The essential craft of directing AI to create photography that brands actually pay for. In two focused hours you go from typing prompts to directing them. Let's build.
A reel of some of my best films, made entirely with AI. This is the bar. By the end of class you will have the exact system to direct work like this yourself.
Embedded preview not loading? Use the button to open it in Canva.
You don't need all of them, but you should know what each one is for. Think with the LLMs, generate with the image tools, then add motion, voice, and the final edit. Tap any logo to open it.
Sixty seconds of warm-up, then we build. Get set up, lock in the one rule, and keep the promise in mind.
ChatGPT, your image model (Nano Banana 2 or Midjourney), and a folder of reference shots. We build live, so follow along in real time.
You are the director, not the audience. Every image is a decision you make on purpose, never a surprise you sit and wait for.
By the tenth lesson you will write a single prompt that beats most of what gets posted online. That is the bar we are setting today.
Your AI images right now are...
Seven blocks. Each one is a skill you can use the moment class ends. Follow the timeline, it's paced for a live 2-hour session.
Why pros direct AI instead of gambling on it.
Locking your vision before you spend a single credit.
The 4-block structure and the repeatable JSON system.
The biggest quality levers, made interactive.
Skin, texture, and the details that fool the eye.
Nano Banana 2, GPT Image 2, Midjourney, Flux & more, who wins what.
The single shift that separates a hobbyist from someone who gets paid: you treat AI like a camera with infinite settings, not a slot machine.
It does exactly what you say, not what you mean. Vague in, random out. Brief it like a junior creative.
The tool is free; taste is rare. You can't prompt what you can't see. Your taste ceiling is your output ceiling.
First output is a draft. Change one thing per pass, compare, keep. Controlled iteration beats infinite re-rolls.
Random output comes from a random brief. These three habits decide the image before AI touches it.
Collect 100 references in your target aesthetic. Tag why you saved each. Quality reference in = quality prompt out.
6–9 aligned images, not a Pinterest dump. Each one answers a question: light? color? texture? mood?
Subject? Mood in one word? Light? Palette? Lens? Answer all five or don't hit generate.
Subject, Style, Lighting, Camera. Miss one and AI fills the gap with randomness. Tap each block to expand.
A paragraph prompt is hard to repeat. JSON gives AI a clean, parseable structure, so the same recipe produces the same look every time, and you change one field without rewriting everything. Build one live ↓
Light is the language of mood. Click a setup, watch the sphere react and read what it's for.
These are the giveaways that scream "generated". Name them, then kill them with the right keywords. The fix for most: texture, asymmetry, and grain.
The Genesis Prompt is a 4-phase architecture for building one powerful, precise image instruction, Foundation, Composition, Sensory Detail, Narrative. Pull exact language from the keyword vault below; click any chip to drop it into your build.
The technical blueprint, genre, era, lighting, fidelity.
The director's frame, angle, framing, perspective.
The textural language, materials, surfaces, imperfections.
The final intent, emotion, mood, atmosphere.
AI chases mathematical perfection, but our brains read imperfection as real. The Reality Serum is the antidote: you intentionally "damage" the AI's perfection with the story of the real world. Spot the symptom, apply the cure.
Style isn't magic, it's a documented system. The Alpha → Master pipeline turns one style definition into endless on-brand images. Then a toolkit of named methods for every job.
Your style locked as reusable rules: light, lens, palette, mood.
Alpha Prompt as system instructions + a one-line brief.
Expanded, technically precise, but always on-style.
Same DNA every time. Consistency by design.
Your Alpha Prompt provides the style; your one-line input provides the specifics. The LLM does the rest.
Style is a documented system, not a vibe. Dissect work you admire and define your recurring choices.
Flat AI images become immersive when you add environmental overlays, the atmosphere is a character in the frame.
Start with an image you love and work backwards into a prompt you can reuse.
Product photography without the photoshoot, swap your real product into any lifestyle scene.
A casting-director prompt that builds a reusable character profile. Copy it into ChatGPT.
Midjourney is a playground, chase the vibe through reference and iteration, not perfect paragraphs.
No model wins everything. Pick a model below to see its strengths, weaknesses, and best use case, and watch the radar update. Scores are an opinionated field guide, not lab benchmarks.
7 dimensions, scored 0–10. Compare the highlighted model against the field average.
Where each model sits, fast & easy vs. slow & controllable.
The fastest way to choose. Match your job to the model that's purpose-built for it.
The client should never spot the AI. Tick each box, your readiness score updates live.
Tick the boxes as you inspect your image.
Bookmark these. Generators, upscalers, and reference wells, all live.
Google's photoreal, edit-and-consistency powerhouse.
gemini.google.com ↗Best-in-class prompt adherence & text rendering.
openai.com ↗Aesthetic king for editorial & fashion mood.
midjourney.com ↗Realism & open ecosystem, train your own LoRAs.
bfl.ai ↗Typography & text-in-image specialist.
ideogram.ai ↗Turn your best stills into consistent motion.
klingai.com ↗Add real resolution & texture for print.
magnific.ai ↗Real-time generation, enhance & upscaling.
krea.ai ↗Build your 100-image visual library here.
pinterest.com ↗