ui-imagegen-director

name

ui-imagegen-director

description

Direct high-quality frontend UI work by using Codex Image Gen for visual mockups, then implementing the selected direction in code with browser-based visual QA. Use when the user asks for a better-looking web UI, UI mockup, app screen, dashboard, landing page, visual redesign, or wants Image Gen/GPT Image 2 to guide frontend implementation. Avoid for backend-only work or tiny style tweaks that do not need visual direction.

UI ImageGen Director

Use this workflow when visual direction is the weak link: the user wants a polished frontend, Codex is producing generic UI, or a generated mockup would give the implementation a stronger target.

This skill composes three activities:

1. Create or refine a UI direction with Image Gen.
2. Translate the selected direction into real frontend code.
3. Verify the rendered result in a browser and iterate visually.

Workflow

1. Inspect the product context before designing:

   • audience and job-to-be-done
   • screen type: dashboard, settings, editor, commerce, landing page, game, etc.
   • existing framework, design system, component library, and constraints
   • whether the page is an operational tool, expressive site, or immersive/game UI

2. Pick one clear visual direction before calling Image Gen.

   • Prefer a concrete direction such as refined B2B operations, editorial analytics, luxury fintech, industrial console, playful maker tool, scientific workspace, or focused commerce catalog.
   • Treat boldness as intentionality, not visual noise. Dense operational tools can be excellent without becoming decorative.
   • Decide the memorable design move: typography, information density, layout rhythm, product imagery, surface treatment, motion, or interaction model.

3. Write an Image Gen prompt that is implementation-oriented.

   • Ask for a UI screenshot or screen mockup, not a marketing poster, unless the user explicitly wants marketing.
   • Include layout regions, expected content density, states, and constraints.
   • Prohibit generic AI UI patterns: default SaaS cards everywhere, purple-blue gradient dominance, meaningless decorative blobs, fake glass panels, ungrounded icons, oversized hero sections for tools, and placeholder text that hides hierarchy.
   • Ask for visual clarity that can be translated into components: spacing, type hierarchy, navigation, controls, table/list structure, empty/error/loading states, and responsive intent.

4. Use Image Gen when a new visual target or raster asset is needed.

   • Invoke the built-in image generation path, or explicitly use $imagegen when asking another agent or writing a reusable prompt.
   • For project-bound assets, keep generated images in the project and wire references to project-local files.
   • Do not use a generated screenshot as the final UI. It is a reference unless the user asked for a bitmap asset.

5. Critique the generated direction before coding.

   • Check whether it fits the product, audience, and workflow.
   • Check implementation feasibility: component structure, responsive behavior, text density, asset needs, animation cost, and accessibility.
   • If it misses the intent, do one targeted Image Gen revision instead of drifting through many unrelated variants.

6. Implement the chosen direction in code.

   • Reuse existing components, tokens, routing, state, and data patterns.
   • Encode the visual direction with real layout, typography, CSS variables/design tokens, states, and responsive constraints.
   • Use generated bitmap assets only where they are the intended asset: hero imagery, backgrounds, illustrations, product imagery, textures, or sprites.
   • Keep operational tools efficient and scannable. Avoid landing-page composition for dashboards, CRMs, editors, and admin tools.

7. Verify visually in a browser after frontend changes.

   • Start or reuse the dev server.
   • Open the relevant route with the Browser plugin or Playwright when available.
   • Check desktop and mobile widths when layout is responsive.
   • Look for text overflow, overlapping UI, unstable dimensions, broken assets, inaccessible contrast, blank canvases, and mismatch with the selected direction.
   • Iterate in code until the rendered UI is coherent, not merely until tests pass.

8. Prepare the final report.

Final Report

Include:

• chosen visual direction
• generated asset or mockup paths, if any
• implementation files changed
• browser/device checks performed
• remaining visual risks or follow-up work

Prompt Template

Use this shape when asking Image Gen for the visual target:

$imagegen
Create an implementation-oriented UI mockup for <product/screen>.

Audience and job: <who uses it and what they need to accomplish>
Screen type: <dashboard/settings/editor/landing page/etc.>
Visual direction: <one concrete aesthetic direction>
Memorable design move: <typography/layout/density/media/motion/etc.>
Required regions: <navigation/header/main content/actions/states>
Content density: <sparse/balanced/dense> with realistic labels and data hierarchy
Implementation constraints: design must translate into web components and responsive CSS
Avoid: generic SaaS card walls, purple-blue gradient dominance, decorative blobs, fake glass, unreadable tiny text, marketing hero layout unless requested
Output: high-quality UI screenshot-style mockup, clear hierarchy, realistic spacing, no watermark

Then implement from the selected mockup:

Use the selected UI mockup as visual reference. Do not paste it as the UI.
Recreate the direction with real components, CSS, responsive layout, accessible states, and project-local assets where needed.
After implementation, use the browser to compare the rendered route against the direction and fix visible issues.

When to Skip Image Gen

Skip generation and implement directly when:

• the user requests a small deterministic CSS fix
• the project already has a strict design system and the task is just applying it
• the UI is primarily data correctness, form behavior, or accessibility remediation
• the requested asset is an existing SVG/vector/icon-system edit that should stay code-native