Elite Frontend Image Direction Skill for Muse
Generate frontend web UI through an image-generation workflow.
- What
- Generate frontend web UI through an image-generation workflow.
- Cost
- Free
- Needs
- none — no account, API key, or paid plan required. (Using the skill later needs a Muse environment that can generate images — one image call per section, so a full page can mean 6–12 image calls and consumes generation budget; installing it needs no image generation at all.)
- Install
- Copy the installer prompt below into your Muse — your agent does the rest.
Frontend art direction for image generation with one ironclad rule: one separate horizontal image per website section, never collapsed into a single tall slice. Enforces a combinatorial variation engine (theme paradigm, background mode, typography character, hero architecture, section system), brief-to-direction mapping (minimalist / editorial / cinematic / SaaS / agency / e-commerce), conversion-aware structure, one disciplined palette across all frames, and an extensive anti-AI-slop checklist (no default purple gradients, no cloned left-text/right-image heroes, no fake KPI strips). Each generated image is meant to be an implementation-friendly reference a developer or coding model can build from. Discovered via skills.sh. Honest note: you need a Muse environment that can generate images; generating a full 8-section page means 8 image calls, which consumes generation budget — and these are design comps, not production code. Skill Harbor never reviews the code, review it yourself before use. Not verified.
Curated by Skill Harbor — free, open-source skill (MIT) by @leonxlnx.
Version:
Quick answer
Elite Frontend Image Direction is a free Muse skill that directs website imagery. It uses one horizontal image per section, never one tall slice reused, with a combinatorial variation engine and an anti-AI-slop checklist, producing design comps rather than production code. Image calls consume generation budget, so each variation has a real cost though the skill itself is free.
Install
Copy the install package below, then paste it into Muse🧪 Install ran in our test sandbox · Oct 19, 2026📌 Pinned version: commit beb16b0MOVED — READ THIS FIRST: this skill has moved repositories. It used to live in leonxlnx/taste-skill; upstream it now lives in Leonxlnx/imagegen-skills. In taste-skill, all that remains at skills/imagegen-frontend-web/SKILL.md is a 459-character stub saying "Moved to github.com/Leonxlnx/imagegen-skills"; that stub carries disable-model-invocation: true (it runs only when called by name), the install name is unchanged ("imagegen-frontend-web"), and the stub gives the install command npx skills add https://github.com/Leonxlnx/imagegen-skills --skill "imagegen-frontend-web". This prompt follows that direction by direct pinned fetch from the new repository instead of running npx, so that the exact tested commit and the fingerprint below are verified before anything is installed. Step 0, before installing anything: read this install prompt, summarize in a few lines what this build does and what installing it involves — including the repository move above — then ask me to confirm before you proceed. Prerequisites: none — no account, API key, or paid plan required. (Using the skill later needs a Muse environment that can generate images — one image call per section, so a full page can mean 6–12 image calls and consumes generation budget; installing it needs no image generation at all.) Install "imagegen-frontend-web" for me — an elite frontend image-direction skill by @leonxlnx. It generates premium, conversion-aware website design reference images (landing pages, marketing sites and product comps) that a developer or a coding model can accurately recreate; these are design comps, not production code. Its hard output rule comes first: generate ONE separate horizontal image (16:9, 16:10 or 21:9) FOR EVERY section, always, no exceptions — 1 section = 1 image, 4 sections = 4, 8 sections = 8, 12 sections = 12; a "landing page" with no count defaults to 6 sections, a "full website template" to 8; never combine several sections into one frame or one tall slice, and if you can only render one image at a time, output them sequentially in the same response, announced one by one ("Section 1 of 8: Hero", "Section 2 of 8: Trust bar", …), until every section has its own image. Before anything else it also fights the most overused AI hero pattern — left-text / right-image is allowed but must not be the first instinct; prefer centered over a background image, bottom-left or bottom-right over image, top-left lead, stacked center, image-as-canvas, off-grid editorial, mini minimalist, or the inverted classic, and use the classic only when it is genuinely the strongest choice. Its baseline dials are DESIGN_VARIANCE 8, VISUAL_DENSITY 4, ART_DIRECTION 8, IMPLEMENTATION_CLARITY 9, IMAGE_USAGE_PRIORITY 9, SPACING_GENEROSITY 8, LAYOUT_VARIATION 8 and CONVERSION_DISCIPLINE 8, but the user's brief always overrides them ("clean" lowers density, "crazy creative" raises variance and art direction, "premium SaaS" keeps clarity high, "editorial" allows stronger type and asymmetry); a brief-to-direction mapping covers minimalist / clean / typography-only / swiss, editorial / magazine / fashion, cinematic / atmospheric / premium / luxury / bold, SaaS / product / dashboard / fintech / infra, agency / creative studio / portfolio, and e-commerce / shop / store briefs, each with its own hero scale, background mode, gradient stance and composition, and restraint is never forced out of a brief that asks for atmosphere, nor atmosphere into one that asks for restraint. A combinatorial variation engine picks, per site: 1 theme paradigm (pristine light, deep dark, bold studio solid, quiet premium neutral), 1 background character, 1 typography character (six options, from Swiss rational sans to elegant editorial serif + sans), 1 hero architecture (six options), 1 dominant section system (six options), exactly 4 signature components from a list of 12 (diagonal staggered square masonry, 3D cascading card deck, hover-accordion slices, gapless bento, brand marquee, polaroid arc, vertical rhythm lines, off-grid editorial, product UI panel stack, testimonial quote wall, oversized metrics strip, layered image crops), exactly 2 motion-implied languages, 1 hero scale (giant statement, mid editorial, mini minimalist), 1 narrative / concept spine (artifact, journey, tool, living system, stage, archive) and exactly 1 second-read moment; per section it picks a composition anchor (at least 3 different anchors across the site, the left-third-caption classic sparingly and never twice in a row), a background mode (12 modes, varied across the page, with full-bleed image backgrounds used confidently when the brief allows) and a CTA variation (varied at least once, the primary action always unmistakable). Around that: a frontend reference rule (every image must clearly communicate layout, hierarchy, spacing, typography scale, rhythm, CTA priority, component styling and image treatment), hero minimalism rules (a strong opening scene, a short powerful headline of usually 5–10 words, concise support text, negative space and contrast, no stuffed pills / fake stats / badges), a creativity escalation rule (actively push at least 3 of composition, typography, scale contrast, hero concept, image treatment, section rhythm, framing, visual tension and layout structure — intentional, not chaotic), image-first art direction (imagery is a core material, including as backgrounds, not decoration), extensive anti-AI-slop rules (no cloned left-text/right-image blocks, no default purple/blue AI gradients, glowing edges, blobs or glassmorphism without reason, no gradient headline as a shortcut for premium, banned generic copy — unleash, elevate, revolutionize, next-gen, seamless, powerful solution, transformative platform — and banned fake brands such as Acme, Nexus, Flowbit, Quantumly, NovaCore, no fake KPI strips or pointless dashboards, no unreadable logo marquees, no over-packed sections), typography-first discipline, section rhythm and density / spacing discipline (the page breathes, section spacing stays even and generous, denser sections alternate with calmer ones), colour and material rules (ONE controlled palette across the whole site — 1 primary, 1 secondary, 1 sparing accent, a neutral scale; gradients are allowed when low-chroma, palette-matched and professional, including tonal grades, vignettes and noise-textured depth, and banned when rainbow / mesh-blob, purple-to-blue or pink-to-orange slop; full-bleed images must tonally match the palette with overlays keeping text readable), image / media direction, default site packs (4 sections: hero, features, social proof, CTA; 8 sections: hero, trust bar, features, product showcase, benefits, testimonials, pricing, CTA; 12 sections: those plus feature grid, product preview, problem / solution, workflow, metrics / proof / integration and FAQ, closing on CTA + footer), a multi-image consistency rule (same palette, type scale, spacing, CTA family, image treatment and tonal voice across all frames, while anchors, background modes and section sizes vary), an 18th-section extra edge (cross-section foreground/background contrast varied at least twice, one unmistakable primary action per tier, at least two distinct image crops in multi-section sites, data-viz only where the site type needs it, cultural / tonal alignment with a named industry or region, mobile-implied fidelity, a real conversion path — hook, proof, action — and a composition-variety check that rejects a set whose same anchor repeats more than 2 sections in a row or whose same background mode repeats more than 3, and that, for non-minimalist briefs, pushes for at least one full-bleed / duotone / atmospheric background and at least one mini minimalist section), a 21-point clarity check run before finalizing (hierarchy obvious, hero clean, no AI tells, premium not template-like, codeable, frames belonging together, imagery strong and varied, breathing spacing, exactly one second-read moment, anchors and background modes mixed, hero scale clean, conversion path clear, palette consistent, every image horizontal and one-section-only, total image count equal to the section count, hero not a reflexive left-text / right-image default — if the count is wrong, regenerate the missing sections), and a response behaviour: infer the site type and conversion goal, infer the section count (defaults above), commit out loud to it ("Generating N horizontal images, one per section"), plan and choose as above, lock the palette, apply the extra edge and the clarity check, then generate every per-section horizontal image, labelled, until the full set is delivered — do not stop early, do not summarize, do not return only one image, and do not ask unnecessary follow-up questions when a strong interpretation is possible. Repository: https://raw.githubusercontent.com/Leonxlnx/imagegen-skills/beb16b0d0f4dae9a6f5cb04ea41e97b93a54ab8e/skills/imagegen-frontend-web/SKILL.md (the skill is the single file skills/imagegen-frontend-web/SKILL.md at that commit in the Leonxlnx/imagegen-skills repository: no helper files, no scripts, no references subfolder) Pinned version: install the code at commit beb16b0d0f4dae9a6f5cb04ea41e97b93a54ab8e exactly (Skill Harbor tested this version). Do not substitute a newer version. If this commit is unreachable, stop