Six references, each with a job

The problem with a mood board

A client sends twenty pictures in a shared folder. They are all “the feeling we want”. Half are lit differently from the other half, three are illustrations and the rest photographs, and two contain a competitor’s product. You generate something that averages them and it satisfies nobody, because nobody agreed on what they were agreeing to.

A reference pack fixes this by attaching a job to each image. Not “this is nice”, “this one is here for the light”.

The six slots

Six is usually enough, and forcing the limit is part of the method.

1. Light. One image that demonstrates the lighting: direction, hardness, colour temperature. This is the single most influential reference and the one people forget.
2. Palette. One image whose colours are right, even if its subject is wrong.
3. Composition. A frame whose arrangement you want to echo: subject placement, camera height, how much room around things.
4. Subject or product. What the thing actually is. For a real product this is a photograph of the real product, and it is non-negotiable.
5. Texture and finish. Grain, gloss, matte, film stock, print quality. This decides whether the result looks like a photograph, a render or a scan.
6. The anti-reference. One image that is close to the brief and wrong, with a sentence saying why. This is worth more than two of the others, because it converts taste into a statement people can check.

Write one line under each. “This one: the hard sunlight from the left, not the model, not the styling.” That sentence is what stops the client saying at review that they never meant the person to look like that.

Why this is not just admin

The labels are how you use the references technically. Later in the course you will be giving images to the model directly rather than describing them, and each slot maps onto a different mechanism:

• The composition reference becomes a depth or edge control.
• The palette reference becomes a colour-matched starting image at low denoise.
• The light and texture references become style adapter inputs, or vocabulary you extract by interrogating them.
• The product reference becomes a composite, not a generation.

A pile of twenty unlabelled images cannot be routed this way. Six labelled ones can be, and the routing is most of the craft in the middle of this course.

Where references come from

Use what is in front of you. Photographs you took, stills from films, museum collections that publish open-licence scans, the client’s own archive, free stock libraries. There is no need to be precious about looking at other people’s work to describe what you want: that is what art direction has always been.

Two cautions that matter in practice. First, a reference is for direction, not for reproduction: taking a living illustrator’s picture and conditioning on it heavily enough that the output is recognisably theirs is a problem regardless of what any court eventually decides about training data, and the later block on rights returns to it. Second, if the client supplies references, ask where they came from. “We found it online” is the beginning of an awkward conversation later, not the end of one.

The one-line test

Before generating, read the pack back as a sentence: “Hard side light like ref 1, the muted greens of ref 2, arranged like ref 3, our actual bottle from ref 4, with the matte film texture of ref 5, and specifically not the plastic studio look of ref 6.”

If that sentence is coherent, you have a brief. If it contradicts itself (hard sunlight and soft studio texture) you have found the disagreement now, in a message, rather than at the review where somebody has to lose.

Building the pack when the client sends nothing

Often you get no references at all, and asking for some produces silence. Do not wait. Assemble a pack yourself and send it as a question: “Which of these three directions?”

Three columns of three images each, each column internally consistent, each clearly different from the others. Warm and documentary; cool and graphic; soft and domestic. The client picks one in under a minute, which is a decision they can make, whereas “what do you want it to feel like” is one they usually cannot.

This is the oldest trick in art direction and it works because recognition is much easier than description. You are not asking them to design anything. You are giving them three things to react to, and a reaction is a specification you can build on.

Two practical notes. Make the three different: three variations on the same idea teaches you nothing. And keep the column they chose, labelled, because it becomes the pack, and the two they rejected become your anti-references with a reason attached.