Art Direction prompts

Write an art direction brief.

THE PROJECT: [WHAT IS BEING MADE]
WHAT IT IS FOR: [PURPOSE AND WHERE IT APPEARS]
AUDIENCE: [WHO SEES IT]
WHAT IT SHOULD COMMUNICATE: [THE MESSAGE OR FEELING]
WHO IS EXECUTING: [designer / photographer / illustrator / AI / me]
CONSTRAINTS: [BRAND, BUDGET, TIMELINE, TECHNICAL]
REFERENCES I LIKE: [DESCRIBE THEM AND WHY]

Produce:

1. THE OBJECTIVE - what this visual must achieve, not what it should look like. Everything else derives from this, and briefs that skip it produce attractive work that does the wrong job.

2. THE AUDIENCE AND CONTEXT - where and how it will be seen. A visual seen at distance on a poster, at thumbnail size in a feed, or held in the hand needs entirely different decisions.

3. THE ATTRIBUTES - three or four words describing the intended feel, each translated into visual terms. 'Warm' is not direction; 'warm: soft directional light, low contrast, muted earth palette, human hands in frame' is. Do this translation for each, because it is the whole value of a brief.

4. THE REFERENCES, ANALYSED - I described references I like. Do not simply pass them on. Identify what specifically works in each: is it the lighting, the composition, the colour, the subject treatment, or the restraint? A reference without analysis gets copied wholesale, which is both derivative and usually wrong for a different context.

5. THE MUST-HAVES - what has to be in it: product, logo, text space, specific elements. Be exact about text space; leaving room for a headline is the most commonly forgotten requirement.

6. THE MUST-NOT-HAVES - the clichés of this category, the things that conflict with the brand, and anything previously rejected.

7. THE TECHNICAL SPECIFICATION - dimensions and aspect ratios needed, resolution, file formats, colour space, and whether it must work cropped to other ratios. Crops are the second most forgotten requirement.

8. THE DELIVERABLES AND TIMELINE - what is delivered, when, and how many rounds of revision are included.

9. THE SUCCESS CRITERIA - how the work will be judged, stated in advance. Without this, feedback becomes taste and the process becomes circular.

10. WHAT IS OPEN - what the executor gets to decide. A brief that specifies everything wastes the person you hired; one that specifies nothing produces something you did not want. Be explicit about which decisions are theirs.

Help me give feedback on this visual work.

THE WORK: [DESCRIBE IT]
WHAT IT WAS BRIEFED TO DO: [THE OBJECTIVE]
MY REACTION: [WHAT YOU THINK, HOWEVER VAGUE]
WHO MADE IT: [their role and relationship to you]
WHAT STAGE: [first concept / refinement / near final]
WHAT CAN STILL CHANGE: [SCOPE FOR CHANGE]

Produce:

1. MY VAGUE REACTION, DIAGNOSED - I described how I feel about it. Translate that into a specific visual cause. 'It feels cheap' usually means a specific thing: too many typefaces, poor spacing, low-quality imagery, or a muddy palette. 'It doesn't pop' usually means insufficient contrast or hierarchy. Name the likely actual cause. This is the most useful thing here, because it converts taste into something actionable.

2. AGAINST THE BRIEF - does it achieve the stated objective? Assess this before anything aesthetic. Work that is beautiful and does not do the job is a bigger problem than work that is plain and does.

3. THE HIERARCHY CHECK - what does the eye go to first, second, third? Is that the intended order? Most visual problems are hierarchy problems.

4. THE SPECIFIC OBSERVATIONS - grouped, with each phrased as an observation about effect rather than a verdict on quality:
   - Composition and hierarchy
   - Typography
   - Colour and contrast
   - Imagery
   - Spacing and alignment
   - Technical: does it work at the required sizes and crops

5. TASTE VERSUS CRAFT - separate what is a genuine problem from what is my preference. Be explicit. Imposing personal taste under the guise of feedback is the main reason creative relationships break down.

