The AI Realism Playbook

01. The core principle: direction over luck

Ninety-five percent of AI images look like generic noise — not because the tools are weak, but because the decisions behind them are random. A prompt is not a wish. It's a technical brief handed to a photographer and a lighting designer at the same time.

Random prompt → random image. Directed prompt → repeatable, on-brand result.

A user writes "cool car" and hopes for luck. A director specifies time of day, lens, lighting direction, material, and mood — and gets a predictable result. Consistency matters more than any single "perfect" output. Style is a repeatable system of decisions.

The core formula

SHOT + LENS + LIGHT + TEXTURE + COMPOSITION + STYLE REFERENCE

The 3-pillar system

Layer every prompt in this order: subject → camera → light → texture → composition → style. Each layer adds precision. Never write a prompt as one flat sentence.

02. Prompt anatomy

Subject & action

The skeleton of the frame — who's in it, what they're doing, and where. Avoid abstraction.

Camera & shot type

• Extreme close-up (ECU) — eye, lips, ring. Intimacy, tension.
• Close-up (CU) — face/chest. Emotion, drama.
• Medium shot (MS) — waist-up. Balance of emotion & context.
• Full shot (FS) — full body. Figure + environment.
• Over-the-shoulder (OTS) — from behind subject. Presence, cinematic.
• POV — first-person. Full immersion.
• Dutch angle — tilted frame. Energy, instability.

Lens & focal length

Lighting

Source — natural light (alive, realistic, soft); studio strobes/softbox (clean, commercial); neon/LED (futurism, cyberpunk); candlelight (intimacy, warmth); practical in-frame lights (lived-in realism).

Direction — front-lit (flat), side-lit / Rembrandt (drama, depth), backlit (silhouette, halo), rim-light (edge separation).

Color temperature — warm 2700–3500K (cozy, nostalgic); daylight 5000–5600K (neutral, realistic); cool 6000–7000K (techy, cold).

Material & texture — the plastic-AI killer

The single biggest realism unlock: name the material. Skin should have visible pores, micro-wrinkles, and natural tone irregularities. Metal can be matte, brushed, polished, or scratched. Fabric — silk (glossy), denim (rough), wool (warm), leather (patina). Ambiguous materials give you smooth, plastic outputs. Specificity gives you a photograph.

Composition

Rule of thirds (balanced, natural), centered (formal, iconic), negative space (air, drama, room for text), leading lines (guides the eye), foreground/mid/background layering (depth).

Style references — anchor every image

Photographers (Annie Leibovitz — dramatic portrait; Bruce Gilden — harsh street; Helmut Newton — fashion). Films (Blade Runner 2049 — neon + fog; Dune — sandy monumentality; Drive — night & neon). Eras (1970s ads, 1990s grunge, 1920s art deco).

A full layered prompt

Motocross rider mid-lean on a dirt track, captured in gritty 90s analog style,
vertical black and white frame with strong motion blur, shot handheld with a
35mm lens at f/2.8, 1/30s, ISO 400. Flat overcast daylight, diffuse shadows,
helmet highlights softly blown out, granular dirt with lateral dust streaks.
Styled like vintage Marlboro campaigns, evoking raw speed and unpolished realism.

03. The visual taste system

Visual taste is what separates "AI content" from premium, intentional creative work. It is your aesthetic identity — the repeatable decisions that make visuals consistent and recognizable: color language, lighting style, composition habits, texture choices, mood, and reference anchors. Taste is trainable — you build it through a system, not by scrolling more.

The 4-step creative director method

1. Curate like a director. Stop scrolling for entertainment. Build a moodboard of references that feel premium, emotional, and aligned with your style. Target: 20–30 strong references to start, then +5 per day.
2. Find the patterns. Treat your moodboard as a dataset, not a collage. Identify 2–4 repeating patterns: a color palette, a lighting signature, texture choices, composition habits, recurring moods.
3. Translate into prompt language. "Cinematic lighting" is too vague. Better: "muted blue-gray palette, matte textures, soft haze, asymmetrical composition, gritty editorial mood, 35mm lens."
4. Micro-practice. Recreate one reference exactly. Generate 5 consistent variations. Change only lighting, then only angle. Build a small signature pack — a mini set that visibly matches.

ChatGPT shortcut: feed your moodboard to ChatGPT and ask it to extract palette, lighting, textures, composition, mood, keywords, ready-to-use prompts, and an aesthetic "NO list" — what breaks your style. The result is a repeatable system.

04. Aesthetic selection algorithm

Use this decision chain before writing any prompt: Goal → Audience → Emotion → Aesthetic → Substyle → Style Kit → Prompts.

The 7 base aesthetics — pick exactly one

Prompt formula — copy/paste template

[SUBJECT], [ACTION/SCENE]. Style: [BASE AESTHETIC] + [SUBSTYLE].
Palette: [3-5 color words]. Lighting: [soft/hard + direction + contrast].
Composition: [negative space / centered / rule-of-thirds].
Texture: [clean / grain / haze / bloom]. Camera: [lens + depth of field].
Mood: [2-3 words]. Post: [filmic grade / matte / glossy].
Negative: low-res, text, watermark, messy background

05. Character lock — visual DNA

The Sprint Rule: to grow fast, stop being random. One account = one character: same face, same skin aesthetic, same vibe. Every project starts by locking a Fixed DNA block before any scene or context is added.

