image-prompt

What this skill does

<!-- SYNC: Source at animalz-intelligence-os/modules/image-prompt-creator/.claude/skills/image-prompt/SKILL.md -->
---
name: image-prompt
description: "Create production-quality image generation prompts for Nano Banana (Google's AI image generator). Guides through brand selection, concept exploration, prompt crafting, and iterative revision. Use when asked to 'create an image prompt', 'make a featured image', 'generate image for blog', or similar."
---

# Image Prompt Creator

Create production-quality image generation prompts for Nano Banana (Google's AI image generator).

$ARGUMENTS — Optional: brand name, or description of what's needed. If empty, start from brand selection.

## Plugin root

This skill is part of the `image-prompt-creator` plugin. Resolve all relative paths (`brands/`, `docs/`, `examples/`) from the plugin root — the directory containing this skill's `skills/` folder.

At the start: read `docs/nano-banana-guide.md` for prompting reference. Keep it loaded throughout.

---

## Phase 1: Brief

### Step 1 — Brand selection

List available brands by scanning `brands/*/brand-config.md`. Present them:

> **Which brand are you creating images for?**
> Available: {list of brand names from folder names}
> Or type a new brand name to set one up.

Always pause for the user to confirm — even if only one brand exists. The user may want to add a new brand or may have opened the skill by mistake.

If no brands exist yet, tell the user:

> **No brands configured yet.** Let's set one up. What's the brand name? I'll create the folder and guide you through the brand config. See `brands/README.md` for the full template.

If the user names a brand that doesn't exist yet, guide them through creating `brands/{slug}/brand-config.md` using the template in `brands/README.md`.

Read the selected `brands/{name}/brand-config.md`. Also check if `brands/{name}/references/` exists and scan for subfolders (e.g., `featured-blog-images/`, `social-graphics/`). These are curated style references from previous successful images — use them to inform your prompt writing and optionally recommend attaching them as style references.

If $ARGUMENTS includes a brand name, still confirm it but don't ask from scratch.

### Step 2 — Input intake

Ask:

> **What do you need?** Describe your image — what it's for, any ideas you have, and what source material you can share.
>
> Examples: "featured image for this blog article", "social ad showing our product", "infographic visualizing this data", "I have a rough idea for an illustration."
>
> You can provide article text, documents, other images, voice notes, or just a rough idea. Type `help` for more guidance.

If the user types `help`, show this expanded guidance:

> **Input types you can provide:**
> - Article or document text (paste or attach) — I'll extract visual metaphors and concepts
> - A specific concept or visual idea — I'll go straight to crafting the prompt
> - A vague direction ("something about growth" or "needs to feel technical") — I'll propose concepts
> - Reference images — competitors, mood boards, styles you like
> - Voice memo transcript — describe your idea verbally
> - An existing draft prompt you want improved
>
> **Image types you can create:**
> - Blog featured images (illustrated or photographic)
> - In-article diagrams and infographics
> - Social media graphics (LinkedIn, Twitter/X)
> - Ad creatives (photographic scenes, conceptual)
> - Landing page visuals
> - Presentation graphics
> - Newsletter header images
>
> **How this works:**
> After you share your input, I'll either go straight to crafting a prompt (if your concept is clear) or propose 3-4 visual concepts for you to choose from. Then I'll write a production-ready Nano Banana prompt with brand colors, composition details, and iteration notes.

If $ARGUMENTS includes a description, use it as the input and skip the question.

### Step 3 — Reference images

If the user hasn't mentioned reference images, ask:

> **Do you have any reference images to include?** These could be style references, mood boards, competitor examples, or related visuals. They're optional but help a lot for style consistency.
>
> If yes, share them and describe what each one should be used for (style reference, color reference, layout reference, etc.).

Also check the brand's `references/` folder. If reference images exist for the relevant image type (e.g., `references/featured-blog-images/` for a blog hero), mention them:

> I found {N} reference images in the brand folder for {image type}. I'll use these to inform the style direction, and may recommend attaching one as a style reference to Nano Banana.

### Checkpoint

Before proceeding, confirm:
- Brand: {name}
- Image intent: {what it's for, where it will appear}
- Concept clarity: {clear concept vs. needs exploration}
- Reference images: {listed with roles, or "none"}
- Aspect ratio: {derived from placement — 16:9 for blog hero, 4:3 for inline, etc.}

---

## Phase 2: Concept (conditional)

**If the user provided a clear visual concept:** Summarize it back in 2-3 sentences. Confirm. Move to Phase 3.

**If the concept is vague or the user wants ideas:** Generate 3-4 concepts. For each:

> **Concept {A/B/C/D}: "{Short title}"**
> {2-3 sentence visual description — what the viewer would see}
> Type: {illustrated / photographic / diagrammatic}
> Why: {How this concept serves the content and brand}

Draw on the source material (article text, brief, etc.) to find visual metaphors. Consider both illustrated and photographic approaches.

Ask the user to pick one or combine elements from multiple concepts.

### Checkpoint

Confirm the selected concept in 2-3 sentences before proceeding.

---

## Phase 3: Prompt craft

### Step 1 — Write the prompt

Craft a complete, production-ready Nano Banana prompt. The output has two parts:

**Part 1 — Setup block** (for the user's reference, NOT pasted into Nano Banana):

> **Reference images to attach:**
> 1. `{filename}` — {what it is}
> 2. `{filename}` — {what it is}
> 3. `{filename}` — {what it is}
>
> **Resolution:** {2K start -> 4K upscale}
> **Aspect ratio:** {ratio} ({placement context})

Pull from brand config's "Reference images to attach" table + any user-provided references + curated references from the brand's `references/` folder. Use your judgment on whether to include style references from the brand folder — they're powerful but risk over-indexing if the model copies them too literally.

**Part 2 — The prompt** (self-contained, copy-pasteable into Nano Banana):

The prompt must be completely self-contained. The user should be able to copy-paste it as one block. Structure it as:

1. **Opening paragraph — image role descriptions.** Tell Nano Banana what each attached image is for. Example: "I'm attaching three reference images. Image 1 is my color palette — use these exact hex codes. Image 2 shows my brand's character designs — match this line art style for any animal characters. Image 3 is a style reference only — match its illustration technique but do not copy its subject or composition."

2. **The scene description** — narrative sentences following the prompting principles from the guide:
   - Style direction (line work, color approach, mood)
   - Composition with spatial language (proportions, layout zones)
   - All colors by hex code from brand config
   - Mood and tone direction
   - Anti-instructions (what to avoid, what NOT to include)
   - Format and cropping notes

**After the prompt, add iteration notes** (for the user's reference, NOT pasted):
- What to check after Round 1
- Likely Round 2 adjustments
- Any known model quirks

### Step 2 — Brand enforcement

Before presenting the prompt, cross-check against `brand-config.md`:
- All hex codes match the brand palette (no made-up colors)
- Style vocabulary aligns with brand guidelines
- Mascot/critter usage matches brand guidelines
- Anti-patterns from brand config are covered in anti-instructions

### Checkpoint

P