motiful/product-shots / skills/product-shots-ad-creative
product-shots-ad-creative
VerifiedDesigns high-performing ad creatives across Instagram, Facebook, TikTok, LinkedIn, Google (Display / Demand Gen), YouTube, Pinterest, and X/Twitter — locking platform dimensions, safe zones, text overlay policy, character limits, industry visual DNA, ad objective hierarchy, composition patterns, CTA strategy, and prompt sanitation. Use when the user says "create an ad", "design an ad creative", "make an ad", "广告创意", "帮我做个广告图", "Instagram ad", "TikTok ad", "Facebook ad", "Google Display ad", "LinkedIn ad", "YouTube ad", "Pinterest ad", "promotional image", "campaign creative", "ad banner", "投放图", or any platform-specific ad request. Routed from `product-shots` when `asset_type == "ad"` or `is_promotion == True`.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 17.0 KB |
| references/ad-objective-rules.md | markdown | 5.2 KB |
| references/banned-words.md | markdown | 6.4 KB |
| references/color-and-cta-strategy.md | markdown | 8.4 KB |
| references/composition-patterns.md | markdown | 6.4 KB |
| references/failures-defaults-output.md | markdown | 7.3 KB |
| references/hard-constraints.md | markdown | 13.8 KB |
| references/industry-style-rules.md | markdown | 7.1 KB |
| references/multi-platform-output.md | markdown | 6.0 KB |
| references/platform-style-profiles.md | markdown | 7.5 KB |
| references/quick-start-and-fidelity.md | markdown | 6.2 KB |
| references/workflow-and-self-check.md | markdown | 11.4 KB |
references/banned-words.md
View on GitHub---
name: banned-words
description: Section 20 — Prompt Banned Words + Required Positive Instruction. The list of platform names, ad-UI terms, brand-tool names that must NEVER enter the image generator prompt (they trigger fake UI / watermarks / screenshot artifacts), plus the positive-preservation instruction that protects native packaging text on the product itself. Loaded by the parent skill's Execution Procedure Step 0 and re-validated at the Self-Check Gate alongside hard-constraints.md.
---
## Execution Procedure
```
def register_protected_phrase(phrase) → registered_id
# Called by quick-start-and-fidelity preflight for each non-null copy field
# (brand_name / slogan / cta_text / price_or_offer).
# Appends the phrase to the session's protected_phrases registry. Returns the
# registry id used by sanitize_prompt() to swap the phrase in / out around
# banned-word filtering. Idempotent — re-registering the same phrase returns
# the existing id.
def sanitize_prompt(prompt, protected_phrases) → sanitized_prompt
# 1. swap each protected_phrase with __USER_CONTENT_i__ placeholder
# 2. apply PROMPT_SAFE_REPLACEMENTS (banned phrase → safe paraphrase)
# 3. remove remaining PROMPT_BANNED_WORDS (case-insensitive whole-word)
# 4. restore __USER_CONTENT_i__ placeholders with the original phrases (quoted)
# 5. inject the Required Positive Instruction (Preserve Native Packaging Text)
# 6. collapse extra whitespace and return.
# Called by SKILL.md EP Step 5 for every per-platform prompt before image-gen dispatch.
```
## Section 20 — Prompt Banned Words
The following words are **banned in image-generation prompts**. AI image models render them as fake in-image UI, watermarks, or screenshot artifacts. They are allowed inside the skill (for format selection / industry matching / routing) but **must never appear in the text sent to the generator**.