6. THE STAGE CALIBRATION - I said what stage this is at. First concepts should be judged on direction, not detail; commenting on kerning at concept stage is wasted effort on both sides. Adjust the feedback accordingly.

7. THE PRIORITY - if one thing changes, what. Every round of feedback should have a clear priority rather than twenty equal notes.

8. THE WRITTEN FEEDBACK - as I should send it. Specific, referencing the objective, with the priority clear, and separating requests from observations.

9. WHAT TO ASK RATHER THAN TELL - where the designer may have had a reason I cannot see.

10. WHAT NOT TO SAY - the phrases that are useless or damaging: 'make it pop', 'I'll know it when I see it', 'can we see some other options', and any comparison to a competitor without saying what specifically about it.

Help me develop a visual concept.

THE MESSAGE: [WHAT IT NEEDS TO COMMUNICATE]
WHO IT IS FOR: [AUDIENCE]
WHERE IT APPEARS: [CONTEXT AND FORMAT]
WHAT IT IS SELLING OR SAYING: [THE SUBSTANCE]
CONSTRAINTS: [BRAND, BUDGET, PRODUCTION LIMITS]
WHAT HAS BEEN DONE BEFORE: [YOUR PREVIOUS WORK OR COMPETITORS']

Produce:

1. THE MESSAGE REDUCED - the single idea, in one short sentence. A visual can carry one idea. If my message contains three, say which to keep.

2. THE LITERAL VERSION - the obvious depiction. Name it, because it is what will be produced by default and it is usually the weakest option. Knowing the obvious one lets you avoid it deliberately.

3. TEN CONCEPTS - genuinely varied approaches to expressing the idea:
   - Show the result rather than the product
   - Show the problem rather than the solution
   - Show the detail that implies the whole
   - Use scale or contrast to make the point
   - Show the absence of something
   - Use an unexpected juxtaposition
   - Show the process rather than the outcome
   - Show the person rather than the thing
   - Use type as the image
   - Show nothing and let the words carry it
   For each: what it would look like, why it communicates the message, and its risk.

4. THE CLICHÉ LIST - the visual clichés in this category. Name them so they are avoided: the handshake, the lightbulb, the jigsaw piece, the person on a mountain, the arrows going up, and whatever is specific to my sector.

5. THE THREE STRONGEST - with reasoning. Assess each on: does it communicate without explanation, is it distinctive, does it suit the format, and can it actually be produced within my constraints.

6. THE PRODUCTION REALITY - for each of the three, what making it would involve and what it would cost. A concept that cannot be executed within the constraints is not a concept.

7. THE ONE-SECOND TEST - in the context where this appears, how long does the viewer look? If it is a feed or a poster, the concept must land immediately. Assess each option against that.

8. WHAT COULD GO WRONG - misreadings, unintended associations, and how each concept might land differently with a different audience.

9. THE RECOMMENDATION, with the reasoning stated so it can be argued with.

Help me build a mood board.

THE PROJECT: [WHAT IS BEING MADE]
THE DIRECTION I WANT: [DESCRIBE IT]
WHO IT IS FOR: [who will use this board - a designer, a photographer, yourself]
WHAT I HAVE GATHERED SO FAR: [DESCRIBE THE REFERENCES]
WHAT MUST NOT BE COPIED: [COMPETITORS OR WORK THAT IS OFF LIMITS]

Produce:

1. WHAT THE BOARD IS FOR - a mood board communicates properties, not pictures to imitate. Its failure mode is that someone reproduces a reference. State this at the top of the board.

2. THE PROPERTY CATEGORIES - organise the board by what each reference demonstrates rather than by source:
   - LIGHTING: quality, direction, colour
   - COLOUR: palette and how it is distributed
   - COMPOSITION: framing, space, where subjects sit
   - TEXTURE AND MATERIAL
   - SUBJECT TREATMENT: how people or objects are handled
   - TYPE AND GRAPHIC ELEMENTS, if relevant
   A board with one section per property is far more useful than a grid of pretty images.

