/img-to-frontend

name: img-to-frontend
description: "Use when the user wants an end-to-end premium visual design workflow for any type of website or web page created directly from images."

Img to Frontend

Overview

Use this skill to run an image-first design-to-code workflow for visually appealing awwward winning websites. It must adapt to the user's brief, audience, content model, interaction needs, and brand intent rather than forcing a fixed category or visual trope.

The workflow is intentionally staged:

1. Invoke `$imagegen` to create four distinct website/page design images. 2. Stop so the user can select the image or images to use as the design reference. 3. Translate only the selected image into a precise implementation prompt. 4. Build the real page and iterate against screenshots until structure, hierarchy, typography, spacing, color, media treatment, component geometry, interactions, and responsive behavior are close to the reference.

Load `references/visual-iteration-checklist.md` for design variety, implementation, close visual replication, premium website work, complex UI sections, or repeated visual refinement.

When To Use

Use this skill when the user asks to:

• Create several distinct website or page concepts before coding.
• Turn an image or generated design into a real frontend page.
• Produce a detailed prompt another coding agent can use to build a design.
• Replicate a screenshot as real HTML/CSS/React rather than as an image.
• Iterate on visual details like spacing, scale, typography, media treatment, visual systems, component alignment, interaction states, or responsive breakpoints.

Do not use this skill for ordinary business-logic changes, backend features, simple copy edits, or design critique that will not lead to generated images, a detailed implementation prompt, or frontend code.

Non-Negotiable Image Gate

The first deliverable is always four generated images. Do not jump straight to text prompts, implementation prompts, or code.

Required behavior:

• Load and use the `$imagegen` skill for phase 1.
• Generate four unique website/page images with the built-in image generation tool by default.
• Make one image-generation call per concept unless the image tool supports equivalent multi-output generation in the current environment.
• Render or present all four images to the user, one at a time.
• Stop after presenting the images and ask which one the user wants to explore.
• Do not write the detailed build prompt until the user selects a specific image.
• Do not implement frontend code until the selected-image prompt phase is complete or the user explicitly skips it.

If image generation is unavailable or fails, report that blocker and ask whether the user wants text-only fallbacks. Do not silently substitute text briefs for the image stage.

Core Workflow

1. Design Exploration

Start by identifying the site context, audience, goal, style constraints, viewport, and must-have sections. If the user already gave enough context, do not pause for more input.

Generate four distinct website/page images through `$imagegen`. Each image must have its own visual language, not minor palette swaps. Vary structure, hierarchy, typography strategy, interaction model, information architecture, content emphasis, density, and brand behavior.

Choose directions that fit the site's purpose and audience. Different website types should develop from their own communication needs, not from a shared default template.

Use `ui-mockup` as the image-generation use case unless the user asks for a different raster style. The images should be polished enough for the user to choose between real visual directions, not rough wireframes.

For each direction, briefly label:

• Overall art direction and emotional tone.
• Page layout and content hierarchy.
• Typography treatment and scale relationship.
• Color palette with exact roles.
• Visual centerpiece appropriate to the site's purpose, content, and user task.
• Credibility, context, or trust treatment when relevant.
• Motion idea, if appropriate.
• What makes it meaningfully different from the other three.

2. Selection Pause

After the four images are created, stop and wait for the user to pick one or request revisions. Do not write the detailed build prompt and do not begin implementation until the user selects an image or explicitly asks to skip selection.

When the user references a selected image, verify which image they mean from the conversation context. If there is ambiguity, ask a short clarification before writing the prompt or coding.

3. Deep Build Prompt

Before coding, convert the selected image into a granular implementation prompt. The prompt should be specific enough that another frontend agent can build a near exact replica without guessing.

Include:

• Canvas and viewport assumptions, including desktop target size and responsive requirements.
• Exact page structure, grid proportions, maximum widths, gutters, and alignment rules.
• Typography hierarchy, font choices, letter spacing, line-height, weights, italics, uppercase treatments, and no-wrap constraints.
• Color tokens for background, surfaces, borders, primary accents, muted text, active states, status states, shadows, and media treatments.
• Component inventory covering only the modules that actually appear in the selected design, described by their role, hierarchy, layout behavior, and content type.
• Detailed behavior for any complex visual systems present in the selected image, including their structure, anchors, scale rules, states, responsive behavior, and fallback behavior.
• Animation rules that are restrained and purposeful.
• Explicit "do not" constraints that prevent fake implementation, generic styling, broken geometry, unreadable content, and overlapping responsive modules.
• Acceptance criteria and screenshot comparison checklist.

4. Codebase Discovery

Inspect the project before editing. Determine the framework, routing mo