SuperAPI AI Directory

mengtofrontend

Frontend & websites

View repository

Updated 10-15 23:44

About this skill

MengToFrontend

Overview

Use this skill as a pre-launch quality pass for landing pages. The principle is: AI can generate the first draft, but a professional page needs typography judgment, real references, visual cleanup, and one or two deliberate human touches.

Source inspiration: Meng To's X tutorial, "Design Skills: How to Avoid AI Slop in Landing Pages" (https://x.com/MengTo/status/2062484065748701429).

Workflow

  1. Inspect the page as a user would: desktop and mobile first viewport, then the full page.
  2. Identify visible "AI slop" signals before editing: generic layout, mismatched images, unnatural text spacing, overused gradients, inconsistent shadows, vague copy, or decorative details without product relevance.
  3. Fix the highest-impact issues in this order: typography, real visual references, image quality, layout hierarchy, micro-interactions, final originality.
  4. Verify the final result visually with screenshots or browser inspection when a local app or URL is available.
  5. Return a concise launch-readiness checklist with remaining risks.

Typography

  • Check letter-spacing on headings, labels, buttons, nav items, and all-caps text.
  • Avoid negative tracking on normal body text and compact UI labels unless the existing brand system clearly uses it.
  • Use tighter tracking only when the font and size support it; inspect at mobile sizes because AI-generated pages often collapse into cramped text there.
  • Test at least two plausible font directions when typography feels generic: one neutral/productive option and one brand-specific option.
  • Match type scale to context. Avoid hero-sized type inside compact cards, sidebars, dashboards, and small panels.
  • Prefer fewer font weights with clear hierarchy over many arbitrary weights.

Real References

  • Ask for or gather real image references when visual authenticity matters: product photos, venue photos, competitor screenshots, brand references, material textures, UI examples, or founder-provided assets.
  • Prefer real product/place/person imagery over atmospheric stock-like art.
  • If generated images are needed, ground prompts in concrete references and inspect the result for artifacts before shipping.
  • Do not let decorative AI visuals replace proof: screenshots, product state, customer workflow, actual object, or real environment should be visible when relevant.

Image Refinement

  • Treat AI images as draft assets. Check hands, faces, logos, text, product geometry, lighting direction, shadow contact, depth, and cropping.
  • Add practical refinements where possible: replace fake texture with real texture, correct lighting mismatch, remove artifacts, simplify busy backgrounds, and align image contrast with surrounding UI.
  • Make image crops intentional. Avoid dark, blurred, or overly cropped media when users need to inspect the subject.
  • Keep visual style consistent across all images. Mixed photo styles, icon styles, and illustration styles create an obvious generated feel.

Prompt Quality

  • Use specific prompts for brand/product names, audience, material, camera angle, lighting, mood, and layout role.
  • Include negative constraints for common failure modes: no fake text, no warped logo, no extra fingers, no distorted interface, no illegible labels.
  • For landing pages, prompt for the actual business subject instead of abstract "modern SaaS" visuals.
  • Keep generated visual prompts tied to page sections: hero proof image, feature detail, testimonial context, or product state.

Human Touch

  • Add one or two custom details that connect to the product, not random decoration.
  • Use subtle micro-interactions only where they clarify state or make the interface feel responsive: hover, focus, reveal, active tab, form validation, or a tasteful hero motion.
  • Replace generic copy with concrete claims, audience language, and product-specific nouns.
  • Ensure icons, dividers, badges, and background treatments serve hierarchy or comprehension.

Audit Output

When auditing, report findings in this format:

Launch readiness: [Ready / Needs polish / Not ready]

Top fixes:
1. [Issue] - [specific correction]
2. [Issue] - [specific correction]
3. [Issue] - [specific correction]

Checklist:
- Typography and letter spacing: [pass/fail + note]
- Real references and imagery: [pass/fail + note]
- Image refinement: [pass/fail + note]
- Prompt/design specificity: [pass/fail + note]
- Human touch and micro-interactions: [pass/fail + note]
- Mobile visual quality: [pass/fail + note]

When implementing fixes, edit the page directly and then provide a short summary plus verification performed.