### Exception: User-Specified Content Passthrough
User-supplied content is **never filtered**. The following user-provided values are protected and passed through verbatim:
```
"user_content_exceptions": [
"brand_name", # user-supplied brand name
"slogan", # user-supplied slogan
"cta_text", # user-supplied CTA copy
"price_offer" # user-supplied price / offer
]
```
Sanitation algorithm (high-level):
```
1. Extract user-specified content (brand_name, slogan, cta_text, price_offer)
2. Temporarily replace user content with placeholders __USER_CONTENT_i__
3. Apply replacement rules (banned phrase → safe paraphrase) — see PROMPT_SAFE_REPLACEMENTS
4. Remove remaining banned words (case-insensitive whole-word match)
5. Restore user content — wrap in quotes
6. Collapse extra whitespace
```
### Banned Words List
```python
PROMPT_BANNED_WORDS = {
# Platform names (strictly forbidden)
"platform_names": [
"instagram", "ig", "facebook", "fb", "tiktok", "youtube",
"linkedin", "pinterest", "google display", "x", "twitter",
"rednote", "xiaohongshu", "小红书"
],
# Ad UI terms (forbidden)
"ad_ui_terms": [
"ad", "advertisement", "sponsored", "promoted",
"feed ad", "story ad", "reel ad",
"like button", "share button", "comment section", "follow button",
"ad manager", "boost post", "campaign"
],
# Brand / tool names (forbidden)
"brand_tool_names": [
"meta", "google ads", "canva", "lovart", "figma",
"adobe", "photoshop"
]
}
```
#### Safe Replacements
```python
PROMPT_SAFE_REPLACEMENTS = {
"Instagram Feed Ad": "Product promotion image, vertical 4:5, polished lifestyle aesthetic",
"Instagram Story Ad": "Vertical 9:16 full-screen image, dynamic mobile-first design",
"Facebook ad": "Product showcase image for social media",
"TikTok ad": "Vertical 9:16 product video still, casual handheld phone aesthetic, raw authentic feel",
"Google Display banner ad": "Product-focused banner image, horizontal 1.91:1, clean composition, no text overlay",
"LinkedIn ad": "Professional business image, square or landscape format",
"with Shop Now button": "with bold CTA text 'Shop Now'",
"with like button": "with engagement prompt",
"sponsored post style": "promotional content style"
}
```
### Required Positive Instruction — Preserve Native Packaging Text
This is the **positive companion** to the banned-words filter. When the prompt instructs "no text overlay" (e.g., for Google Display), the AI image model may incorrectly delete printed text on the product packaging itself (brand name on a bottle, ingredient list on a box, volume label, etc.). The Required Positive Instruction prevents that.
Verbatim instruction to inject:
```
clean product shot with no added graphic text, captions, or button overlays,
BUT fully preserve all native printed text on the product packaging —
brand name, product name, label, volume, and any markings as they physically
appear on the product.
```
**Rationale**: prevents the AI from misinterpreting "no text" as "delete brand name from product packaging". This dual-direction constraint (negative banned-words filter + positive preservation rule) is the defining defensive prompt-engineering pattern of this skill.
## Why These Are MUST-level
- **Wrong dimensions** → ad rejected by the platform, or destructively cropped at delivery.
- **Safe-zone violation** → CTA / brand name / face hidden behind UI on real devices (TikTok right rail, Reel bottom 35%, Shorts bottom 35%).
- **Text overlay violation on Google Display** → **immediate ad rejection** (zero tolerance — there is no warning, no review).
- **TikTok QR code or other-platform watermark** → **immediate ad rejection**.
- **LinkedIn font < 60pt or contrast < 4.5:1** → unreadable on feed scroll, brand looks unprofessional.
- **Char limit overrun** → ad copy truncated mid-sentence with "..." (especially harsh on FB at 27 chars vs IG 40).
- **Google CJK miscount** → ad copy fails preflight check (Chinese / Japanese / Korean chars cost double).
- **Banned word in prompt** → image contaminated with fake platform UI, fake username, fake like-counts, fake "Sponsored" tags. Looks like a screenshot, not a designed visual. May trigger platform takedowns for impersonation.
- **Missing Required Positive Instruction** → product packaging text gets erased by the image model, brand name disappears from the bottle / box / label.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |