Two Months, a Batch of AI Landing Pages, and the Comments Say They All Look the Same

Two months, a batch of landing pages, and the same controversy.

Designer Elaya said in a tweet that over the past two months, using Higgsfield, Claude, and Cursor, plus an AI designer skill they wrote themselves, they made a batch of landing pages. The skill has been open-sourced. The first thing you see in the video is a blue sea, a clifftop town, and a church bell tower; in the upper left of the page are Metricra and social icons, and on the right is a four-column navigation. It looks good and consistent—and it immediately drew a comment: they look like every AI home page that is over-designed and misses the point.

This controversy is more interesting than the tool itself.

The video shows the visual of one of the pages: a seascape illustration as the base, with navigation and content laid over the image. Elaya made more than one page like this in two months.

The skill is, plainly put, a Markdown rules file. Tools like Claude Code, Cursor, Codex, and Windsurf can read project rules and generate pages according to them. The repo's main landing-page-design covers the process of building a landing page from scratch: first ask intake questions, then define page structure and conversion copy, and finally provide a visual system, including typography, spacing, border radius, and motion.

It solves the most common problem when AI writes frontend: it can generate, but the default styling is flat and lacks a stable set of design constraints. Writing the specs into a rules file makes the output much more complete. The cost is right there too: the more fixed the spec, the easier it is for pages to look like the same template.

Someone in the comments asked whether intake questions actually change the output, or whether the visual system is carrying the result. Elaya did not elaborate. This question is key.

Installation is not complicated. For Claude Code, copy the skill directory to .claude/skills:

mkdir -p .claude/skills
git clone https://github.com/elayadesign/ai-design-skills.git /tmp/ai-design-skills
cp -r /tmp/ai-design-skills/skills/<skill-name> .claude/skills/

For Cursor, pull SKILL.md into .cursor/rules:

mkdir -p .cursor/rules
curl -o .cursor/rules/<skill-name>.md https://raw.githubusercontent.com/elayadesign/ai-design-skills/main/skills/<skill-name>/SKILL.md

For other tools, copy SKILL.md into the corresponding rules file: Codex uses AGENTS.md, Windsurf uses .windsurfrules, Cline uses .clinerules, and the general approach is to paste it into the system prompt. Claude.ai has no terminal, so upload it to a Project, or attach it in the chat and say follow this. If you want to adapt it to your own style, fork the repo and change the design values at the bottom of the file—typography, colors, spacing, and motion are all there.

The controversy is also worth laying out. Some users say these pages have almost the same hero and only swap the background; some say it is essentially a pretty video plus text; others remind people to add a reduced-motion version—after turning off blur and slide-in, the content should appear immediately and the page should still look finished. Elaya replied that Gemini can also use it, but that they hate Gemini and recommend Claude, Cursor, and Codex.

Putting these comments together, the value and boundaries of this skill are clear. It can lock down the design process and produce stable, complete, shippable output; it can also easily lock aesthetics into a template. For indie developers, it first solves going from nothing to something; for product teams, the next problem is going from having something to not all being the same.

Repo: https://github.com/elayadesign/ai-design-skills

Portfolio: https://elaya.design/

Original tweet: https://x.com/elayadesigns/status/2114049200749175174