Design static ad creatives for social media ads, display banners, and digital advertising campaigns. Build production-ready ads via the design subagent and present them as iframes on the canvas.

When to Use

User needs ad creatives for Facebook, Instagram, LinkedIn, Google Display, or TikTok
User wants banner ads or display advertising assets
User needs multiple ad variants for A/B testing
User wants ad copy and visual design together
User wants to iterate on ad creative based on performance data

When NOT to Use

Organic social media content (use content-machine skill)
Video ads or animated content (use storyboard skill for planning)
Full landing pages (use the artifacts skill)

Image-First Design Philosophy

Every ad should be built around a striking, full-bleed generated image. The image IS the ad — text is overlaid on top, not placed in a separate zone. This produces the punchy, scroll-stopping aesthetic that performs on modern social feeds.

Always generate images using generateImage from the media-generation skill unless:

The user uploads/attaches their own product photos or brand imagery
The ad is purely typographic by explicit request
The platform is text-only (e.g., Google RSAs)

Image style direction: Photorealistic, editorial, dramatic. Think fashion magazine meets Apple product launch. Single striking subjects, dramatic studio lighting, dark/moody backgrounds, high contrast. Avoid: abstract digital art, generic stock photo aesthetic, flat illustrations, busy compositions. Each image should have one clear focal point that makes someone stop scrolling.

Prompt strategy for generated images:

Lead with a specific, tangible subject (object, scene, or person)
Specify lighting style (studio, dramatic, golden hour, Rembrandt)
Specify camera quality ("shot on Hasselblad," "editorial photography," "fine art")
Add negative prompts: "text, words, letters, logos, watermark, blurry, low quality, cartoon, illustration"
Each ad in a set should use a different visual metaphor — same brand, different visual world

Default Format

Default to square (1:1) for ad creatives. Use portrait/iPhone format (9:16, iframe size 608×1080) when:

User asks for Instagram Stories or Reels ads
User asks for TikTok ads
User mentions "mobile ads," "iPhone," "portrait," or "stories format"
User asks for 9:16 or vertical format

Methodology

Step 1: Creative Brief

Gather these inputs:

Platform & Format: Which platform? (Google Ads, Meta, LinkedIn, TikTok, X/Twitter) Which format? (Search RSAs, feed, stories, display)
Objective: Awareness, consideration, or conversion?
Target audience: Who will see this ad? What stage of awareness? (Problem-aware, solution-aware, product-aware)
Key message: Single most important thing to communicate
CTA: What action should the viewer take?
Brand assets: Logo, colors, fonts, product images. If the user hasn't provided their actual logo, note this and ask at the end.
Performance data (if iterating): Which headlines/descriptions are performing best/worst? What angles have been tested?

Step 2: Platform Specifications (2025-2026)

Enforce these programmatically. Count characters in code, don't eyeball.

Meta (Facebook/Instagram) — Visual

Placement	Dimensions	Safe zone (keep critical elements inside)
Feed (square)	1080×1080	~100px margin all edges
Feed (portrait) — preferred	1080×1350 (4:5)	4:5 outperforms 1:1 on CTR; mobile-first
Stories	1080×1920	Top 14% (270px) + bottom 20% (380px) = dead zones
Reels	1080×1920	Top 14% + bottom 35% (670px) — like/comment/share UI is taller
Universal safe core	1010×1280 centered	Design inside this box → works everywhere

Upload at 2x pixel density (2160×2160 for a 1080 slot) for Retina sharpness. JPG/PNG, 30MB max.

20% text rule: officially removed — Meta no longer rejects text-heavy images, but the delivery algorithm still quietly throttles them. Keep text minimal; move details to primary text.

Meta — Text

Element	Limit	Notes
Primary text	125 chars visible feed / ~72 chars visible in Reels	Write for 72
Headline	40 chars rec	Below image
Description	30 chars rec	Often hidden on mobile

Google Ads

Responsive Search Ads (RSA):

Element	Limit	Qty	Rule
Headlines	30 chars	up to 15, min 3	Each must work standalone — Google combines randomly
Descriptions	90 chars	up to 4, min 2	Complement, don't repeat headlines
Display path	15 chars × 2	—

Pin sparingly — pinning drops Ad Strength and limits ML optimization. Supply ≥5 headlines + ≥5 descriptions for ~10% more conversions at same CPA.

Responsive Display Ads (RDA) — default Display format:

Asset	Spec	Qty
Landscape image	1200×628 (1.91:1)	up to 15 total
Square image	1200×1200 (1:1)	required
Portrait image	1200×1500 (4:5)	optional, expands inventory
Logo square	1200×1200 (128 min)	up to 5
Logo wide	1200×300 (4:1)	optional
Short headline	30 chars	up to 5
Long headline	90 chars	1
Description	90 chars	up to 5

All images ≤5MB, JPG/PNG only (no GIF in RDA). Keep file size <150KB for fast load.

