Ad Designer

Generate marketing ad images using Nano Banana Pro (Gemini 3 Pro Image). Accepts campaign-planner creative briefs, reads brand bible for visual style, constr...

installs

stars

karma

SkillRank score ↗

7.9/ 10

evaluated by implexa, claude-haiku-4-5 · 2026-05-31

ad-designer wraps gemini image generation with brand-aware prompt construction, aspect ratio mapping, and a mandatory self-review loop to catch hallucinated text and logos before delivery. supports draft-first workflow and multiple brief processing.

structure

9.0

trigger phrases

8.0

procedure

8.0

edge cases

7.0

documentation

8.0

strengths

view original SKILL.md from clawhubclick to expand

---
name: ad-designer
description: Generate marketing ad images using Nano Banana Pro (Gemini 3 Pro Image). Accepts campaign-planner creative briefs, reads brand bible for visual style, constructs marketing-optimized prompts, and produces platform-ready images at correct aspect ratios. Supports 1:1, 9:16, 16:9, 4:5 formats. Includes self-review loop to catch hallucinated logos, wrong text, and quality issues. Draft-first workflow (1K fast iteration, 4K final). Outputs to /tmp/marketing/assets/images/.
---

# Ad Designer

Generate marketing ad images from campaign-planner creative briefs. Wrap nano-banana-pro with brand-aware prompt construction, aspect ratio handling, and a self-review loop that catches quality issues before marking images final.

## Roles

- **Ad Designer** (this skill) — sole executor
- **nano-banana-pro** — image generation engine (upstream dependency)
- **campaign-planner** — upstream source of creative briefs
- **brand-bible-builder** — upstream source of brand visual identity

---

## Prerequisites

Verify before starting:

```bash
command -v uv
test -n "$GEMINI_API_KEY"
test -f ~/.codex/skills/nano-banana-pro/scripts/generate_image.py
```

If `uv` is missing: tell the user to install it with `curl -LsSf https://astral.sh/uv/install.sh | sh`.
If `GEMINI_API_KEY` is unset: tell the user to export it from Google AI Studio.
If the script is missing: tell the user the nano-banana-pro skill must be installed first.

---

## Step 1: Collect Inputs

Identify which inputs are present in the conversation or filesystem:

| Input | Source | Where to find it |
|-------|--------|-----------------|
| Creative brief(s) | campaign-planner skill | `/tmp/marketing/campaigns/<brand>-campaign-plan.json` → `ad_creatives[]` array, or inline in conversation |
| Brand bible | brand-bible-builder skill | `/tmp/marketing/brands/<brand>/brand-bible.md` or inline |
| Aspect ratio | User or brief | Defaults to 1:1 if not specified |
| Resolution mode | User | Defaults to draft (1K) unless user says "final" or "4K" |

If the creative brief is missing, ask:

> "I need a creative brief to generate the ad. Run `/campaign-planner` first, or paste the brief details: headline, visual direction, CTA, and target audience."

If the brand bible is missing, proceed with neutral visual defaults — note this in the output and recommend running `/brand-bible-builder` for on-brand results.

---

## Step 2: Read Brand Bible Visual Identity

Open the brand bible and extract these fields into a working reference:

- **Primary colors** — hex codes or descriptors (e.g. `#1A2E5A deep navy`, `#F5A623 amber`)
- **Secondary / accent colors** — supporting palette
- **Font style** — serif vs sans-serif, weight, mood descriptor (e.g. "bold geometric sans", "light editorial serif")
- **Imagery style** — photography vs illustration, lifestyle vs product-only, color grading mood (warm/cool/muted/vibrant)
- **Layout pattern** — minimal and clean vs content-dense, whitespace-heavy vs layered
- **Tone** — formal, casual, bold, playful, premium, approachable

Keep this as a 6-field internal reference. Do not show it to the user. Use it when constructing prompts in Step 3.

---

## Step 3: Construct Image Generation Prompt

Apply these rules every time. They are non-negotiable.

### Core Rules (from best-practice image generation)

**Rule 1 — Less is more.** Short, focused prompts outperform long descriptive ones. Target 15–30 words for the visual description. Cut adjectives that repeat the same idea.

**Rule 2 — No logos, brand names, or company names.** Never include the brand name, product name, logo description, or any trademark in the prompt. The generation engine hallucinates these badly and they must be added in post-production.

**Rule 3 — Only include text that is explicitly specified.** If the brief specifies a headline to appear in the image, include it exactly as written — no paraphrasing. If no text is specified, do not add any text to the prompt. Do not include CTAs, taglines, or copy unless the brief explicitly says to render text.

**Rule 4 — Describe emotion and composition, not data.** Instead of "a woman saving money on her bills," write "confident woman, bright home office, warm morning light, relaxed expression." The emotion communicates the message; the model handles the rest.

**Rule 5 — Translate brand colors without naming the brand.** Convert hex codes to descriptors: `#1A2E5A` → "deep navy", `#F5A623` → "warm amber". Use these color words in the prompt naturally.

### Prompt Construction Formula

Assemble the prompt in this order:

```
[SUBJECT] — who or what is the focal element
[SETTING] — environment or background context
[MOOD / EMOTION] — feeling the image should evoke
[COMPOSITION] — framing, shot type, perspective
[LIGHTING] — quality and direction of light
[COLOR PALETTE] — 2–3 colors derived from brand bible
[STYLE] — photography/illustration style from brand bible
[AVOID] — explicit exclusions (text, logos, clutter, etc.)
```

Do not use all 8 fields every time. Use only those relevant to the brief. A 5-field prompt is often stronger than an 8-field one.

### Example Construction

Brief input:
```
headline: "Finally, sleep that works"
visual_direction: "Woman waking up refreshed, natural light bedroom"
cta: "Try free tonight"
target_persona: "Aisha, 28, KL, stressed professional"
```

Brand bible: warm coral `#E8735A`, off-white `#FAF7F2`, editorial sans-serif, warm and minimal photography style.

Constructed prompt:
```
Young woman waking up peacefully, sunlit minimalist bedroom, linen textures,
warm coral and off-white tones, soft morning light through sheer curtains,
close-up on relaxed expression, warm editorial photography style
```

What was excluded: brand name, CTA text, product name, headline text (no text was specified for visual rendering).

---

## Step 4: Select Aspect Ratio

Map the brief's platform or format to the correct ratio flag and pixel dimensions:

| Platform / Format | Ratio | Pixel Dimensions | Use Case |
|-------------------|-------|-----------------|----------|
| Instagram / Facebook Feed | 1:1 | 1080 × 1080 | Square feed post |
| Instagram Story / Reel | 9:16 | 1080 × 1920 | Full-screen vertical |
| Facebook Feed (portrait) | 4:5 | 1080 × 1350 | Taller feed post, more screen area |
| YouTube / LinkedIn Cover | 16:9 | 1920 × 1080 | Landscape / banner |

If the brief does not specify a platform, default to 1:1.

For carousel ads: generate all cards at the same ratio. Use 1:1 unless the brief specifies otherwise.

Pass the ratio to the script via `--resolution`. The nano-banana-pro script handles internal sizing. Generate at 1K for drafts, 4K for finals.

---

## Step 5: Generate Images

Create the output directory:

```bash
mkdir -p /tmp/marketing/assets/images
```

### Single image generation

Use this command pattern. Always use absolute path for the script.

```bash
uv run ~/.codex/skills/nano-banana-pro/scripts/generate_image.py \
  --prompt "<constructed prompt>" \
  --filename "/tmp/marketing/assets/images/<timestamp>-<brief-id>-draft.png" \
  --resolution 1K
```

Filename format: `yyyy-mm-dd-hh-mm-ss-<brief-id>-draft.png`
- `brief-id`: lowercase kebab-case derived from the brief headline (e.g. `finally-sleep-works`)
- Use current date/time for timestamp

### Draft-first workflow

Always follow this sequence:

1. **Draft at 1K** — generate first at 1K for fast feedback. Present to user.
2. **Iterate** — if issues found in self-review (Step 6), adjust prompt and regenerate at 1K. Max 3 attempts.
3. **Final at 4K** — only after user approves the draft or after self-review passes. Change filename suffix from `-draft` to `-final`, set `--resolution 4K`.