3. THE ANNOTATION - every reference needs a line saying what specifically to take from it, and what to ignore. An unannotated reference is an instruction to copy the whole thing. This is the single most important element.

4. THE ANTI-BOARD - references showing what to avoid, annotated with why. Often clearer than positive direction, and it prevents the obvious wrong turns.

5. MY EXISTING REFERENCES, ASSESSED - from what I described: are they coherent, or do they pull in different directions? Mood boards often contain contradictory references because each was chosen in isolation. Identify any conflicts and say which to drop.

6. WHAT IS MISSING - the property categories my current references do not address. Gaps get filled by whoever executes, with whatever they prefer.

7. THE ORIGINALITY LINE - the difference between taking a property and reproducing a work. Name where any of my references risk being copied too closely, particularly the ones I said are off limits.

8. THE BOARD STRUCTURE - how to lay it out so the direction is readable at a glance, and the summary line that states the direction in words. The words matter as much as the images; a board with no written direction gets interpreted freely.

9. THE HANDOVER - what to say when presenting it, and the question to ask to check it has been understood: ask them to describe the direction back in their own words.

Help me use AI image generation as a directed process.

WHAT I AM TRYING TO MAKE: [THE OUTPUT]
WHAT IT IS FOR: [PURPOSE AND CONTEXT]
WHERE I AM STUCK: [generating endlessly without getting closer / the results are generic / I cannot describe what I want]
MODEL: [WHICH]
ATTEMPTS SO FAR: [ROUGHLY HOW MANY, AND WHAT YOU HAVE TRIED]

Produce:

1. THE DIAGNOSIS - which failure mode I am in:
   - NO CLEAR TARGET: generating and judging by reaction rather than against a specification. The most common, and the most wasteful.
   - TOO MANY VARIABLES: changing several things per attempt, so nothing is learned
   - WRONG VOCABULARY: describing feelings instead of visual properties
   - MODEL LIMIT: asking for something the model cannot do

2. THE SPECIFICATION - before generating anything else, write down what the image must contain and what it must achieve. Judging output against a specification rather than a reaction is what turns generation into a process. Draft it from what I told you.

3. THE CONTROLLED ITERATION METHOD:
   - Establish a baseline prompt that produces something roughly in the right area
   - Change one element at a time, and record what each change does
   - Keep a log: prompt, result, what changed. Without a log, you re-try things you already tried.
   - Fix the seed where the model allows it, so changes are attributable to the prompt rather than to randomness

4. THE VOCABULARY TRANSLATION - the feelings I described, converted into the visual properties that produce them: lighting, composition, colour, texture, subject treatment. Models respond to visual description and ignore emotional adjectives.

5. THE ELEMENT ISOLATION - if several things are wrong, fix them in order: subject first, then composition, then lighting, then style, then detail. Fixing style before the subject is right wastes both.

6. THE STOPPING RULE - decide in advance how many attempts a direction gets before you abandon it. Generation is compelling precisely because the next one might work, and that is how afternoons disappear.

7. THE COMPOSITE APPROACH - when the model cannot produce the whole image, generate elements separately and combine them in an editor. This is normal practice, not a failure, and it is often faster than prompting for perfection.

8. THE CAPABILITY CEILING - for what I described, what the model genuinely cannot do. Say so plainly so I stop trying.

9. WHEN TO STOP GENERATING - the point at which hiring a photographer, illustrator or designer is cheaper and better than continuing. Include the honest arithmetic: hours spent generating have a cost.

Help me establish a visual direction for this project.

THE PROJECT: [WHAT IT IS]
WHAT MATERIALS ARE NEEDED: [LIST EVERYTHING - web, social, print, video, signage]
WHO PRODUCES THEM: [ONE PERSON / SEVERAL / EXTERNAL SUPPLIERS]
EXISTING BRAND: [WHAT MUST BE FOLLOWED, or 'nothing fixed']
TIMEFRAME: [HOW LONG THIS RUNS]
AUDIENCE AND TONE: [WHO AND WHAT FEELING]

Produce:

1. THE UNIFYING IDEA - what makes these materials recognisably one thing. It should be one strong decision, not five. Options: a distinctive colour treatment, a consistent compositional device, a photographic approach, a typographic treatment, or a recurring graphic element. Recommend one and say why.

