# Improv UI Brainstorm

> Generate divergent AI-image UI concept explorations for apps, websites, dashboards, onboarding, paywalls, empty states, App Store screenshots, and marketing UI screenshots. Use when the user asks to brainstorm, riff, improvise, imagine what a UI could look like, generate multiple UI concepts with Imagen/image generation, explore product screens visually, or compare UI directions. Do not use for exact production implementation, pixel-perfect redesigns, or coding the UI.

- Skill: `aihow-to/improv-ui-brainstorm` (Agent Skill, multi-file: 7 files)
- Install (CLI): `npx skillmds@latest add aihow-to/improv-ui-brainstorm`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aihow-to/improv-ui-brainstorm/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: aihow-to (https://skillmd.com/u/aihow-to)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/aihow-to/improv-ui-brainstorm

---


# Improv UI Brainstorm

Use this skill to turn a product idea, screenshot, rough brief, or design direction into a set of intentionally different UI concept images. The goal is controlled divergence: each image should test a distinct product/design hypothesis, not a minor style variant.

## Workflow

1. **Extract the brief.** Identify product, audience, platform, business model, existing references, current UI state, and any hard constraints. Ask only for missing details that materially affect the concepts.
2. **Name the essence.** Write: `Essence: This product is really about [human transformation / user job], not merely [software category].` Every concept must preserve this essence.
3. **Build a direction matrix.** Assign each image a different role, hypothesis, and optional improv move before generating. For reusable roles, read `references/direction-lenses.md`; for creative variation, read `references/improv-moves.md`.
4. **Write generation prompts.** Use concrete UI surfaces, exact important copy, platform cues, interaction state, visual mood, and explicit negative constraints. For prompt patterns, read `references/image-prompt-patterns.md`.
5. **Generate the set.** Default to 10 images unless the user asks for another count. If image generation is unavailable, output the exact prompts instead.
6. **Synthesize.** Finish with an index of concepts and a short recommendation. For the output format, read `references/critique-rubric.md`.

## Creative Rules

- Preserve the user's known brand/product truth, but do not invent a rigid design system unless asked.
- Anchor every concept to the product essence while varying surface, metaphor, hierarchy, tone, and commercial framing.
- Vary the product hypothesis first: trust, conversion, onboarding, retention, empty state, information density, monetization, virality, calm, speed, or delight.
- Make each image app-relevant: real controls, visible state, plausible content, and a clear reason for every major element.
- Keep marketing screenshots distinct from in-product UI. Marketing images may have external headlines, device mockups, and callouts; product UI should feel usable.
- Use platform-native cues when a platform is known. Do not turn native app screens into generic landing pages.
- Avoid repeating the same palette, layout, and hierarchy across the whole set unless the user explicitly asks for tight variations.
- Avoid fake awards, fake testimonials, misleading platform endorsement, unreadable microcopy, impossible device chrome, and decorative clutter that does not serve the product.

## Default Output

After image generation, provide:

- where the images were saved, if the tool reports a path.
- one line per concept: role, hypothesis, and what to steal.
- the strongest 2-3 directions.
- the riskiest ideas to avoid or validate before implementation.

Do not over-explain individual prompts unless the user asks for prompt text.