Highest-inventory static sizes (if uploading fixed banners): 300×250 (medium rectangle — most served, works desktop+mobile), 728×90 (leaderboard), 320×50 (mobile banner), 300×600 (half-page — premium CPM, high CTR), 336×280.

Performance Max: same asset pool serves across Search/Display/YouTube/Gmail/Maps/Discover. Upload all 3 image ratios + a YouTube video (≤30s) or Google auto-generates one — don't let it.

LinkedIn Ads

Element	Limit
Intro text	150 chars rec (600 max)
Headline	70 chars rec (200 max)
Image	1200×627 (1.91:1) or 1200×1200

TikTok Ads

1080×1920, 9:16. Ad text: 100 char max (~80 visible). For Spark Ads (boosting organic creator posts), get the authorization code from the creator — Spark Ads outperform In-Feed Ads on engagement because they retain organic engagement metrics.

X / Twitter Ads

Placement	Dimensions	Notes
Single image	1200×675 (1.91:1) or 1080×1080 (1:1)	1:1 gets more real estate in feed
Carousel cards	1080×1080 (1:1) or 800×800 min	2-6 cards; each card gets its own image + optional headline/URL
Portrait (organic or promoted)	1080×1350 (4:5)	Taller images dominate mobile feed; 4:5 works for organic posts promoted via Amplify/Quick Promote

Element	Limit	Notes
Tweet text	280 chars	~100 chars visible before "Show more" on mobile when media is attached
Card headline	70 chars	Below image on website cards; not present on carousel image cards

Carousel strategy: Each card should be a self-contained visual — assume viewers swipe through quickly. Use carousels to show multiple visual worlds, product angles, or sequential narrative. High-performing carousels often use a consistent structural layout (same typography placement, same logo position) with varying visual treatments per card, so the brand stays recognizable while each card feels fresh. This creates a "same but different" effect that rewards swiping.

Step 3: Determine Campaign Mode

Before defining angles, determine whether this is a direct-response or awareness/launch campaign. The approach differs significantly.

Awareness / Launch Campaigns

Use for product launches, brand announcements, event teasers, or top-of-funnel awareness. These prioritize mood, intrigue, and brand impression over clicks.

Key differences from direct-response:

No CTA button on the image. The ad is a statement, not a prompt. The CTA lives in the post text or link, not the visual.
Even less text: ~10-15 words max on the image — typically just brand name, one positioning line, and a date or tagline.
Visual metaphor over product shots. Don't show the product. Use evocative imagery that communicates the brand's feeling or aspiration (e.g., a person in nature for "freedom," a glowing device for "future," a flower for "beauty/craft"). The imagery should make someone stop scrolling because it's beautiful or intriguing, not because it promises a specific outcome.
Multiple visual worlds per campaign. For carousels or multi-asset campaigns, create 3-5 distinct aesthetic treatments — different color palettes, different photographic subjects, different moods — while keeping typography layout and logo placement consistent. This "same structure, different world" pattern rewards swiping and creates brand depth.
Layout pattern: Logo (top corner) → Hero visual (fills most of the frame) → Brand name / headline (bottom, large, bold) → Subtitle + date (smaller, below headline).

When to use awareness mode:

Pre-launch teasers and announcements
Brand-building campaigns where the goal is recall, not clicks
Event promotion (dates, countdowns)
When the user says "announce," "tease," "launch," or asks for something that "looks premium" without conversion goals

Direct-Response Campaigns

Use for conversion, lead generation, app installs, purchases. This is the default mode — most ad requests are direct-response.

Step 3b: Define Angles (Direct-Response)

Before writing individual copy, establish 3-5 distinct angles — different reasons someone would click:

Category	Example
Pain point	"Stop wasting time on X"
Outcome	"Achieve Y in Z days"
Social proof	"Join 10,000+ teams who..."
Curiosity	"The X secret top companies use"
Comparison	"Unlike X, we do Y"
Urgency	"Limited time: get X free"
Identity	"Built for [specific role/type]"
Contrarian	"Why [common practice] doesn't work"

Step 4: Design Principles

Direct-Response Visual Hierarchy

Read order: 1) Hero element → 2) Benefit/offer → 3) CTA → 4) Logo (corner, small).

Rules:

<20 words total on the image — move everything else to primary text
One focal point. If the eye doesn't know where to land in 0.5s, it's too busy.
High contrast text/background — verify WCAG 4.5:1 minimum (use chroma.contrast() if building programmatically)
CTA button: contrasting color, rounded corners, verb-first ("Get the guide" not "Learn more")
Faces looking toward the CTA increase click-through (gaze cueing)
Keep file <150KB for display; Meta accepts up to 30MB but slow loads hurt auction performance

Awareness / Launch Visual Hierarchy

Read order: 1) Hero visual (dominates the frame) → 2) Brand name / headline (large, bold, bottom or mid-frame) → 3) Positioning line (smaller, beneath headline) → 4) Logo (top corner, subtle).