2. THE SPECIFICATION - the decisions that apply everywhere, with exact values:
   - Colour: the palette and the proportions each is used in
   - Type: the typefaces, sizes and weights, with a fallback
   - Imagery: the treatment, subject approach, and what is never used
   - Composition: the recurring device or grid
   - Spacing: a numeric scale

3. THE FORMAT ADAPTATION - a table of every material I listed, with how the direction applies to each and what changes. A vertical social post and a wide web banner cannot use the same composition; say what carries across and what adapts.

4. THE RECOGNITION TEST - if someone sees two of these materials in different places, would they connect them? Name the element that does that work, and confirm it survives in every format including the most constrained one.

5. THE MOST CONSTRAINED FORMAT - design for it first. The identity that works in a small square avatar will work everywhere; one designed for a full-width banner often will not scale down.

6. THE PRODUCTION REALITY - given who produces these, how complex the system can be. If several people or external suppliers are involved, it must be simple enough to apply correctly without supervision. Complexity is the enemy of consistency.

7. THE TEMPLATES - the specific templates to build so that producing on-direction material is easier than producing off-direction material. This is what actually delivers consistency.

8. THE ONE-PAGE REFERENCE - the whole direction on a page, with examples.

9. WHAT IS FIXED AND WHAT IS FLEXIBLE - be explicit. A system with no flexibility produces monotony; one with no fixed elements produces incoherence.

10. THE DRIFT CHECK - over my stated timeframe, how to review whether materials are still on direction, and who decides.

Help me judge whether this is finished.

THE WORK: [DESCRIBE IT]
WHAT IT NEEDS TO DO: [THE OBJECTIVE]
WHERE IT WILL APPEAR: [CONTEXT AND FORMAT]
HOW LONG I HAVE BEEN WORKING ON IT: [TIME]
WHAT I KEEP CHANGING: [WHAT YOU ARE FIDDLING WITH]
DEADLINE: [WHEN IT IS DUE]

Produce:

1. THE OBJECTIVE TEST first - does it do the job it was made for? This is the only question that determines whether it is finished. Everything else is refinement. Assess it against the stated objective.

2. THE CONTEXT TEST - view it as it will actually be seen: at the real size, in the real place, on the real device, among whatever surrounds it. Work judged at full screen on a designer's monitor and used as a thumbnail is not finished, it is untested. Say specifically how to check this.

3. THE FUNCTIONAL CHECKLIST - the things that must be right and are checkable rather than a matter of judgement:
   - Legible at the required size
   - Works in the required crops and formats
   - Sufficient contrast
   - Technical specifications met: dimensions, resolution, colour space, file size
   - Nothing important in a region that will be cropped or covered
   - Correct: names, numbers, dates, spelling

4. WHAT I KEEP CHANGING, DIAGNOSED - I said what I am fiddling with. Repeatedly adjusting one element usually means one of three things: that element is genuinely wrong, a different element is wrong and this one is where you can reach, or the work is finished and you are avoiding declaring it. Say which is most likely.

5. THE DIMINISHING RETURNS POINT - given how long I have been working on it, whether further changes are improving it. Most work reaches its quality plateau well before the person stops adjusting it.

6. THE SECOND-GUESSING CHECK - changes that are lateral rather than improvements: swapping one acceptable option for another. Name any I described.

7. WHAT WOULD ACTUALLY IMPROVE IT - if anything. Be specific and limited. If the answer is that nothing material would, say so.

8. THE FRESH EYES TEST - the practical ones: leave it overnight, view it mirrored or upside down, view it at thumbnail size, and show it to someone who has not seen it before and ask what they notice first.

9. THE VERDICT - finished / one specific thing to fix / needs more work in a named area. Be decisive.

10. THE DEADLINE REALITY - given when it is due, what is worth doing in the time remaining and what is not.