Ad Creative Generation

Ad Creative Generation

Generate ad creatives — both visuals and copy — that match a brand's identity. This skill orchestrates brand extraction, copywriting, and image generation into a single workflow.

Requirements

This skill assumes the Hyper MCP is connected to your agent so the tools below are available. Brand extraction also requires Firecrawl to be configured under your Hyper integrations.

Tool surface

For deeper image-tool selection guidance see the image-generation skill. To turn finished creatives into running campaigns see google-ads and meta-ads.

Out of scope

• Picking the right image model when the task isn't ad creative — use image-generation.
• Creating, launching, or budgeting actual ad campaigns — use google-ads, meta-ads, tiktok-ads, pinterest-ads, or amazon-ads.
• Searching live competitor ads — use meta-ads-library.

Critical rules

• Always extract branding first before generating visuals for a website/brand.
• Always pass both logo and screenshot as reference images — SVG logos are auto-converted to PNG.
• DO NOT display image URLs — they are automatically shown in chat.
• Read `brand.screenshot.description` to understand the product — never guess from the company name.
• Default to logo + headline + product screenshot for SaaS / product ads unless the user requests a different style.
• Use OpenAI for the initial ad creative by default — openai_image_edit for branded / reference-based work, openai_image_generation for loose first-pass concepts.
• Do not default to Nano Banana for the first creative pass — use it when the user explicitly asks or when readable text inside the image is the main requirement.
• `openai_image_edit` supported sizes: 1024x1024 (square), 1024x1536 (portrait / story), 1536x1024 (landscape / banner). Match to platform placement.
• Respect character limits — Google RSA headlines are 30 chars, descriptions 90 chars; Meta primary text is 125 chars visible.
• Change one variable per variant so test results are attributable.
• Match aspect ratio to placement — feed is 1:1, story is 9:16 (use 1024x1536), banner is 16:9 (use 1536x1024).

Routing Table

Based on what the user needs, read the appropriate reference file:

When the task spans multiple areas (e.g., "create ad creatives for this website"), follow the full workflow below.

Core Workflow

Phase 1: Brand Extraction

Call firecrawl_extract_branding with the website URL. This single call returns branding data, a saved logo, and a website screenshot:

brand = firecrawl_extract_branding(url="https://example.com")

The result contains:

• brand.logo.file_id — logo image to use as reference
• brand.screenshot.file_id — website screenshot to use as reference
• brand.branding — colors, typography, spacing, personality

Both logo and screenshot are saved images ready to pass as reference_images to image generation tools. SVG logos are automatically converted to PNG, so always include the logo — no need to filter by format.

For deeper guidance on brand extraction, read `references/brand-extraction.md`.

Phase 2: Ad Copy Generation

Using the brand personality, tone, and value proposition from the extraction, write ad copy variants. Structure copy by what's being tested:

1. Hook variants — different opening angles (number-led, question, pain-point, benefit-first) 2. Body variants — different messaging (social proof, feature highlight, urgency) 3. CTA variants — different calls to action (Learn More, Shop Now, Get Started, Try Free)

Change only one element per variant so results can be attributed to specific changes.

For copy frameworks, hook patterns, and variant strategy, read `references/ad-copy-frameworks.md`.

Phase 3: Visual Creative Generation

Generate ad images using brand assets as references. Always pass both logo and screenshot as reference images — SVG logos are automatically converted to PNG by the branding tool.

Read `brand.screenshot.description` to understand what the product actually does and what the UI looks like. Do not guess from the company name.

Preferred tool for the first creative pass: `openai_image_edit`

Default approach (best practice for SaaS / product ads)

By default, a strong ad creative has three layers: brand (logo), copy (headline), and product (realistic screenshot showing the core value prop). This is the recommended starting point when the user hasn't specified a creative direction. Use OpenAI for this initial composition unless the user explicitly asks for Nano Banana or the image is primarily a text-rendering task.

openai_image_edit(
    requests=[{
        "prompt": (
            "Social media ad creative for [company name]. "
            "Top: the [company] logo. "
            "Headline: '[headline from ad copy phase or site hero text]'. "
            "Below the headline: a clean, realistic product screenshot of [describe the actual UI "
            "based on brand.screenshot.description]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    size="1024x1024",
    quality="high"
)

Following user direction

If the user asks for a specific creative style (lifestyle imagery, abstract, illustration, people using the product, etc.), follow their direction. The defaults above are a starting point, not a constraint. The reference images still supply brand consistency regardless of creative direction.