### Carousel ads

Generate each card as a separate image. Naming convention:
```
yyyy-mm-dd-hh-mm-ss-<brief-id>-card-01-draft.png
yyyy-mm-dd-hh-mm-ss-<brief-id>-card-02-draft.png
```

Ensure visual consistency across cards:
- Use the same color palette descriptor in every card's prompt
- Use the same style descriptor in every card's prompt
- Use the same lighting descriptor in every card's prompt
- Vary only the subject and composition per card

---

## Step 6: Self-Review Loop (CRITICAL)

After generating each image, perform a structured self-review before presenting to the user. Do not skip this step.

Run through this checklist for every generated image:

| Check | Pass Condition | Fail Action |
|-------|---------------|-------------|
| Text accuracy | Any specified text appears correctly and is readable | Simplify prompt; add explicit text instruction |
| No hallucinated logos | No brand marks, wordmarks, icons, or watermarks visible | Add "no logos, no text, no watermarks" to avoid clause |
| No hallucinated brand names | No company names or product names rendered as text | Add "no brand names, no product names" to avoid clause |
| Aspect ratio | Matches requested format | Re-run with correct ratio flag |
| Color alignment | Dominant colors match brand palette intent | Strengthen color descriptors in prompt |
| Overall quality | Image is coherent, professional, on-brief | Rebuild prompt from scratch using simpler structure |

### Fail response

If any check fails:

1. Identify the specific failure from the table above.
2. Apply the corresponding fail action to the prompt.
3. Regenerate at 1K.
4. Re-run the self-review checklist on the new image.
5. Stop after 3 total attempts. If still failing on attempt 3, present the best result to the user with a note explaining what the issue is and what was tried.

Do not present a failing image silently. Always tell the user what was wrong and what fix was applied.

---

## Step 7: HITL — Present for User Review

After self-review passes (or after 3 attempts), present the results:

```
Ad image(s) generated for: [brief headline]

Draft(s) saved to /tmp/marketing/assets/images/:
  [timestamp]-[brief-id]-draft.png   ← [ratio] | 1K

Self-review: [PASS / notes on any issues found and fixed]

Review the image and reply with one of:
  A) Approve → I'll upscale to 4K final
  B) Adjust → describe what to change
  C) Skip → move to next brief
```

Wait for the user's response before generating the 4K final or moving to the next brief.

If the user says "adjust" or describes changes:
- Apply the change to the prompt (keep changes minimal — one diff at a time)
- Regenerate at 1K
- Re-run self-review
- Present again

If the user approves:
- Regenerate at 4K with `-final` filename suffix
- Confirm: "Final saved: `/tmp/marketing/assets/images/[filename]-final.png`"

---

## Step 8: Process Multiple Briefs

If the input contains multiple briefs (e.g. from a campaign-planner JSON with 3–5 image ad entries), process them sequentially. Complete the full cycle (generate → self-review → HITL approval → final) for each brief before moving to the next.

After all briefs are processed, present a summary:

```
All ad images complete.

Generated:
  /tmp/marketing/assets/images/[file-1]-final.png   [ratio]
  /tmp/marketing/assets/images/[file-2]-final.png   [ratio]
  /tmp/marketing/assets/images/[file-3]-final.png   [ratio]

Next step: run /meta-ads-publisher to upload these to your ad account,
or /campaign-planner to review the full campaign.
```

---

## Error Handling

| Error | Action |
|-------|--------|
| `Error: No API key provided.` | Tell user to run `export GEMINI_API_KEY=<key>` and retry |
| `Error loading input image:` | Verify the `--input-image` path exists; fix path and retry |
| `quota / 403 / permission` API error | Wrong key or quota exceeded; ask user to check Google AI Studio quota |
| Script not found at `~/.codex/skills/...` | nano-banana-pro skill is not installed; tell user to install it from ClawHub |
| `uv: command not found` | Run `curl -LsSf https://astral.sh/uv/install.sh | sh` then retry |
| Image generation returns blank/empty | Prompt may be too restrictive; remove avoid clauses and simplify to subject + style only |

