The most useful mental model for prompting is this: you are briefing a collaborator with flawless technical execution, no taste of their own, no knowledge of your brand and no memory of the last thing you asked for.
Everything that works about prompting follows from that model. Everything that fails follows from forgetting it.
Structure a visual prompt like a photography brief
Amateur prompts describe subjects. Professional prompts describe photographs. Cover these six elements in order:
- Subject. Who or what, with the specificity a casting note would carry.
- Action and expression. What they are doing and how they appear while doing it.
- Environment. Where, including the material and architectural detail that makes a place specific rather than generic.
- Light. Source, direction, quality and colour temperature. This single element controls more of the result than any other.
- Camera. Focal length, distance, angle, depth of field.
- Treatment. Colour grade, contrast, film characteristics, mood.
Compare a woman drinking coffee against: a woman in her early thirties in a linen shirt, mid conversation and laughing, in a narrow city apartment kitchen with worn wooden counters, soft overcast window light from camera left, shot on an 85mm lens at close range with shallow depth of field, warm muted grade with lifted shadows.
Same subject. Entirely different category of output.
Light first, if you change only one thing. Specifying the light source, its direction and its quality produces a larger realism improvement than any other single addition to a prompt.
Build a brand prompt system, not individual prompts
The mistake that costs the most time is treating each image as a fresh prompt. Consistency does not come from writing better sentences. It comes from a reusable system with fixed and variable parts.
A workable structure:
- Brand constants. Colour palette, lighting philosophy, grade, lens conventions, level of realism. These never change across your library.
- Character constants. The locked identity description for each character, applied identically every time.
- Scene variables. Setting, action, styling, time of day. These change per asset.
- Format variables. Aspect ratio, crop and composition per platform.
- Negative constraints. An explicit list of what must never appear.
Written this way, producing a new asset means changing the variables rather than rewriting the brief, which is what makes a library of four hundred images look like one brand.
Art direction, not just generation
Proklisi treats AI as a creative tool rather than a shortcut. Every image, video and caption is produced with intent, designed to hold brand identity and drive outcomes.
Prompting for copy and script
The same principles apply to language models, with different variables. A weak copy prompt asks for a caption. A strong one specifies:
- Audience, described specifically enough to exclude people
- Voice, with an example of prior copy that hit the mark
- Objective, the single action you want taken
- Constraints, length, structure, claims that cannot be made
- Format, hook, body, call to action, and where each begins
Then ask for ten variants rather than one. Selection is where a person adds most value, and it is fast.
Five failure patterns and their fixes
| Symptom | Cause | Fix |
|---|---|---|
| Output looks generic | Prompt describes a category, not a specific instance | Add concrete detail: materials, wear, specific objects |
| Too perfect, reads as synthetic | No imperfection specified | Ask for texture, asymmetry, imperfect framing, clutter |
| Character drifts between images | Identity described rather than locked | Use a fixed reference set and identical identity text |
| Inconsistent across a set | Brand constants not applied | Move light, grade and lens into a fixed template |
| Unwanted elements keep appearing | No negative constraints | Maintain an explicit exclusion list |
The part prompting cannot do
No prompt supplies a point of view. Models generate from the distribution of what already exists, which makes them structurally inclined toward the average. Left to itself, generated output converges on a recognisable house style that belongs to nobody.
Distinctiveness comes from a deliberate creative direction imposed from outside the model: a defined aesthetic, an unusual constraint, a specific reference world. That decision is human work, and it is the reason two teams using identical tools produce work of completely different quality.