A skill that makes Claude write the art direction for a project as numbered rules that a photographer, an illustrator, a designer or an image tool can follow, and that someone else can check from the picture alone. Give it a chosen idea, your brand guidelines and your references. It counts every file in every shape, describes each reference in words (the one quality to take, what to ignore) and finds the three to five decisions that make a picture yours with the logo covered. Then it writes rules for light, colour, composition, casting, styling, set, type over images and illustration, each with a number, a test and a tolerance, followed by do and don't pairs, hard stops and a handoff for each maker.
Inside are five reference files. A rule library holds the numbers working photographers and retouchers use: light on a clock, lighting ratios, colour temperature, camera height, lens feel and crop maths, with mood words like premium or warm translated into figures. Another file covers handoffs, safe zones, contrast thresholds and US legal flags, with settings for Midjourney, Nano Banana, GPT Image and Firefly, all checked on 17 October 2026 with sources. There is a review method and a full worked example for a fictional cookware brand. It works one stage at a time and ends each with one question and a sensible default. It will not name a living artist as a style or cast from a real person's face. Where a product detail is missing, it leaves a marked gap and asks.
It cannot take photographs or make images itself. It writes the rules and the prompts for your own photographer and tools, and checks what comes back when you paste it in. Turn on code execution so it can sample colours from your pictures, compute contrast, draw safe-zone overlays and export the spec as a PDF or a deck. Colours sampled from a photo are estimates, so the codes in your brand guide win. Legal points are flagged for your lawyer, not settled. You get the most from it by sharing past pictures you love and ones you wince at, and by saying early who makes what, because that one answer shapes every handoff.
Claude skill · version 1.0
Art direction spec
Checking your access.
In the download · 9 files
- SKILL.mdThe method Claude follows: intake, signature, rules, guardrails, handoffs and check, one stage at a time.
- README.mdHow to install it in claude.ai or Claude Code, how to start, and the licence.
- references/references-in-words.mdTwelve qualities to read in any picture, the reference line, translating a named photographer's style into rules, safe sources, the brand audit and finding the signature.
- references/rule-library.mdThe numbers behind every rule: light on a clock, lighting ratios, colour temperature, camera height, lens feel, framing, casting, styling, set, type and image, illustration, retouching, mood words translated and crop maths.
- references/handoffs-and-specs.mdWhat each handoff carries, settings for Midjourney, Nano Banana, GPT Image and Firefly, safe zones, file specs, contrast thresholds, US legal flags and usage terms, each sourced and checked on 17 October 2026.
- references/review-and-drift.mdThe check sheet, how to measure each rule from the picture alone, the drift each kind of maker falls into, writing one-change fixes and when to stop regenerating.
Install it and begin
01Install it in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/art-direction-spec.zip -d ~/.claude/skills/# Start it with /art-direction-spec. If this is your first skill, type /reload-skills once.On claude.ai there is nothing to unzip: upload the zip as it is. Every step is on Install and use a Claude skill.
02Your first message to ClaudeText Use the art direction spec skill. The idea is [THE CHOSEN IDEA IN ONE SENTENCE]. The pictures will run on [WHERE THEY WILL RUN]. They will be made by [WHO MAKES THEM, FOR EXAMPLE A PHOTOGRAPHER] with [THE BUDGET OR SHOOT DAYS]. I have attached our brand guidelines and references. Start with the header and tell me what is missing.Attach your brand guidelines, the brief or chosen idea, three to eight references and a few past pictures you like. Rough references are fine; the second stage writes down what to take from each one.
Side by side
What the brief says, and the rule the spec writes
| What a brief usually says | What the spec writes instead | |
|---|---|---|
| Light | Natural, warm light. | L1 MUST: one soft window key from camera left, 9 to 10 o'clock, 20 to 40 degrees up. Test: one shadow per object, falling right. Tolerance: one hour on the clock. |
| Colour | On-brand colours. | C1 MUST: brand red only on the product and one prop, under 10% of the frame, counted on a 10 x 10 grid. |
| Composition | A hero shot of the product. | K1 MUST (4:5): the product fills 55 to 65% of the frame height, and the top 28% is plain wall kept for words. |
| Casting | Real, diverse people. | P1 MUST: people mid-task, eyes off camera in 8 of 10 frames, at least three age bands across the set, matched to your own customer data. |
| Type and image | Make the text pop. | T2 MUST: every type colour computed over every background. 4.5:1 for normal text; 3:1 only for type at 72 px or more in a 1080 px frame. |
| References | Like that famous photographer's work. | R1, for the light only: a hard key at 10 o'clock and a crisp shadow its own height, falling right. Ignore the set and the crop. No names. |
| Do and don't | Make it look premium. | Do: one hard key drawing a highlight along the rim. Don't: a soft wraparound light with no shadow at all. |
Every rule carries a decision, a number, a test a checker can apply to the picture alone, and a tolerance.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
- MidjourneySubscription
Generates images from a written description, with the strongest default aesthetic of anything in the category.
Bad at That default aesthetic is the problem as often as it is the draw. It has a house look, it pulls everything towards it, and getting a plain honest photograph of a real product out of it is harder than it should be. Text inside an image is unreliable.
- GPT ImagePay per use
Generates and edits images with unusually good instruction-following, and can hold text inside a picture better than most.
Bad at Faces drift between generations unless you feed a reference every time. Skin can come out plasticky at default settings, and a heavy fidelity setting costs many times more and takes many times longer.
- Adobe FireflyFree tier, paid above it
Adobe's generative models, trained on licensed and public-domain material and wired into Photoshop and Illustrator.
Bad at Raw image quality trails the leaders. What you are buying is the indemnity and the fact that it sits inside files you already have.
- Google AI StudioFree tier, paid above it
Where Gemini API keys, projects, billing and spend caps are made, with a playground for the models.
Bad at Free-tier prompts are used to improve Google's products, and the billing screens changed several times in 2026.
- ColorCheckerSubscription
A printed chart shot at the start of a session so colour can be made objectively correct later.
Bad at Only works if somebody remembers to shoot it. Fades with age and must be replaced.
Why it matters
Most art direction arrives as mood words and a folder of pictures. The photographer copies the wrong picture, the image tool redraws the label, the designer sets type across the product, and nobody can say what went wrong because nothing was written down to check against. This skill makes Claude turn each word into a number and each reference into one named quality, so a stranger can make the work and a second stranger can judge it.
- Type
- technique
- Period
- ongoing