Workflow: 1) choose your aesthetic → 2) copy the Fixed DNA text → 3) convert it with the JSON Engine → 4) add a scenario and generate.

Five Fixed-DNA starting points

• High-contrast organic. Deep warm skin with organic vitiligo patches, dark brows, green eyes, silver hair, natural oily sheen, unretouched texture. Organic patterns create a unique visual fingerprint for instant retention.
• Ultra-light ethereal. Albinism aesthetic — translucent skin, visible veins, platinum lashes, violet-blue iris. Pale skin pops against dark feeds — rare, mystical, expensive.
• Hyper-freckled. Heavy freckles, emerald eyes, ginger curly hair, rosacea flush, chapped lips. Imperfections read as "real" and build trust fast.
• Glass skin. Extremely hydrated but visibly pored, sharp jawline, ice-blue eyes, jet-black bangs. Uncanny-valley effect — ideal for tech and beauty niches.
• Raw realism. Uneven skin tone, cheek redness, open pores, peach fuzz, messy bun. Reads 100% native/UGC — the audience won't clock it as AI.

The DNA Extractor prompt (from a reference photo)

Analyze the uploaded image and create a "Visual DNA" description for AI image
generation. Focus ONLY on biological and physical features: Face (bone
structure, eye shape/color, nose, lips) · Skin (exact tone, texture, specific
imperfections — pores, freckles, moles — and age) · Hair (color, texture,
style). Do NOT describe clothing, background, or lighting. Keep it under 50
words. High density of adjectives.

The Universal Constructor (build from scratch)

PART 1 — FIXED DNA (Biology): Portrait of a [AGE] year old [GENDER],
[ETHNICITY]. Skin: [SKIN TEXTURE]. Eyes: [EYE COLOR], looking at camera.
Hair: [HAIR COLOR & STYLE]. Details: visible pores, slight imperfections,
authentic human look.

PART 2 — VARIABLE CONTEXT (Scene): Location: [WHERE?]. Action: [DOING WHAT?].
Lighting: [LIGHT SOURCE]. Tech specs: shot on iPhone 17 Pro, raw photo,
amateur footage, unedited, realistic shadows.

The JSON Engine — for Nano Banana Pro

{
  "subject": "Main subject description",
  "dna_features": "Eyes, skin, hair, bone structure (Fixed Part)",
  "context_modifiers": "Clothing, location, lighting (Variable Part)",
  "camera_settings": "iPhone 17 Pro, flash, raw photo, unedited",
  "negative_prompt": "cartoon, 3d render, plastic, illustration"
}

Variable context menu — lifestyle triggers

• The bathroom mirror (morning) — mirror selfie, dirty mirror texture with water spots, holding iPhone 17 Pro, fluorescent lighting, messy room, raw photo.
• The night flash (party) — hard direct flash, red-eye reduction, dark background, motion blur, Snapchat quality.
• The car sunset (travel) — driver's seat, golden-hour sunlight on the face, seatbelt on, lens flare, front camera, Instagram-story vibe.
• The gym elevator (fitness) — elevator mirror selfie, full body, harsh overhead lighting, fingerprint smudges, street-style outfit.

06. The 24-shot angle lock pack

Once a character's Fixed DNA is locked, generate the full 24-shot angle set to build a consistent reference library — every prompt anchors back to "the reference character" so identity never drifts.

Base tail (append to every prompt): clean white studio background, soft studio lighting, photorealistic, maintain character consistency — hair, eyes, nose, lips, skin texture, spots, moles, pores, and natural facial features.

The set covers front views (85mm shallow DOF), close-ups (100mm macro), full body front / left / right / back (50mm), plus expression variants — neutral, soft smile, laughing, serious, chin-lowered, slight head-turn. Twenty-four shots per character, generated once, used forever as reference anchors.

07. Hybrid reality workflow

Put your locked AI character into your real footage:

1. Record yourself on a tripod — static camera is required.
2. Screenshot a clear frame from the video.
3. Open Nano Banana Pro → Inpainting tool.
4. Mask only the face.
5. Paste your Fixed DNA text only.
6. Generate — AI keeps the room and clothes, swaps only the face.

08. Model & tool stack — which engine for which job

Swap the stack as the frontier moves — none of these were the answer 12 months ago and none will be in 12 months. What stays: the prompt structure, the DNA lock, the angle set.

09. End-to-end workflow checklist

1. Lock the Fixed DNA (one character, one aesthetic).
2. Generate the 24-shot angle pack — save as reference library.
3. For every scene: paste DNA at the top, then layer shot + lens + light + texture + composition + style.
4. Run a realism upscale pass — enhance skin texture, natural imperfections, film-grain grade.
5. Deliver: raw shots + edit-ready MP4, same DNA everywhere.