The Reference Framework: How to Direct With Images, Not Adjectives

The Reference Framework is a method for turning a reference deck into direction. A reference only directs when you annotate why it is there, naming four layers on every image: subject, light, palette, and composition, plus a fifth discipline of stating what to ignore. Without annotation, a reference is decoration, not instruction.

Every director keeps a folder of images. They are the fastest way to say what words struggle to: this quality of light, that register of a face, this restraint. But the folder is where the discipline usually ends. The images get dropped into a deck, arranged for rhythm, and handed over with a sentence like "here is the world we are building." Then the shoot comes back looking like a compromise, and no one can say exactly where it went wrong.

It went wrong at the reference. A pretty image with nothing attached to it is not direction, it is a Rorschach test. The Reference Framework extends the Essenzi Brief Standard to the layer beneath it: how a single image becomes a single line of instruction that a photographer, a retoucher, or an AI model can execute without you in the room.

Why Mood Boards Become Mood Soup

A reference deck fails for one reason: an image carries many qualities at once, and the deck never says which one you meant. One frame holds a subject, a lighting setup, a palette, a composition, a mood, a location, a wardrobe, a decade. When you paste it in unannotated, you are handing over all of them and hoping the reader picks the same one you did.

They will not. Everyone reads a reference through their own craft. The photographer sees the light and the lens. The stylist sees the coat. The retoucher sees the grade. The client sees the model's expression and wonders if she looks approachable enough. Five people open the same deck and privately build five different campaigns. The deck feels aligned because everyone likes the images. It is not aligned, because no one agreed on what the images were for. That is mood soup: shared taste, zero instruction.

The Four Layers to Name on Every Reference

An annotated reference answers four questions with decisions, not adjectives. Name all four and the image stops being ambiguous. Below is the framework as a spec block: the layer, the question it settles, and the kind of annotation that settles it.

Subject

Subject is not "a model." It is who or what is in the frame and, more importantly, the energy they carry. A reference of a woman looking past the camera might be there for her stillness, not her face; for the way she occupies the frame without performing for it. Name the register, not the person: "receives the camera rather than projecting," "somewhere to be but no hurry to get there." That is what you are actually pointing at.

Light

Light is the layer that most decides how an image reads, and the one collaborators most often read differently. Name four things: source (single window, hard strobe, mixed practical), direction (camera left, above, behind), quality (soft and wrapping, or hard with a defined edge), and temperature (cool daylight, warm tungsten). A reference tagged only "beautiful light" has said nothing. Tagged "single north window, camera left, soft, cool, shadows allowed to fall," it has given a gaffer a plan.

Palette

Palette is the colour that carries the mood, stated as a relationship rather than a swatch. Which tones dominate, which single accent is permitted, and how far it is allowed to spread. "Bone and oatmeal throughout, one oxblood accent held under ten percent of frame, shadows warm." This is the same colour logic our note on brand DNA treats as a signature, applied to one image so that styling, set, and grade all read the frame the same way.

Composition / Frame

Composition is the geometry: how far the camera sits, where the subject falls in the frame, and how much emptiness surrounds them. Negative space is a decision, not an accident, and it carries meaning: generous empty air reads as restraint and singularity, a tight crop reads as intimacy or pressure. Name the distance (full-length, portrait, detail), the placement (centred, off to one side), and the air (deep negative space above, or filled edge to edge).

The Fifth Discipline: What to Ignore

The four layers say what to take. The fifth says what to leave. Almost no reference carries only the quality you want. You love the light in an image but not its garish styling; you want the palette but not the frenetic composition; you are pointing at the subject's energy and the location is irrelevant. If you do not say so, a collaborator will import the wrong element in good faith, and you will spend the shoot correcting a mistake you authored.

So every reference gets an ignore line. "Here for the light and the negative space. Ignore the wardrobe, ignore the era, ignore her expression." This single sentence is what prevents one strong image from over-directing the whole campaign. It is the difference between a reference that points and a reference that shoves.

Turning an Annotated Reference Into a Brief Line

Once a reference is annotated across the four layers and given its ignore line, it collapses into a single line of the brief. That line is portable: it means the same thing to a photographer on set, a retoucher three weeks later, and an AI model at one in the morning. Here is one reference taken all the way through.

The brief line reads like a sentence because it is one. Nothing in it is an adjective doing decorative work; every clause is a decision traceable to the annotation above it. Hand that line to three different people and they converge, because there is nothing left to privately interpret. That convergence is the entire point of the framework.

How to Annotate a Reference

The sequence is the same for one image or forty. Run it in order and do not skip the last two steps, which are the ones most decks omit.

1. Choose the reference for one reason. Before you annotate, know why the image is in the deck at all. If you cannot say the one thing it is carrying, it is decoration and it comes out.

2. Name the subject and its energy. Not the person, the register: how they hold themselves, how they meet or refuse the camera.

3. Name the light in four parts. Source, direction, quality, temperature. If you cannot read one of them from the image, say what you want instead.

4. Name the palette as a relationship. Dominant tones, permitted accent, how the shadows behave. Not hex codes, logic.

5. Name the frame. Distance, placement, negative space, camera height.

6. Write the ignore line. State plainly what in this image is not the point, so no one imports it by accident.

7. Collapse it into one brief line. If the four layers cannot be written as a single executable sentence, one of them is still vague. Fix it there.

Why This Is What Lets AI Execute Faithfully

An AI model is the most literal collaborator you will ever brief. It has no instinct to fall back on and no shared history with your house. It executes what it is told, exactly, which makes it a merciless test of your direction: a vague reference produces a generic image, every time, and the fault is upstream.

An annotated reference solves this because its four layers already speak the language a model needs. Named light, named palette, named framing, and a stated register translate directly into prompt syntax, the same way our guide to editorial photography prompts describes. The ignore line matters even more here than on set, because a model cannot tell which part of a reference you loved and will faithfully reproduce the parts you did not. Direct with images that carry their own instructions, and the machine becomes an instrument. Direct with mood soup, and it hands the soup straight back.

Frequently asked

What is the Reference Framework?

The Reference Framework is a method for turning a reference deck into direction. A reference only directs when you annotate why it is there, naming four layers on every image: subject, light, palette, and composition, plus a fifth discipline of stating what to ignore. Without annotation, a reference is decoration, not instruction.

Why do most mood boards fail to direct a shoot?

Most mood boards are pretty images with no instruction attached, so everyone who opens them reads them differently. A photographer sees the light, a stylist sees the clothes, a retoucher sees the grade. Because the reference never says which of those elements matters, the deck becomes mood soup: a shared feeling that produces inconsistent images.

What are the four layers to annotate on a reference?

The four layers are subject (who or what is in frame and its energy), light (source, direction, quality, and temperature), palette (the specific colours that carry the mood), and composition or frame (distance, geometry, and negative space). Naming all four converts a single image into a line of a brief that a person or an AI model can execute faithfully.

Why does every reference need a "what to ignore" note?

A reference almost never carries only the quality you want. You might love its light but not its casting, or its palette but not its composition. Stating what to ignore prevents a collaborator from importing the wrong element by accident. The ignore note is what stops a single image from over-directing the shoot in ways you never intended.

How does an annotated reference help AI execute faithfully?

An AI model, like a photographer or a retoucher, executes what it is told, not what you felt. An unannotated reference gives it nothing to act on. When a reference is annotated across the four layers, that annotation translates directly into prompt language: named light, named palette, named framing. The same clarity that lets a stranger shoot the image lets a model generate it.