orange grade, FMB depth, Roger Deakins / Blade Runner 2049"
Indie film still "MS, Super 35 at f/2, soft window light, slight 16mm grain, muted natural color, off-center composition, Sofia Coppola feel"
Beauty / cosmetics "BCU, 100mm macro at f/4, butterfly + ring light, high-key fashion grade, centered symmetrical, glossy product feel"
Sports action "MS tracking, 200mm at f/2.8, hard side-light, slight motion blur, saturated Velvia, diagonal composition"
Architectural "Tilt-shift, 50mm equivalent at f/11, overcast diffuse, neutral grade, rule of thirds with leading lines, Iwan Baan style"

Use as starting templates; tune Subject and 1-2 details.

Workflow

  1. Define Subject — what + action + emotion in one phrase.
  2. Choose shot type — EWS to ECU based on intent.
  3. Pick camera + lens + aperture — match to mood (shallow vs deep DoF).
  4. Design lighting — Rembrandt / butterfly / hard / soft / golden hour / blue hour.
  5. Select film stock or color grade — era + mood + saturation.
  6. Apply composition rule — thirds / centered / leading lines / negative space.
  7. Add 1-2 style references if needed (don't stack 5).
  8. Specify aspect ratio + quality.
  9. Construct prompt in Subject → Style → Composition → Technical order.
  10. Generate + verify against the 7-criteria checklist.
  11. Iterate prompt (not just seeds) when 3+ criteria fail.

Output Template

cinematic_prompt:
  subject: "Tired barista mid-yawn behind counter"
  shot_type: medium_close_up
  camera_angle: slight_low
  camera: 35mm_full_frame
  lens: 35mm
  aperture: f/2
  depth_of_field: shallow
  lighting:
    key: window_light_camera_right
    practical: warm_tungsten_in_frame
    setup: motivated_natural
  color_grade: cinestill_800t_with_halation
  composition:
    rule: rule_of_thirds
    negative_space_at: right
  style_reference: roger_deakins
  aspect_ratio: 16:9
  prompt_text: |
    Tired barista mid-yawn behind counter,
    medium close-up at slight low angle,
    35mm full-frame at f/2 with shallow depth of field,
    window light from camera-right with warm tungsten practical,
    Cinestill 800T color grade with halation,
    rule-of-thirds composition with negative space right,
    Roger Deakins cinematography style,
    16:9 cinematic widescreen.
  verification:
    shot_type_match: PASS
    aperture_honored: PASS
    lighting_recognizable: PASS
    color_grade: PASS
    composition: PASS
    style_reference: PASS
    aspect_ratio: PASS

Anti-Patterns

  • Generic "professional photo" with no cinematic vocabulary — flat output.
  • Stacking 5 style references — model loses signal.
  • Conflicting era cues (Cinestill 800T + 1920s sepia) — incoherent.
  • Aperture requested but DoF wrong — model didn't honor; restructure prompt.
  • "Bokeh" without aperture spec — vague.
  • "Beautiful" / "stunning" / "amazing" — meaningless to the model.
  • "Hyperdetailed, 8K, masterpiece" filler — 2022-era prompts; modern models don't need this.
  • Ignoring composition rule — center-default everywhere.
  • Mismatched lens to subject (200mm tele for landscape EWS) — wrong tool.
  • Lighting setup unspecified for portrait — uncontrolled.
  • Color grade unspecified — model defaults to bland.
  • Aspect ratio missing — square default may not match intent.
  • Re-rolling seeds when prompt is the problem — wastes API calls.
  • Director name without directorial signature in mind — name-dropping noise.
  • Mixing photo + painting + 3D references — pick one medium.

Deliverable Contract

A cinematic prompt design is complete when:

  • Subject articulated in one phrase.
  • Shot type + angle specified.
  • Camera + lens + aperture + DoF specified.
  • Lighting setup named.
  • Film stock / color grade named.
  • Composition rule applied.
  • Style reference (≤ 2) chosen if needed.
  • Aspect ratio specified.
  • Prompt text follows Subject → Style → Composition → Technical order.
  • Verification criteria checked post-generation.

References

  • David Bordwell & Kristin Thompson, Film Art: An Introduction (12th ed., 2019).
  • Roger Deakins, Byways (2021) photo book — visual reference.
  • Vittorio Storaro, Writing with Light — color theory in cinema.
  • Ansel Adams, The Negative / The Print — Zone System.
  • Henri Cartier-Bresson, The Decisive Moment (1952).
  • Annie Leibovitz, At Work (2008).
  • Wes Anderson, Accidentally Wes Anderson (community visual reference).
  • Ed Lachman, Christopher Doyle, Hoyte van Hoytema — cinematography interviews.
  • American Cinematographer magazine — current cinematography techniques.
  • Cinestill, Kodak, Fuji film stock documentation — accurate color science.
  • Gemini 3 image / FLUX / SDXL prompt-engineering community guides (2024–2026).
  • Ari Folman, "How to Direct an AI" — practical cinematic-prompt tutorials.
  • Karen Cheng (visual essays) — composition / framing breakdowns.

Source: SKILL.md on GitHub

2 warnings4mo5 checks · Risk SAFE
  • Gen Agent Trust Hub4mo

    This skill provides a secure and well-architected framework for generating Python code to interact with the Google Gemini image generation API. It adheres to security best practices by emphasizing environment-variable-based credential management, providing explicit .gitignore guidance, and incorporating comprehensive error handling for API and safety filter responses. The skill acts exclusively as a code generator, ensuring that no actual API calls or external network requests are executed within the agent context itself.

  • Socket4mo

    No alerts

  • Snyk4mo

    Risk: MEDIUM · 1 issue

  • Runlayer6mo

    2/4 files flagged

  • ZeroLeaks5mo

    1 finding · Score: 86/100

Signed by skilld at 16a3f28. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated last month

README badge

README badge for simota/agent-skills/sketch