and tell me. [FINGERPRINTS] SKILL.md at the pinned commit: SHA-256 6b5c2256522fdba1e3313eafa3d743b960a7b13ec029d3e6586e04b23327c1f8, 36,854 bytes (36,780 characters). If the fetched file does not match, it changed since the review: stop, tell me what you found and let me decide. Updated October 19, 2026: Skill Harbor regenerated this install prompt because the skill moved repositories. At the previous pinned commit (717446e07a78d4e6d7918b1fde1a263378eebac9, in leonxlnx/taste-skill) the skill was fetched from taste-skill; it now lives in Leonxlnx/imagegen-skills at the new pinned commit above, and in taste-skill the old path holds only the 459-character move stub described at the top of this prompt (disable-model-invocation: true on that stub, so it runs only when called by name; install name "imagegen-frontend-web" unchanged; stub's install command: npx skills add https://github.com/Leonxlnx/imagegen-skills --skill "imagegen-frontend-web" — this prompt reaches the same result by direct pinned fetch, with the fingerprint verified, instead of running npx). Corrections: the previous prompt pointed at the taste-skill repository and told you to fetch SKILL.md "and any helper files" from it — the new skill folder holds SKILL.md alone, with no helper files, scripts or references to fetch. The previous prompt's safety boilerplate also said the repository should contain zero secrets in code, credentials only via the secure vault, and allowed hosts declared in SKILL.md — the new file declares none of that: it contains no URL, no script, no credential, no environment variable and no vault; verify that from the file itself, as step 2 below asks. What the skill does at use time is stated before installation, in step 2: nothing is sent anywhere during installation itself, and at use time the file prescribes no network call and no named external service at all — it directs your own host environment's image generation, working from the brief you give it. 1. Fetch skills/imagegen-frontend-web/SKILL.md at the pinned commit, into a temporary folder, then compare it with the fingerprint from Skill Harbor's review of 2026-10-19 (see [FINGERPRINTS] above). If it does not match, it changed since the review: do not install it yet, tell me what you found and let me decide. 2. Safety check: read SKILL.md and review it for anything suspicious (unexpected network calls, shell commands, credential harvesting, code that runs at install); STOP on any red flag and tell me. Tell me in plain words what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Confirm these points from the file: - Nothing is sent anywhere during installation itself, and no code runs during installation: the skill is a single Markdown instruction file (SKILL.md, 36,854 bytes) — no scripts, no executable code, no helper files. - At use time the file prescribes no network call, names no external service, API or host, and contains no URL at all: it instructs the agent to generate the design images with the host environment's own image-generation capability, in the response, from the brief the user gives it — one separate horizontal image per section, labelled "Section X of N: <name>", generated sequentially in the same response if only one can be rendered at a time, with the section count committed out loud first ("Generating N horizontal images, one per section"; defaults: landing page 6, full website 8). Do not put personal information in a brief you would not want reflected in generated design comps and their short copy. - No file reads credentials, API keys or personal information, and nothing asks for any: the file's frontmatter holds only name ("imagegen-frontend-web") and a description — no disable-model-invocation field in this new file itself (that flag is on the old-repository stub), no allowed hosts, no environment variables, no accounts. - The output discipline to confirm from the file: the total number of images must equal the number of sections, every image is horizontal and one-section-only, one palette stays locked across all frames with at least 3 different composition anchors across the site, the hero must not default to left-text / right-image out of habit, and the 21-point clarity check runs before the set is delivered. 3. When I say "yes, install", install it as a skill: copy SKILL.md into the agent's skills directory, in a folder named "imagegen-frontend-web" (the install name is unchanged from the old repository). 4. Verify with no network calls: frontmatter valid, files in place (1 file). 5. Report what was installed, where, and what I still need to do myself (e.g. start a fresh session so the host re-indexes the skill; confirm my environment can generate images; bring a site brief — type of site, brand, section count if I have one in mind, otherwise the file's defaults apply; remember a full page means that many image-generation calls). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.
Saved to your recent installs. Find it anytime on /connect.
🔒 Security, provenance and what this touches
The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?
Verification certificate
Dated facts about what was checked for this listing, and how. This is not a safety certification.
- Pinned version
- beb16b0 · 2026-10-19
- Sandbox install test
- Install ran in our test sandbox · Oct 19, 2026
- Creator identity
- Not verified
- Fingerprint watch
- Last checked · Oct 19, 2026 : no change detected in the files linked by this install prompt
- Install prompt analysis
- 2026-10-13 : No commands, network calls, packages, file writes or API keys detected.
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
Questions
How do I install a skill?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.