Rules:

~10-15 words max on the image. The visual does the heavy lifting.
No CTA button. The image is a brand impression, not a click prompt.
Use photography, 3D renders, or AI-generated imagery — not stock graphics, icons, or wireframes. The hero should feel editorial or art-directed.
Visual metaphor over literal representation. Show a feeling, not a feature. A glowing screen = innovation. A person with outstretched arms = freedom. A flower = craft/beauty. The viewer should feel something before they read anything.
Logo adapts to background: white logo on dark backgrounds, dark logo on light backgrounds. Keep it small and corner-positioned — it brands without competing with the hero.
For carousel/multi-card: maintain identical text layout and logo position across cards. Vary only the hero visual and color palette. This "same frame, different world" structure creates visual rhythm that rewards swiping.
Typography should be bold and confident — large headlines that anchor the bottom of the frame. Lowercase or sentence-case can feel more modern and approachable than ALL CAPS.

Step 5: Generate Variations

For each angle, generate multiple variations. Vary:

Word choice — synonyms, active vs. passive
Specificity — numbers vs. general claims ("Cut reporting time 75%" beats "Save time")
Tone — direct vs. question vs. command
Structure — short punch vs. full benefit statement

Strong headlines: Specific over vague. Benefits over features. Active voice. Include numbers when possible ("3x faster," "in 5 minutes").

Strong descriptions: Complement headlines, don't repeat them. Add proof points, handle objections ("No credit card required"), reinforce CTAs.

Step 6: Validate and Deliver

Before presenting, check every piece against character limits. Flag anything over and provide a trimmed alternative. Include character counts.


## Angle: [Pain Point — Manual Reporting]

### Headlines (30 char max)

1. "Stop Building Reports by Hand" (29)
2. "Automate Your Weekly Reports" (28)
3. "Reports in 5 Min, Not 5 Hrs" (27)

### Descriptions (90 char max)

1. "Marketing teams save 10+ hours/week with automated reporting. Start free." (73)
2. "Connect your data sources once. Get automated reports forever. No code required." (80)

Step 7: Build Ad Creatives — Design Subagents + Canvas

Launch a separate design subagent for each variation in parallel using startAsyncSubagent. Each subagent builds one ad variation as a standalone HTML page. This is much faster than building them sequentially.

Canvas Layout

Square (1:1) — default: Use 1080×1080 iframes. Group by angle in rows, with a landing page iframe alongside:

x=0,    y=0:    [Iframe: angle1-ad]              (1080x1080)
x=1200, y=0:    [Iframe: angle1-landing]         (1280x720)
x=0,    y=1180: [Iframe: angle2-ad]              (1080x1080)
x=1200, y=1180: [Iframe: angle2-landing]         (1280x720)

Portrait (9:16) — for stories/mobile/Instagram/TikTok: Use 608×1080 iframes. Arrange in rows of 5 with 50px gutters:

Row 1: x=0, x=658, x=1316, x=1974, x=2632  (y=0,    each 608x1080)
Row 2: x=0, x=658, x=1316, x=1974, x=2632  (y=1130, each 608x1080)

Parallel design subagent delegation

Launch all variations simultaneously. Each subagent gets one ad + its landing page. Use startAsyncSubagent for each, then waitForBackgroundTasks to collect results.

Before launching subagents, generate all hero images first using generateImage. Each ad needs its own unique, photorealistic hero image. Save them to artifacts/mockup-sandbox/public/images/ and reference them in the component via /__mockup/images/[filename].png.

// Step 1: Generate all hero images BEFORE launching design subagents
// Use generateImage from the media-generation skill
// Save to artifacts/mockup-sandbox/public/images/ad-[name].png
// Each image should be a different visual metaphor — same brand, different world

// Step 2: Launch design subagents in parallel — each builds one ad component
await startAsyncSubagent({
    task: `Create a production-ready ad creative component for the following angle.