---

## File Paths Reference

```
/tmp/marketing/
├── assets/
│   └── images/
│       ├── yyyy-mm-dd-hh-mm-ss-<brief-id>-draft.png    ← draft outputs
│       └── yyyy-mm-dd-hh-mm-ss-<brief-id>-final.png    ← approved 4K finals
├── brands/<brand>/
│   └── brand-bible.md                                   ← brand visual reference
└── campaigns/
    └── <brand>-campaign-plan.json                       ← input brief source
```

don't have the plugin yet? install it then click "run inline in claude" again.

Ad Designer

Generate marketing ad images from campaign-planner creative briefs. Wrap nano-banana-pro with brand-aware prompt construction, aspect ratio handling, and a self-review loop that catches quality issues before marking images final.

Intent

This skill takes creative briefs from campaign-planner, extracts visual direction and brand guidelines from brand-bible-builder, constructs tight image generation prompts that avoid hallucinated logos and text, and produces platform-ready ad images at correct aspect ratios. Use it when you need on-brand ad visuals fast. The skill enforces a draft-first workflow (1K resolution for iteration) followed by 4K finalization only after user approval. Self-review catches common generation failures before output.

Inputs

Required

Creative brief , from campaign-planner skill output or pasted inline. Must contain: headline, visual direction, target persona, platform/format. Source: /tmp/marketing/campaigns/<brand>-campaign-plan.json → ad_creatives[] array, or conversation paste.
Aspect ratio , user specification or inferred from platform. Default: 1:1 (square). Accepted: 1:1, 9:16, 16:9, 4:5.
Resolution mode , "draft" (1K, fast) or "final" (4K, upscaled). Default: draft.

Required External Connections

GEMINI_API_KEY , Google AI Studio API key. Export as env var: export GEMINI_API_KEY=<key>. Required to call Gemini 3 Pro Image generation. Obtain from https://aistudio.google.com/app/apikey. Monitor quota at AI Studio dashboard; quota exhaustion returns 403.
nano-banana-pro , upstream skill dependency. Image generation engine. Script location: ~/.codex/skills/nano-banana-pro/scripts/generate_image.py. Must be installed from ClawHub before running ad-designer. Verify: test -f ~/.codex/skills/nano-banana-pro/scripts/generate_image.py.
uv , runtime package manager. Required to execute the nano-banana-pro script. Verify: command -v uv. Install if missing: curl -LsSf https://astral.sh/uv/install.sh | sh.

Optional

Brand bible , from brand-bible-builder skill output. Provides visual identity (colors, fonts, imagery style, tone). Source: /tmp/marketing/brands/<brand>/brand-bible.md. If missing: proceed with neutral defaults and flag recommendation to run brand-bible-builder for on-brand results.
Campaign JSON with multiple briefs , if input is a full campaign plan JSON, process all ad_creatives[] entries sequentially.

Edge Cases

Empty input , no brief provided. Prompt user for inline details or run campaign-planner first.
Missing brand bible , proceed with neutral visual defaults; output includes disclaimer.
API quota exceeded , Gemini API returns 403. User must check quota in Google AI Studio and wait for reset or upgrade plan.
Network timeout , image generation takes up to 60 seconds. Long delays may indicate API saturation. Retry once after 30 seconds.
Malformed JSON input , campaign JSON missing required fields (headline, visual_direction). Ask user to validate brief structure.

Procedure

Step 1: Collect and Validate Inputs

Input: conversation context or filesystem (/tmp/marketing/campaigns/...) Output: validated brief object, identified brand bible location, selected aspect ratio and resolution mode

Check if creative brief is present in the conversation or at /tmp/marketing/campaigns/<brand>-campaign-plan.json.
Extract required fields: headline, visual_direction, target_persona. If any are missing, ask the user inline.
Check for brand bible at /tmp/marketing/brands/<brand>/brand-bible.md. Note if missing (non-blocking).
Infer or ask user for aspect ratio. Default to 1:1 if not specified.
Infer or ask user for resolution mode. Default to "draft" if not specified.
If input is a campaign JSON with multiple briefs, list all entries and confirm sequence.

