Recommended
Writes prompts for image models (FLUX, Stable Diffusion, ChatGPT images, Gemini and the like) that give a website, a blog or a product one recognisable style and come back without garbled letters. The scene comes first and the site's style recipe follows it word for word; absence is described positively instead of "no text, no watermark", which in a single prompt box summons those very things; anything that would need readable writing (a price list, a chart, a date) becomes a metaphor without characters; the scene ends with a closed list of what is in it; people and hands, the usual weak spots, are shown through objects; the aspect ratio is stated in numbers; and several pictures become several prompts, never one. Use when you write or fix prompts for hero images, blog headers, cards or illustrations, when generated pictures come back with pseudo-letters, cluttered scenes or a different style each time, or when a brief arrives in another language.
Image Prompts in One Style: Clean Pictures for a Site is a tested SKILL.md that writes prompts for image models (FLUX, Stable Diffusion, ChatGPT images, Gemini and the like) that give a website, a blog or a product one recognisable style and come back without garbled letters; an agent buys it once for $0.05 over x402.
Generating the pictures themselves, editing or retouching existing images, logos or typography that must carry real text, photo-realistic portraits of real people, or choosing a brand's colours from scratch. It writes and fixes the prompts you send to an image tool and keeps every picture in the style you already have.
Tested 2026-10-09 with a strong and a weak model.
| Sonnet | Haiku | |||
|---|---|---|---|---|
| with | without | with | without | |
| Prompts right on every check (12 prompts) | 12/12 | 0/12 | 9/12 | 0/12 |
| Prompts free of the defects alone (12 prompts) | 12/12 | 1/12 | 9/12 | 1/12 |
Same request on both sides. The second row leaves out the three structure checks the request never states (scene before style, closed list, frame in numbers) and counts only defects: negations, words that summon writing, faces, collages, language and the exact style; the price follows that stricter row. Read by hand, Sonnet without the skill wrote careful, vivid prompts and kept the house style exactly, but in 10 of 12 it answered the request for a clean picture with a list of bans (no text, no letters, no logos, no watermark), which a one-box tool reads as a request for those things; it also named the menu, the dashboard, the street name and the time on the clock, drew the smiling agent and the laughing hikers with their faces, and put the style before the scene every time. Haiku without the skill made the same moves.
Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.
Our own test text, before and after the skill ran. Excerpts only.
Before
Hero image for our restaurant's website: the chalkboard menu next to the entrance, listing today's dishes and their prices.
After
A soft clay standing board with a rounded frame leans beside an arched restaurant doorway, seen straight on at eye level and centred, with generous empty space around it. On the board face, rows of short rounded clay bars of different lengths sit slightly raised, each row ending in a small round dot, and a single bar in vivid teal is the focal point. Wide landscape, 3:2, 1536x1024.…
Before
Снимка за статия „Как се пише фактура“: фактура на бюро, на която се вижда крайната сума, а до нея химикалка.
After
A single sheet of paper lying flat on a sand-coloured desk, seen from slightly above and centred, with a pen resting diagonally beside it. The sheet carries rows of short navy bars of different lengths stacked down its face, and at the bottom a wide rounded coral block glows as the one accent, drawing the eye to the end of the rows. The pen has a navy body and a coral tip.…
Any language. Tried in: English, Bulgarian.
Perpetual, non-exclusive; use and modify for yourself incl. paid work; no resale or republishing. Holder: Georgi Kalchev, aiskills402.com. Full terms.
Current version 1.0.3, updated 2026-10-09. Whoever bought an earlier version gets new ones free through the same re-download token.
v1.0.3 · 2026-10-09
- Picture probe in FLUX schnell (Cloudflare Workers AI), same three trap briefs: the plain prompts gave a menu board covered in pseudo-letters although they asked for no letters, an invoice amount with wrong figures and a road sign without its house number; the guided prompts gave three clean pictures. ChatGPT had drawn both sides cleanly. Price set by the owner to $0.05 and the Recommended badge added on that evidence. One run per picture.
v1.0.2 · 2026-10-09
- Price set by the owner to $0.03 and no Recommended badge, after a picture probe: three trap briefs sent to ChatGPT with the plain and the guided prompt gave six pictures without garbled letters (ChatGPT wrote the requested amount and street name cleanly); the difference was stray objects in the plain versions. The probe ran in an account whose memory holds the owner's own rules for pictures, so it is not neutral; FLUX was not tried. Card and test note say so. The skill text is unchanged.
v1.0.0 · 2026-10-09
First release: writes prompts for image models that keep a site's pictures in one style and free of garbled writing. The scene comes first and the house style recipe follows word for word; absence is described positively instead of with "no" in a single prompt box, while a separate negative field takes the unwanted things; anything that needs readable writing becomes a metaphor and objects that carry writing are left out or given a blank face; each scene ends with a closed list; people and hands are shown through objects; the frame is stated in numbers; several pictures become several complete prompts; prompts are in English whatever the language of the brief. Twelve cases; test/control.mjs proves right prompts pass and wrong ones fail. Before the run, the meeting case was replaced by a hiking group because the skill's own example (cups and a laptop at a table) was its answer, and two words that matched case checks exactly were taken out of the skill. No model run yet: the baseline, the final price and the sentence on what Sonnet gains are still to be written. Price 10000 is provisional.
Usually because the prompt asks for them, often without meaning to. A sign, a menu board, a document or a dial invites writing, and in a tool with one prompt box a line like 'no text, no logo' is read as a request for text and logos too. The skill turns such objects into shapes and light, gives the unavoidable ones a blank face and describes clean surfaces positively, so there is nothing for the model to spell.
Your style recipe, one paragraph about medium, light, background, palette and the single accent colour, goes into every prompt word for word and after the scene, never paraphrased. Each prompt ends its scene with a closed list of what is in it, states the frame in numbers, and several pictures become several prompts, each one complete, so nothing turns into a collage or drifts between images.
Markedly, on every trap. We gave Sonnet and Haiku twelve image briefs, ten of them built around a trap such as a menu board, an invoice total, a smiling support agent or a user asking for no logo, and code checked each prompt. Left alone, Sonnet wrote rich prompts but answered a request for a clean picture with lists of bans in ten of them, named the writing it meant to avoid and drew faces; with the file loaded, all twelve came out clean. Counting the defects alone, it went from one to twelve; Haiku from one to nine.
In FLUX, clearly; in ChatGPT, not visibly. On 9 October 2026 we sent three trap briefs (a menu board, an invoice total, a road sign) to both tools, once with the prompt written without the file and once with it. In FLUX schnell the plain prompts gave a board covered in pseudo-letters although they asked for none, an invoice amount with the wrong figures and a sign missing its house number; the guided prompts gave three clean pictures. ChatGPT drew all twelve of its pictures cleanly either way, writing a requested amount or street name correctly. One run per picture: a probe, not a benchmark.
SEO & Content
Writes the SEO title, meta description and (on request) the H1 heading for one web page, in the page's own language, inside hard character limits, matching what the searcher wants. Use when asked to write or rewrite a page title, meta description, SEO snippet or H1 for a page, product, service, article or landing page.
Code & Engineering
Reviews a pasted image pipeline (upload or AI generation, storage in R2, resizing, serving through Cloudflare image transformations, a Next.js loader, a sharp or WebP script) for the traps that quietly ship huge or broken images or spend a shared quota. It flags photographs stored as PNG, files stored exactly as a generator returned them, a redirect-on-error setting that hides an exhausted transformation quota, a list of widths that multiplies the unique-transformation count, widths above the original, transformations "turned off" while the images binding or the zone setting stays on, lossy WebP on charts and infographics, a job that re-encodes WebP that is already lossy, a preview subdomain mistaken for a zone, an original overwritten by a backdated fix, and self-made variants that can 404. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use to review an image pipeline, Cloudflare image transformations or WebP handling before it ships.
Writing & Rewriting
Rewrites AI-sounding prose so it reads like a person wrote it — in any language, without dropping facts. Use when asked to humanize a text, make it sound less AI-generated or less like ChatGPT, or clean machine phrasing out of an article, post, email, product description or page.