Brand: [name]
Colors: Primary [hex], Secondary [hex], Accent [hex]
Fonts: [display font] (load from Google Fonts or use Inter)
Logo: [SVG description or path to uploaded logo]
User's stated style preferences: [include any specific preferences]

**Angle: [Pain Point]**
- Hero image: \`/__mockup/images/ad-[name].png\` (already generated — use this exact path)
- Headline: "[headline text]"
- Subtext: "[optional short line]"

Build this file: [ComponentName].tsx in artifacts/mockup-sandbox/src/components/mockups/[folder]/

**CRITICAL — Image-first full-bleed layout:**
The generated photo IS the ad. The component must use this exact structure:

\`\`\`
Outermost div: width: 100vw, height: 100vh, overflow: hidden, position: relative
Hero image: <img> with position: absolute, inset: 0, object-fit: cover, width: 100%, height: 100%
Bottom gradient overlay: position: absolute, bottom: 0, width: 100%, linear-gradient(transparent, rgba(0,0,0,0.7))
Text on top of gradient: position: absolute, bottom section
All font sizes in vw units (8-12vw for headlines in square, 8-20vw in portrait)
\`\`\`

Design rules:
- The hero image fills the ENTIRE frame — no margins, no panels, no colored backgrounds
- Text overlaid directly on the image with gradient for readability
- Headlines: HUGE, bold (font-weight 900), vw units
- Minimal text: brand name + one headline + one short subtext max
- Logo: small, top corner, SVG
- Use text-shadow for extra readability on busy backgrounds
- The vibe: if you saw this in your Instagram feed, you'd stop scrolling`,
    specialization: "DESIGN",
    relevantFiles: ["artifacts/mockup-sandbox/src/components/mockups/[folder]/[Component].tsx"]
});

// Repeat for each angle — all launch simultaneously
// Wait for all to complete
await waitForBackgroundTasks();

After all subagents finish, embed each page as an iframe on the canvas using apply_canvas_actions. Tell the user what was created and offer to focus the viewport.

Styling rules — no exceptions

Viewport-relative sizing. All ad HTML must use 100vw/100vh for the container and vw-based font sizes/padding. Never fixed pixel dimensions for the ad container. The ad must fill whatever iframe it's placed in without clipping or scrollbars.
Image-first layout. The generated hero image fills the entire frame (position: absolute; inset: 0; object-fit: cover). Text is overlaid on the image using a bottom gradient (linear-gradient(transparent, rgba(0,0,0,0.8))). No separate text panels, no side-by-side layouts, no colored backgrounds behind text. The image IS the ad.
Bold, massive typography. Headlines should be huge — 8-12vw for square, 8-20vw for portrait. Font weight 900 (black). Use text-shadow for readability against busy backgrounds. Headlines should feel like they belong on the image, not pasted on top.
Minimal text. Brand name, one headline, one short subtext line max. Move everything else to the platform's text fields. If you can remove a word, remove it.
No decorative gradients in direct-response ads. Use flat, solid colors for any non-image elements. Exception: the bottom readability gradient over the hero image is always allowed. Awareness/launch ads may use dramatic lighting, glows, and atmospheric gradients when they serve the visual metaphor.
Match the concept. If the user said "minimalist," don't add decorative elements. If they said "bold and energetic," don't make it muted. Re-read the user's stated preferences before delegating.
Consistency across angles. All angles should feel like they're from the same brand — same fonts, same color usage patterns, same visual language. The angles differ in message and hero image, not in design system. For awareness carousels: consistency means same text layout and logo placement, but each card can inhabit a completely different visual world (different photography, different color temperature, different mood).
Landing page = real page. The mock landing page should look like where the ad actually leads — hero section echoing the ad's message, value props, testimonials/social proof section, and a CTA. Not a wireframe, not a placeholder.

Export

The user can screenshot each iframe directly, or open the HTML files in a browser at the desired export size. Since the ads use vw/vh sizing, they'll adapt to any viewport. For batch export at specific dimensions: npx playwright screenshot angle1-ad.html --viewport-size=1080,1080.

Iterating from Performance Data

When the user provides performance data:

Analyze winners: Identify winning themes, structures, word patterns, and character utilization in top performers (by CTR, conversion rate, or ROAS)
Analyze losers: Identify themes that fall flat and common patterns in underperformers
Generate new variations: Double down on winning themes, extend winning angles, test 1-2 new unexplored angles, avoid patterns from underperformers
Document the iteration: Track what was learned, what's being tested, and what angles were retired

Research Before Writing

Use webSearch to find examples of top-performing ads in the user's vertical. Search for ad breakdowns, swipe files, and case studies — e.g. webSearch("[industry] top performing Facebook ads 2026") or webSearch("[industry] TikTok ad examples"). The TikTok Creative Center and Meta Ad Library are useful reference sites but require direct browser interaction to filter; web search can surface articles and analyses that reference their data. Reverse-engineer: what hook, what angle, what visual pattern. Don't guess what works — look it up.