Step 2: Read Brand Bible into Working Reference

Input: brand-bible.md file (or default to neutral if missing) Output: internal 6-field reference object (do not show to user)

Open /tmp/marketing/brands/<brand>/brand-bible.md.
Extract and parse these fields:
- Primary colors (hex codes and/or descriptors, e.g. #1A2E5A deep navy)
- Secondary / accent colors
- Font style (serif vs sans-serif, weight, mood)
- Imagery style (photography vs illustration, color grading mood)
- Layout pattern (minimal vs dense, whitespace philosophy)
- Tone (formal, casual, bold, playful, premium, approachable)
Store as internal reference. Keep hidden from user. Use in Step 3 prompt construction.
If brand bible is missing, log: "[brand] brand bible not found; using neutral visual defaults. Recommend running /brand-bible-builder for on-brand results."

Step 3: Construct Image Generation Prompt

Input: creative brief fields, brand reference object (from Step 2), composition rules Output: tight, brand-aware image generation prompt (15-30 words target)

Apply non-negotiable rules:
- No logos, brand names, product names, trademarks in prompt.
- Only include text explicitly specified in brief. If brief says "headline: Finally, sleep that works" and specifies "render text in image", include it exactly. Otherwise, omit all text and CTAs.
- Describe emotion and composition, not data. "Confident woman, bright home office, warm morning light" not "woman saving money."
- Translate brand colors to descriptors. #1A2E5A → "deep navy", #F5A623 → "warm amber". Use naturally in prompt.
- Less is more. Short, focused prompts outperform long ones. Cut adjectives that repeat ideas.
Assemble prompt using this order (use only relevant fields, omit others):
- [SUBJECT] , focal element (who/what)
- [SETTING] , environment or background
- [MOOD/EMOTION] , feeling to evoke
- [COMPOSITION] , framing, shot type, perspective
- [LIGHTING] , quality and direction of light
- [COLOR PALETTE] , 2-3 colors from brand bible (as descriptors)
- [STYLE] , photography/illustration style from brand bible
- [AVOID] , explicit exclusions (no logos, no text, no watermarks, etc.)
Example walkthrough:
- Brief: "Young professional woman waking, natural light bedroom, warm minimal aesthetic, approachable tone"
- Brand: coral #E8735A, off-white #FAF7F2, editorial sans-serif, warm photography
- Constructed: "Young woman waking peacefully, sunlit minimalist bedroom, linen textures, warm coral and off-white tones, soft morning light through sheer curtains, close-up on relaxed expression, warm editorial photography style"
- What was excluded: brand name, CTA text, product name, headline text.
Output the final prompt to user for review before generating.

Step 4: Select Aspect Ratio and Resolution

Input: user platform specification or brief default, resolution mode Output: ratio flag (for script), pixel dimensions, resolution setting (1K or 4K)

Map platform/format to ratio:
- Instagram Feed / Facebook Feed → 1:1 (1080 × 1080)
- Instagram Story / Reel → 9:16 (1080 × 1920)
- Facebook Feed portrait → 4:5 (1080 × 1350)
- YouTube / LinkedIn Cover → 16:9 (1920 × 1080)
- Default (no platform specified) → 1:1
Set resolution based on mode:
- Draft mode → 1K (fast iteration)
- Final mode → 4K (approved output)
Pass ratio to script via --resolution flag. Nano-banana-pro handles internal sizing.

