How to stop generating random content and start directing AI with intention. The framework that separates people who “use AI” from people who direct it.
Most people prompt AI like it's a search engine: type a wish, hope for the best. That's why they get generic, forgettable results. The output ceiling isn't the model. It's the input.
This masterclass teaches you three techniques — Reference Anchoring, Role Briefing, and Constraint Stacking — that, when combined, transform AI generation from random into intentional.
There are two types of AI users. One says “make me an image.” The other says “here's exactly what I see in my head, and here's how to build it.”
The first person gives AI infinite directions to go. The second gives it one. That's the intention gap. And closing it is the single biggest upgrade you can make.
Make a dragon standing on a castle.
The amateur prompt gives the AI a subject. The director prompt gives it a shot brief. More constraints = more focus = better output. Every. Single. Time.
Instead of prompting from scratch, you upload an existing image — a film frame, a viral shot, a masterpiece — and extract what already works.
You're not copying. You're using proven aesthetics as visual DNA and recombining them with your own subject and story. Every great filmmaker does this — they call them “look books” and “mood boards.”
The tonal DNA of the image. Warm vs cool, saturated vs muted, the overall chromatic identity.
Pull from: a Wes Anderson film still, a Saul Leiter photograph, a Kodak Portra contact sheet.
"Match the desaturated teal-and-amber palette of this reference"
Pro move: Don't use one reference — blend 2-3. Pull the color palette from one, composition from another, and texture from a third. This creates something familiar yet completely original. More on this in Section 06.
Most people prompt AI like it's a tool: “make me an image.” That's why they get tool-level results.
Assign the AI a specific role, identity, and expertise before giving it any task. You're not asking a chatbot — you're briefing a specialist. When the AI believes it's a cinematographer, it thinks in lenses and light ratios. When it believes it's a colorist, it thinks in tonal curves and film stocks.
You are a cinematographer who has shot 15 feature films. You think in lenses, light ratios, and focal distances. Every frame must serve the narrative.
What it activates: Activates camera-specific knowledge: lens compression, depth of field physics, sensor characteristics.
Notice the last option: you can stack roles. “You are both a cinematographer AND a color grading specialist” creates intersection expertise — the whole exceeds the parts.
Constraints don't limit the AI — they focus it.
Instead of vague prompts like “make it cinematic,” you stack precise constraints: the camera angle, the lens, the lighting setup, the mood, the texture, the imperfections. Each constraint eliminates a thousand possible outputs. What remains is your image.
low angle, looking up at the subject. 85mm f/1.4, shallow depth of field.
Try it: toggle all 8 constraints on. That's 8 precise instructions. The AI has almost nowhere to improvise — and that's exactly the point. You're not removing creativity. You're directing it.
The Master Prompt Formula. This is the sentence structure that makes every constraint land.
A (camera angle), of (subject), (environment), (camera body), (lens), (lighting)
Extreme wide establishing shot, of a woman in a leather jacket, an empty parking garage at midnight. ARRI Alexa 65, 85mm f/1.4. Golden hour backlight.
Play with it. Swap the angle from “Low angle” to “Bird's eye view” and watch how the entire shot brief transforms. That's one variable. Imagine stacking all six.
The best creators don't use one reference. They blend 2-3, pulling different elements from each. Color from one, composition from another, texture from a third.
This is where originality lives. Each reference is familiar. The combination is yours. Nobody else has blended these exact ingredients in this exact proportion.
Color palette — halation reds, deep blacks, neon bleed
Composition — staged symmetry, suburban uncanny
Texture — high contrast grain, raw urban grit
A portrait that feels like a neon-lit crime scene in a quiet suburb, shot on expired film.
The trick: describe what you're taking from each reference explicitly. “Use the palette from A, the framing from B, and the grain from C.” Vague references produce vague blends.
Before you generate, make the AI critique its own prompt.
Instruct it: “Find three flaws in this prompt from the perspective of a 200 IQ film critic and fix them.” Watch the prompt refine itself through adversarial self-examination. Two rounds of this typically produce a prompt 3-4x more specific than the original.
A woman standing in the rain in a city at night with neon lights.
Step through all 5 tabs. Watch the prompt evolve from 12 words to 90+. Each critique eliminates vagueness. The final draft has a specific time (11:47 PM), a specific location (Shinjuku crossing), and light described by physics, not adjectives.
Each technique is powerful on its own. When you stack all three together — reference, role, and constraints — you stop generating random content and start generating with intention.
Find images that capture the look you want: one for color/mood, one for composition, one for texture. Film stills, photography, concept art, even screenshots from viral AI content.
Before any creative prompt, assign the AI a specific expert identity. "You are a cinematographer who has shot 15 feature films." This changes how it interprets every subsequent instruction.
Tell the AI what you want to create, using the references as visual anchors. "Match the color grading of reference A, the composition of reference B." Your subject + their aesthetic DNA.
Layer in camera body, lens, lighting setup, mood, texture, imperfections, environment. Each constraint eliminates 1000 possible outputs. More constraints = more focus = better results.
Before generating, ask the AI: "Find three flaws in this prompt from the perspective of a 200 IQ film critic and fix them." The prompt improves itself.
Generate 2-4 variants. Your first output won't be perfect. Adjust which elements you pull from each reference. Change one constraint at a time. Dial it in.
Reference gives you direction.
Role briefing gives you expertise.
Constraint stacking gives you precision.
Together, they create intention.
Each recipe uses all three techniques: a role assignment, a constraint-stacked prompt, and named references. Copy them. Study why they work. Make them yours.
You are a cinematographer obsessed with symmetry and one-point perspective. Every frame must feel mathematically precise yet deeply unsettling.
A perfectly symmetrical one-point perspective shot down a long institutional hallway. Fluorescent tube lighting overhead, casting cold blue-white light with hard shadows on the linoleum floor. A single figure stands at the vanishing point, slightly too still. The walls are mint green, institutional, 1970s. Shot on ARRI Alexa Mini LF, 21mm Zeiss Supreme Prime. Deep focus — everything razor sharp from foreground to vanishing point. Slight wide-angle distortion. No warmth. No comfort.
Role assignment (symmetry-obsessed DP) + constraint stacking (specific lens, lighting physics, color temperature) + reference DNA (institutional horror). The "slightly too still" figure adds narrative tension without describing a story.
You are a street photographer and a color grading specialist. You see beauty in mundane moments and grade them into cinema.
A woman has just missed her bus. She's frozen mid-reach, hand extended toward the closing doors, grocery bag swinging from her other arm. A single orange has rolled out onto the wet sidewalk. Late afternoon light rakes across the scene from the left at 15 degrees. Her shadow stretches long behind her. The bus exhaust creates a soft haze. Shot on Leica Q3, 28mm f/1.7. Kodak Portra 400 pushed one stop — warm highlights, lifted shadows, soft grain. The moment between action and acceptance.
Stacked roles (street photographer + colorist) create intersection thinking. The orange is micro-storytelling. "The moment between action and acceptance" gives the AI an emotional vector, not just a visual one.
You are both a product photographer and a cinematographer. You photograph objects the way Terrence Malick photographs people — with reverence, light, and implied narrative.
A handmade ceramic mug sits on a rough-hewn wooden table. Steam rises from black coffee, catching a single beam of morning light that enters from a small kitchen window, camera right. The light carves across the table grain, illuminating the mug's uneven glaze and the potter's fingerprints still visible in the clay. Dust motes float in the light beam. Everything outside the beam falls into soft shadow. Shot on Fuji GFX 100S, 110mm f/2.0 macro. Focus stacked: sharp from the mug handle to the steam. Fuji Pro 400H color science — muted, clinical, cool greens in the shadows.
Treating a product like a character ("potter's fingerprints") activates the model's narrative rendering. Raking light reveals texture at microscopic levels. Focus stacking is a pro-level constraint that eliminates the soft-everywhere look.
You are a cinematographer who specializes in psychological tension. You create unease through framing, not content.
Two people sit across from each other at a small cafe table. Medium close-up, slightly telephoto — 100mm f/2.0 compresses the space between them. One is in sharp focus, the other softly blurred in the foreground, a sliver of their shoulder and jaw visible at frame left. The focused subject stares directly past camera, expression unreadable. Harsh overhead practical light creates deep eye sockets. The cafe behind them is dark, out of focus, irrelevant. Color desaturated to near-monochrome except for the warm amber of a single espresso cup. Shot on RED V-Raptor, Cooke Anamorphic. Kodak Vision3 500T.
The selective focus between two people creates immediate psychological tension. "Expression unreadable" is more powerful than any named emotion — it forces the viewer to project. Anamorphic + desaturation = modern noir.
Run through this before every generation. If you can check all 10, your prompt is ready.
13 ready-to-use system prompts, director-style templates, and workflow recipes. Copy them into any AI tool. Each one embodies the Reference + Role + Constraint framework.
You are a Creative Director who orchestrates AI image generation. Never pass raw user text directly to the model. Pipeline: 1. Analyze Intent: What does the user actually want? What emotion should the viewer feel? 2. Select Domain Mode: Cinema / Product / Portrait / Editorial / Landscape 3. Construct Brief using the 6-Layer Constraint Stack: - Camera Angle (15%): Shot type, perspective, framing - Lens & Camera (20%): Body, focal length, aperture, sensor characteristics - Lighting Physics (25%): Source direction, quality, color temperature, shadow behavior - Subject & Environment (20%): Who/what, where, surface materials, props - Texture & Film Stock (10%): Grain, color science, imperfections - Mood & Narrative (10%): Emotional vector, implied story, tension BANNED: "4k", "8k", "masterpiece", "highly detailed", "best quality", "trending on artstation" INSTEAD USE: Named camera bodies, specific film stocks, lighting physics, and narrative context. Output a single, optimized prompt string. Every word must earn its place.
You are a veteran cinematographer with 20 years of experience shooting feature films and high-end commercials. You think exclusively in the language of cinema: - Every shot has a purpose. You never frame without intent. - You specify camera body, lens, aperture, and focal distance for every setup. - You describe lighting by physics: direction (clock position), quality (hard/soft), color temperature (Kelvin), and falloff. - You understand that film stocks are not filters — they are complete tonal systems with distinct color response curves, grain structures, and contrast behavior. - You reject vague terms: "cinematic" means nothing without specifying WHY it's cinematic. - Your mood baseline is: precise, deliberate, economical with words. When the user describes what they want, translate it into a shot brief that a gaffer, grip, and camera operator could execute on a real set. Then output that brief as an image generation prompt.
You are a prompt engineering critic with the analytical mind of a 200 IQ film critic and the visual vocabulary of a Master cinematographer. When given any image generation prompt, you will: 1. CRITIQUE: Identify exactly 3 weaknesses in the prompt from a visual storytelling perspective. Be brutal but specific. Don't say "add more detail" — say exactly WHAT detail and WHY it matters. 2. REWRITE: Produce an improved version that fixes all 3 weaknesses while preserving the original creative intent. 3. EXPLAIN: For each change, explain what visual difference it will produce in the output. Rules: - Never add banned keywords (8k, masterpiece, trending on artstation, etc.) - Always add a camera body + lens if missing - Always add lighting physics if the original only describes mood - Always add at least one imperfection (grain, dust, aberration) if none exist - Prefer specific over general: "Shinjuku crossing at 11 PM" beats "city street at night"
You are simultaneously a cinematographer, a color grading specialist, and a production designer. When given any scene description, you will analyze it from all three perspectives and produce a unified prompt. Cinematographer lens: What camera, lens, angle, and movement best serves this scene? What depth of field creates the right focus hierarchy? Colorist lens: What color palette, film stock, and tonal treatment creates the right emotional response? What should the color temperature be? Production Designer lens: What should be IN the frame? What materials, textures, props, and environmental details sell the reality? What should be deliberately excluded? Each perspective must contribute at least 2 specific constraints. The final prompt must be a single paragraph that integrates all three viewpoints seamlessly. Mark where each discipline contributed.
You stop generating random content
when you combine all three.
Reference gives you direction. Role briefing gives you expertise. Constraint stacking gives you precision. Together, they create intention.
Start Directing with Luno