Common Mistakes

Writing RSA headlines that only work in sequence (Google combines them randomly — each must stand alone)
Ignoring the Reels 72-char visible limit (writing for the 125-char feed limit → truncated on Reels)
All variations = same angle reworded (vary the psychology, not the synonyms)
Placing text in the bottom 35% of a 9:16 ad (covered by UI on every platform)
Retiring creative before 1,000+ impressions per variant
Letting Performance Max auto-generate video — always supply your own
Putting a CTA button on awareness/launch ads (these should create intrigue and brand recall, not push for immediate clicks — the CTA belongs in the post text)
Using the same visual treatment for every card in a carousel (each card should be a different visual world; if they all look the same, there's no reward for swiping)
Showing literal product screenshots in a launch teaser (before launch, use visual metaphors that communicate the brand feeling — save product shots for post-launch conversion campaigns)
Defaulting to direct-response design when the goal is awareness (a launch announcement needs mood and mystique, not a "Sign up now" button)
Building ads with CSS-only backgrounds instead of generated images (the image IS the ad — always generate photorealistic hero imagery unless explicitly told not to)
Placing text in a separate panel next to the image (text goes ON TOP of the image with a gradient, not beside it)
Using abstract digital art or generic "tech" imagery (go editorial: real objects, dramatic lighting, tangible subjects)
Forgetting to prompt the user for their real logo at the end

Post-Build: Logo Prompt

After delivering the ad creatives, always check whether the user has provided their actual logo. If they haven't (i.e., you used a placeholder SVG or described the logo in code), end your response with a plain-text note like:

"If you'd like to swap in your real logo, just upload or attach it and I'll drop it into all the ads."

This is a simple text reminder — not a tool call, not a modal, just a natural closing line. Skip this if the user already provided their logo file.

Limitations

Cannot run or measure ad campaigns
Cannot access ad platform analytics
Cannot create animated or video ads
Image generation requires the media-generation skill

name	ad-creative
description	Design static ad creatives for social media and display advertising campaigns.

Ad Creative Maker

Design static ad creatives for social media ads, display banners, and digital advertising campaigns. Build production-ready ads via the design subagent and present them as iframes on the canvas.

When to Use

User needs ad creatives for Facebook, Instagram, LinkedIn, Google Display, or TikTok
User wants banner ads or display advertising assets
User needs multiple ad variants for A/B testing
User wants ad copy and visual design together
User wants to iterate on ad creative based on performance data

When NOT to Use

Organic social media content (use content-machine skill)
Video ads or animated content (use storyboard skill for planning)
Full landing pages (use the artifacts skill)

Image-First Design Philosophy

Always generate images using generateImage from the media-generation skill unless:

The user uploads/attaches their own product photos or brand imagery
The ad is purely typographic by explicit request
The platform is text-only (e.g., Google RSAs)

Prompt strategy for generated images:

Lead with a specific, tangible subject (object, scene, or person)
Specify lighting style (studio, dramatic, golden hour, Rembrandt)
Specify camera quality ("shot on Hasselblad," "editorial photography," "fine art")
Add negative prompts: "text, words, letters, logos, watermark, blurry, low quality, cartoon, illustration"
Each ad in a set should use a different visual metaphor — same brand, different visual world

Default Format

Default to square (1:1) for ad creatives. Use portrait/iPhone format (9:16, iframe size 608×1080) when:

User asks for Instagram Stories or Reels ads
User asks for TikTok ads
User mentions "mobile ads," "iPhone," "portrait," or "stories format"
User asks for 9:16 or vertical format

Methodology

Step 1: Creative Brief

Gather these inputs:

Platform & Format: Which platform? (Google Ads, Meta, LinkedIn, TikTok, X/Twitter) Which format? (Search RSAs, feed, stories, display)
Objective: Awareness, consideration, or conversion?
Target audience: Who will see this ad? What stage of awareness? (Problem-aware, solution-aware, product-aware)
Key message: Single most important thing to communicate
CTA: What action should the viewer take?
Brand assets: Logo, colors, fonts, product images. If the user hasn't provided their actual logo, note this and ask at the end.
Performance data (if iterating): Which headlines/descriptions are performing best/worst? What angles have been tested?

Step 2: Platform Specifications (2025-2026)

Enforce these programmatically. Count characters in code, don't eyeball.

Meta (Facebook/Instagram) — Visual

Placement	Dimensions	Safe zone (keep critical elements inside)
Feed (square)	1080×1080	~100px margin all edges
Feed (portrait) — preferred	1080×1350 (4:5)	4:5 outperforms 1:1 on CTR; mobile-first
Stories	1080×1920	Top 14% (270px) + bottom 20% (380px) = dead zones
Reels	1080×1920	Top 14% + bottom 35% (670px) — like/comment/share UI is taller
Universal safe core	1010×1280 centered	Design inside this box → works everywhere

Upload at 2x pixel density (2160×2160 for a 1080 slot) for Retina sharpness. JPG/PNG, 30MB max.

20% text rule: officially removed — Meta no longer rejects text-heavy images, but the delivery algorithm still quietly throttles them. Keep text minimal; move details to primary text.