Step 5: Generate Draft Image at 1K

Input: constructed prompt, aspect ratio, filename conventions, output directory Output: PNG image file at /tmp/marketing/assets/images/yyyy-mm-dd-hh-mm-ss-<brief-id>-draft.png

Create output directory: mkdir -p /tmp/marketing/assets/images
Generate filename using timestamp and brief-id:
- Format: yyyy-mm-dd-hh-mm-ss-<brief-id>-draft.png
- Timestamp: current UTC date and time
- Brief-id: lowercase kebab-case from headline. Example: "Finally, sleep that works" → finally-sleep-works

Execute generation command:

uv run ~/.codex/skills/nano-banana-pro/scripts/generate_image.py \
  --prompt "<constructed prompt>" \
  --filename "/tmp/marketing/assets/images/<timestamp>-<brief-id>-draft.png" \
  --resolution 1K

Wait for completion. Image generation typically takes 20-60 seconds depending on API load.
Confirm file exists: test -f /tmp/marketing/assets/images/<filename> && echo "Generated OK"
For carousel ads, generate each card separately using naming: yyyy-mm-dd-hh-mm-ss-<brief-id>-card-01-draft.png, card-02-draft.png, etc. Use identical color/style/lighting descriptors across all cards; vary only subject and composition per card.

Step 6: Self-Review Loop (CRITICAL)

Input: generated draft PNG image Output: pass/fail verdict with specific notes; if fail, regenerated image after prompt adjustment

Open generated image and visually inspect against this checklist:

Check	Pass Condition	Fail Action
Text accuracy	Any specified text appears correctly, readable, no typos	Simplify prompt; add explicit text instruction
No hallucinated logos	No brand marks, wordmarks, icons, watermarks visible	Add "no logos, no text, no watermarks" to avoid clause
No hallucinated brand names	No company/product names rendered as text	Add "no brand names, no product names" to avoid clause
Aspect ratio	Matches requested format without letterboxing	Re-run with correct ratio flag
Color alignment	Dominant colors match brand palette intent	Strengthen color descriptors in prompt
Overall quality	Image is coherent, professional, on-brief, no artifacts	Rebuild prompt from scratch; simplify structure

If all checks pass: log "Self-review: PASS" and proceed to Step 7.
If any check fails:
- Identify the specific failure from the table.
- Apply the corresponding fail action to the prompt.
- Regenerate at 1K using the adjusted prompt.
- Re-run self-review on the new image.
- Track attempt count (1 of 3 max).
On the 3rd failed attempt: stop and present the best result to user with a note: "Attempted 3 times. Issue identified: [specific failure]. Recommendation: [what was tried]."
Never present a failing image silently. Always tell the user what failed and what fix was applied.

Step 7: Present Draft for User Review (HITL)

Input: self-reviewed draft image, brief headline, aspect ratio, self-review verdict Output: user response (Approve, Adjust, or Skip)

Present results in this format:

Ad image(s) generated for: [brief headline]

Draft(s) saved to /tmp/marketing/assets/images/:
  [timestamp]-[brief-id]-draft.png   ← [ratio] | 1K

Self-review: [PASS / notes on any issues found and fixed]

Review the image and reply with one of:
  A) Approve → I'll upscale to 4K final
  B) Adjust → describe what to change
  C) Skip → move to next brief

Display the draft image inline if possible.
Wait for user response. Do not proceed without explicit user direction.

Step 8: Handle User Response

Input: user reply (Approve, Adjust, or Skip) Output: final 4K image (if Approve) or regenerated draft (if Adjust) or proceed to next brief (if Skip)

If user approves:
- Regenerate the image at 4K using same prompt, new filename with -final suffix.
- Command: same as Step 5 but with --resolution 4K and filename ending in -final.png.
- Confirm: "Final saved: /tmp/marketing/assets/images/[filename]-final.png"
- Proceed to next brief if multiple exist (Step 9).
If user adjusts (describes changes):
- Parse the user's requested change.
- Apply change to prompt (keep minimal, one diff at a time).
- Regenerate at 1K.
- Run self-review (Step 6) on new draft.
- Re-present (Step 7).
- Limit adjustment loops to 3 iterations total across all attempts.
If user skips:
- Mark this brief as "skipped" in log.
- Proceed to next brief in campaign (Step 9).