Meta — Text

Element	Limit	Notes
Primary text	125 chars visible feed / ~72 chars visible in Reels	Write for 72
Headline	40 chars rec	Below image
Description	30 chars rec	Often hidden on mobile

Google Ads

Responsive Search Ads (RSA):

Element	Limit	Qty	Rule
Headlines	30 chars	up to 15, min 3	Each must work standalone — Google combines randomly
Descriptions	90 chars	up to 4, min 2	Complement, don't repeat headlines
Display path	15 chars × 2	—

Pin sparingly — pinning drops Ad Strength and limits ML optimization. Supply ≥5 headlines + ≥5 descriptions for ~10% more conversions at same CPA.

Responsive Display Ads (RDA) — default Display format:

Asset	Spec	Qty
Landscape image	1200×628 (1.91:1)	up to 15 total
Square image	1200×1200 (1:1)	required
Portrait image	1200×1500 (4:5)	optional, expands inventory
Logo square	1200×1200 (128 min)	up to 5
Logo wide	1200×300 (4:1)	optional
Short headline	30 chars	up to 5
Long headline	90 chars	1
Description	90 chars	up to 5

All images ≤5MB, JPG/PNG only (no GIF in RDA). Keep file size <150KB for fast load.

Performance Max: same asset pool serves across Search/Display/YouTube/Gmail/Maps/Discover. Upload all 3 image ratios + a YouTube video (≤30s) or Google auto-generates one — don't let it.

LinkedIn Ads

Element	Limit
Intro text	150 chars rec (600 max)
Headline	70 chars rec (200 max)
Image	1200×627 (1.91:1) or 1200×1200

TikTok Ads

X / Twitter Ads

Placement	Dimensions	Notes
Single image	1200×675 (1.91:1) or 1080×1080 (1:1)	1:1 gets more real estate in feed
Carousel cards	1080×1080 (1:1) or 800×800 min	2-6 cards; each card gets its own image + optional headline/URL
Portrait (organic or promoted)	1080×1350 (4:5)	Taller images dominate mobile feed; 4:5 works for organic posts promoted via Amplify/Quick Promote

Element	Limit	Notes
Tweet text	280 chars	~100 chars visible before "Show more" on mobile when media is attached
Card headline	70 chars	Below image on website cards; not present on carousel image cards

Step 3: Determine Campaign Mode

Before defining angles, determine whether this is a direct-response or awareness/launch campaign. The approach differs significantly.

Awareness / Launch Campaigns

Use for product launches, brand announcements, event teasers, or top-of-funnel awareness. These prioritize mood, intrigue, and brand impression over clicks.

Key differences from direct-response:

No CTA button on the image. The ad is a statement, not a prompt. The CTA lives in the post text or link, not the visual.
Even less text: ~10-15 words max on the image — typically just brand name, one positioning line, and a date or tagline.
Visual metaphor over product shots. Don't show the product. Use evocative imagery that communicates the brand's feeling or aspiration (e.g., a person in nature for "freedom," a glowing device for "future," a flower for "beauty/craft"). The imagery should make someone stop scrolling because it's beautiful or intriguing, not because it promises a specific outcome.
Multiple visual worlds per campaign. For carousels or multi-asset campaigns, create 3-5 distinct aesthetic treatments — different color palettes, different photographic subjects, different moods — while keeping typography layout and logo placement consistent. This "same structure, different world" pattern rewards swiping and creates brand depth.
Layout pattern: Logo (top corner) → Hero visual (fills most of the frame) → Brand name / headline (bottom, large, bold) → Subtitle + date (smaller, below headline).

When to use awareness mode:

Pre-launch teasers and announcements
Brand-building campaigns where the goal is recall, not clicks
Event promotion (dates, countdowns)
When the user says "announce," "tease," "launch," or asks for something that "looks premium" without conversion goals

Direct-Response Campaigns

Use for conversion, lead generation, app installs, purchases. This is the default mode — most ad requests are direct-response.

Step 3b: Define Angles (Direct-Response)

Before writing individual copy, establish 3-5 distinct angles — different reasons someone would click:

Category	Example
Pain point	"Stop wasting time on X"
Outcome	"Achieve Y in Z days"
Social proof	"Join 10,000+ teams who..."
Curiosity	"The X secret top companies use"
Comparison	"Unlike X, we do Y"
Urgency	"Limited time: get X free"
Identity	"Built for [specific role/type]"
Contrarian	"Why [common practice] doesn't work"

Step 4: Design Principles

Direct-Response Visual Hierarchy

Read order: 1) Hero element → 2) Benefit/offer → 3) CTA → 4) Logo (corner, small).