Step 9: Process Multiple Briefs (if applicable)

Input: campaign JSON with multiple ad_creatives[] entries Output: all drafts and finals processed and saved

If the input brief source is a campaign plan JSON containing multiple ad entries, repeat Steps 1-8 for each brief sequentially.
Complete the full cycle (generate → self-review → HITL approval → final) for each brief before moving to the next.
Log each brief's status: [headline] → FINAL | SKIPPED | ADJUSTED.

Step 10: Final Summary and Handoff

Input: all completed briefs (drafts and finals) Output: summary message with file listing and next-step recommendations

After all briefs are processed, present summary:

All ad images complete.

Generated:
  /tmp/marketing/assets/images/[file-1]-final.png   [ratio]
  /tmp/marketing/assets/images/[file-2]-final.png   [ratio]
  /tmp/marketing/assets/images/[file-3]-final.png   [ratio]

Next step: run /meta-ads-publisher to upload these to your ad account,
or /campaign-planner to review the full campaign.

Confirm all final files exist and are non-empty.
Provide file paths as absolute paths for copy/paste into downstream tools.

Decision Points

Missing Creative Brief

If brief is not in conversation or filesystem: ask user to run /campaign-planner first or paste brief inline (headline, visual direction, CTA, target audience).
If brief is incomplete (missing headline or visual direction): ask user to fill in the missing fields.

Missing Brand Bible

If brand bible does not exist at /tmp/marketing/brands/<brand>/brand-bible.md: proceed with neutral visual defaults. Add disclaimer to output: "[brand] brand bible not found; using neutral visual defaults. Recommend running /brand-bible-builder for on-brand results."

Platform / Aspect Ratio Not Specified

If user does not specify platform or aspect ratio: default to 1:1 (square, 1080 × 1080). Confirm with user before generating.

Self-Review Fails on First Attempt

If any check fails (e.g. hallucinated logo, wrong aspect ratio, color mismatch): adjust prompt using the fail action from the checklist and regenerate at 1K. Re-run self-review.
After 3 failed attempts: present the best result with a note explaining the issue and what was tried. Do not block.

Text Rendering in Image

If the brief specifies a headline or copy to render in the image: include the exact text in the prompt (e.g. "Include text: 'Finally, sleep that works'"). Flag in self-review that text must be accurate and readable.
If the brief does not specify text: omit all text, CTAs, and taglines from the prompt. Let the visual stand alone.

User Approves the Draft

If user says "Approve" or "looks good": upscale to 4K final. Regenerate with same prompt, new filename with -final suffix.
If user says "Adjust" or describes specific changes: apply one change at a time to the prompt. Regenerate at 1K. Re-run self-review. Re-present.
If user says "Skip" or "move on": mark brief as skipped. Proceed to next brief if multiple exist.

API / Infrastructure Errors

If GEMINI_API_KEY is unset: user sees "Error: No API key provided." Tell user: export GEMINI_API_KEY=<key> (obtain from https://aistudio.google.com/app/apikey). Retry after setting.
If API returns 403 (quota/permission): user exceeded API quota or key is invalid. Tell user to check Google AI Studio quota and plan. Retry after quota reset or upgrade.
If nano-banana-pro script is not found: nano-banana-pro skill is not installed. Tell user to install from ClawHub first.
If uv is not found: tell user to run curl -LsSf https://astral.sh/uv/install.sh | sh then retry.
**If image generation returns

Ad Designer

related skills

Ad Designer

Intent

Inputs

Required

Required External Connections

Optional

Edge Cases

Procedure

Step 1: Collect and Validate Inputs

Step 2: Read Brand Bible into Working Reference

Step 3: Construct Image Generation Prompt

Step 4: Select Aspect Ratio and Resolution

Step 5: Generate Draft Image at 1K

Step 6: Self-Review Loop (CRITICAL)

Step 7: Present Draft for User Review (HITL)

Step 8: Handle User Response

Step 9: Process Multiple Briefs (if applicable)

Step 10: Final Summary and Handoff

Decision Points

Missing Creative Brief

Missing Brand Bible

Platform / Aspect Ratio Not Specified

Self-Review Fails on First Attempt

Text Rendering in Image

User Approves the Draft

API / Infrastructure Errors