Rules:

<20 words total on the image — move everything else to primary text
One focal point. If the eye doesn't know where to land in 0.5s, it's too busy.
High contrast text/background — verify WCAG 4.5:1 minimum (use chroma.contrast() if building programmatically)
CTA button: contrasting color, rounded corners, verb-first ("Get the guide" not "Learn more")
Faces looking toward the CTA increase click-through (gaze cueing)
Keep file <150KB for display; Meta accepts up to 30MB but slow loads hurt auction performance

Awareness / Launch Visual Hierarchy

Rules:

~10-15 words max on the image. The visual does the heavy lifting.
No CTA button. The image is a brand impression, not a click prompt.
Use photography, 3D renders, or AI-generated imagery — not stock graphics, icons, or wireframes. The hero should feel editorial or art-directed.
Visual metaphor over literal representation. Show a feeling, not a feature. A glowing screen = innovation. A person with outstretched arms = freedom. A flower = craft/beauty. The viewer should feel something before they read anything.
Logo adapts to background: white logo on dark backgrounds, dark logo on light backgrounds. Keep it small and corner-positioned — it brands without competing with the hero.
For carousel/multi-card: maintain identical text layout and logo position across cards. Vary only the hero visual and color palette. This "same frame, different world" structure creates visual rhythm that rewards swiping.
Typography should be bold and confident — large headlines that anchor the bottom of the frame. Lowercase or sentence-case can feel more modern and approachable than ALL CAPS.

Step 5: Generate Variations

For each angle, generate multiple variations. Vary:

Word choice — synonyms, active vs. passive
Specificity — numbers vs. general claims ("Cut reporting time 75%" beats "Save time")
Tone — direct vs. question vs. command
Structure — short punch vs. full benefit statement

Strong headlines: Specific over vague. Benefits over features. Active voice. Include numbers when possible ("3x faster," "in 5 minutes").

Strong descriptions: Complement headlines, don't repeat them. Add proof points, handle objections ("No credit card required"), reinforce CTAs.

Step 6: Validate and Deliver

Before presenting, check every piece against character limits. Flag anything over and provide a trimmed alternative. Include character counts.


## Angle: [Pain Point — Manual Reporting]

### Headlines (30 char max)

1. "Stop Building Reports by Hand" (29)
2. "Automate Your Weekly Reports" (28)
3. "Reports in 5 Min, Not 5 Hrs" (27)

### Descriptions (90 char max)

1. "Marketing teams save 10+ hours/week with automated reporting. Start free." (73)
2. "Connect your data sources once. Get automated reports forever. No code required." (80)

Step 7: Build Ad Creatives — Design Subagents + Canvas

Canvas Layout

Square (1:1) — default: Use 1080×1080 iframes. Group by angle in rows, with a landing page iframe alongside:

x=0,    y=0:    [Iframe: angle1-ad]              (1080x1080)
x=1200, y=0:    [Iframe: angle1-landing]         (1280x720)
x=0,    y=1180: [Iframe: angle2-ad]              (1080x1080)
x=1200, y=1180: [Iframe: angle2-landing]         (1280x720)

Portrait (9:16) — for stories/mobile/Instagram/TikTok: Use 608×1080 iframes. Arrange in rows of 5 with 50px gutters:

Row 1: x=0, x=658, x=1316, x=1974, x=2632  (y=0,    each 608x1080)
Row 2: x=0, x=658, x=1316, x=1974, x=2632  (y=1130, each 608x1080)

Parallel design subagent delegation

Launch all variations simultaneously. Each subagent gets one ad + its landing page. Use startAsyncSubagent for each, then waitForBackgroundTasks to collect results.

// Step 1: Generate all hero images BEFORE launching design subagents
// Use generateImage from the media-generation skill
// Save to artifacts/mockup-sandbox/public/images/ad-[name].png
// Each image should be a different visual metaphor — same brand, different world

// Step 2: Launch design subagents in parallel — each builds one ad component
await startAsyncSubagent({
    task: `Create a production-ready ad creative component for the following angle.

Brand: [name]
Colors: Primary [hex], Secondary [hex], Accent [hex]
Fonts: [display font] (load from Google Fonts or use Inter)
Logo: [SVG description or path to uploaded logo]
User's stated style preferences: [include any specific preferences]

**Angle: [Pain Point]**
- Hero image: \`/__mockup/images/ad-[name].png\` (already generated — use this exact path)
- Headline: "[headline text]"
- Subtext: "[optional short line]"

Build this file: [ComponentName].tsx in artifacts/mockup-sandbox/src/components/mockups/[folder]/

**CRITICAL — Image-first full-bleed layout:**
The generated photo IS the ad. The component must use this exact structure:

\`\`\`
Outermost div: width: 100vw, height: 100vh, overflow: hidden, position: relative
Hero image: <img> with position: absolute, inset: 0, object-fit: cover, width: 100%, height: 100%
Bottom gradient overlay: position: absolute, bottom: 0, width: 100%, linear-gradient(transparent, rgba(0,0,0,0.7))
Text on top of gradient: position: absolute, bottom section
All font sizes in vw units (8-12vw for headlines in square, 8-20vw in portrait)
\`\`\`

Design rules:
- The hero image fills the ENTIRE frame — no margins, no panels, no colored backgrounds
- Text overlaid directly on the image with gradient for readability
- Headlines: HUGE, bold (font-weight 900), vw units
- Minimal text: brand name + one headline + one short subtext max
- Logo: small, top corner, SVG
- Use text-shadow for extra readability on busy backgrounds
- The vibe: if you saw this in your Instagram feed, you'd stop scrolling`,
    specialization: "DESIGN",
    relevantFiles: ["artifacts/mockup-sandbox/src/components/mockups/[folder]/[Component].tsx"]
});

// Repeat for each angle — all launch simultaneously
// Wait for all to complete
await waitForBackgroundTasks();

After all subagents finish, embed each page as an iframe on the canvas using apply_canvas_actions. Tell the user what was created and offer to focus the viewport.

Styling rules — no exceptions

Viewport-relative sizing. All ad HTML must use 100vw/100vh for the container and vw-based font sizes/padding. Never fixed pixel dimensions for the ad container. The ad must fill whatever iframe it's placed in without clipping or scrollbars.
Image-first layout. The generated hero image fills the entire frame (position: absolute; inset: 0; object-fit: cover). Text is overlaid on the image using a bottom gradient (linear-gradient(transparent, rgba(0,0,0,0.8))). No separate text panels, no side-by-side layouts, no colored backgrounds behind text. The image IS the ad.
Bold, massive typography. Headlines should be huge — 8-12vw for square, 8-20vw for portrait. Font weight 900 (black). Use text-shadow for readability against busy backgrounds. Headlines should feel like they belong on the image, not pasted on top.
Minimal text. Brand name, one headline, one short subtext line max. Move everything else to the platform's text fields. If you can remove a word, remove it.
No decorative gradients in direct-response ads. Use flat, solid colors for any non-image elements. Exception: the bottom readability gradient over the hero image is always allowed. Awareness/launch ads may use dramatic lighting, glows, and atmospheric gradients when they serve the visual metaphor.
Match the concept. If the user said "minimalist," don't add decorative elements. If they said "bold and energetic," don't make it muted. Re-read the user's stated preferences before delegating.
Consistency across angles. All angles should feel like they're from the same brand — same fonts, same color usage patterns, same visual language. The angles differ in message and hero image, not in design system. For awareness carousels: consistency means same text layout and logo placement, but each card can inhabit a completely different visual world (different photography, different color temperature, different mood).
Landing page = real page. The mock landing page should look like where the ad actually leads — hero section echoing the ad's message, value props, testimonials/social proof section, and a CTA. Not a wireframe, not a placeholder.

Export

Iterating from Performance Data

When the user provides performance data:

Analyze winners: Identify winning themes, structures, word patterns, and character utilization in top performers (by CTR, conversion rate, or ROAS)
Analyze losers: Identify themes that fall flat and common patterns in underperformers
Generate new variations: Double down on winning themes, extend winning angles, test 1-2 new unexplored angles, avoid patterns from underperformers
Document the iteration: Track what was learned, what's being tested, and what angles were retired

Research Before Writing

Common Mistakes

Writing RSA headlines that only work in sequence (Google combines them randomly — each must stand alone)
Ignoring the Reels 72-char visible limit (writing for the 125-char feed limit → truncated on Reels)
All variations = same angle reworded (vary the psychology, not the synonyms)
Placing text in the bottom 35% of a 9:16 ad (covered by UI on every platform)
Retiring creative before 1,000+ impressions per variant
Letting Performance Max auto-generate video — always supply your own
Putting a CTA button on awareness/launch ads (these should create intrigue and brand recall, not push for immediate clicks — the CTA belongs in the post text)
Using the same visual treatment for every card in a carousel (each card should be a different visual world; if they all look the same, there's no reward for swiping)
Showing literal product screenshots in a launch teaser (before launch, use visual metaphors that communicate the brand feeling — save product shots for post-launch conversion campaigns)
Defaulting to direct-response design when the goal is awareness (a launch announcement needs mood and mystique, not a "Sign up now" button)
Building ads with CSS-only backgrounds instead of generated images (the image IS the ad — always generate photorealistic hero imagery unless explicitly told not to)
Placing text in a separate panel next to the image (text goes ON TOP of the image with a gradient, not beside it)
Using abstract digital art or generic "tech" imagery (go editorial: real objects, dramatic lighting, tangible subjects)
Forgetting to prompt the user for their real logo at the end

Post-Build: Logo Prompt

"If you'd like to swap in your real logo, just upload or attach it and I'll drop it into all the ads."

This is a simple text reminder — not a tool call, not a modal, just a natural closing line. Skip this if the user already provided their logo file.

Limitations

Cannot run or measure ad campaigns
Cannot access ad platform analytics
Cannot create animated or video ads
Image generation requires the media-generation skill