Things to avoid (unless the user specifically asks)

• Made-up visual elements (robot mascots, random platform logos, abstract graphics)
• UI chrome (buttons, nav bars, form inputs) — ad platforms add their own CTAs
• Guessing what the product does from the company name
• Generic marketing clip art

Visual Tool Selection

Platform Quick Reference

For full platform specs, read `references/google-ads-creatives.md` or `references/meta-ads-creatives.md`.

Ad Copy Variant Quick Reference

When generating variants from existing top performers:

1. Identify the winning elements: hook type, CTA, messaging angle, format 2. Generate variants that preserve winning elements while changing one variable 3. Group variants by what's being tested (hook, body, CTA) 4. Pair variants for A/B testing — each pair should isolate one variable

For detailed frameworks, read `references/ad-copy-frameworks.md`.

• Follow user direction — if the user wants lifestyle, illustration, abstract, or any other style, follow their lead. The reference images still supply brand consistency regardless of creative direction.

Ad Copy Frameworks

Frameworks for writing ad copy variants and structuring A/B tests.

Hook Patterns

The hook is the first line of ad copy — it determines whether someone keeps reading. Each pattern works differently depending on audience and offer.

Number-Led Hook

Leads with a specific, concrete number. High CTR when the number is surprising or aspirational.

• "Cut your onboarding time from 3 weeks to 3 days"
• "47% of teams switch within the first month"
• "Save 12 hours per week on reporting"

Question Hook

Opens with a question that the target audience would answer "yes" to. Creates immediate relevance.

• "Still manually reconciling invoices every month?"
• "What if your team could ship 2x faster?"
• "Tired of tools that promise AI but deliver spreadsheets?"

Pain-Point Hook

Names a specific frustration the audience experiences. Works best for problem-aware audiences.

• "Your CRM shouldn't need a full-time admin"
• "Scattered tools. Missed deadlines. Sound familiar?"
• "Manual data entry is costing you more than you think"

Benefit-First Hook

Leads with the outcome, not the problem. Works for solution-aware audiences.

• "Automate your entire onboarding workflow in one click"
• "Get real-time insights without writing a single query"
• "Ship landing pages in minutes, not sprints"

Social Proof Hook

Opens with credibility — customer results, user count, or testimonials.

• "Join 10,000+ teams that ditched manual reporting"
• "'We cut our CAC by 40% in the first quarter' — [Customer]"
• "Rated #1 on G2 for ease of use"

Body Frameworks

Problem-Agitate-Solve (PAS)

1. State the problem 2. Amplify the pain 3. Present the solution

Best for: Cold audiences, pain-point messaging

Feature-Advantage-Benefit (FAB)

1. Name the feature 2. Explain what it enables 3. State the outcome for the user

Best for: Product-aware audiences, feature launches

Before-After-Bridge (BAB)

1. Describe current state (before) 2. Describe desired state (after) 3. Show how to get there (bridge = your product)

Best for: Aspirational messaging, transformation offers

CTA Types

Match CTA to funnel stage. Don't use "Shop Now" for awareness campaigns or "Learn More" for retargeting.

Variant Generation Strategy

The One-Variable Rule

Each variant should change exactly one element from the control. This lets you attribute performance differences to specific changes.

Example from a winning control ad:

Hook: "Cut your onboarding time from 3 weeks to 3 days"

Body: Customer quote about time savings

CTA: "Get Started Free"

Variant set (8 variants):

Hook variants (change hook, keep body + CTA): 1. Different metric: "Reduce support tickets by 60% in 30 days" 2. Different metric: "Onboard new hires in hours, not weeks" 3. Question-based: "What if onboarding took 3 days instead of 3 weeks?" 4. Pain-point: "3-week onboarding is costing you your best hires"

Body variants (change body, keep hook + CTA): 5. Feature highlight instead of social proof 6. Stat-driven instead of quote-driven

CTA variants (change CTA, keep hook + body): 7. "Book a Demo" instead of "Get Started Free" 8. "See How It Works" instead of "Get Started Free"

Variant Grouping

Always group and label variants by what's being tested:

GROUP A: Hook Testing (4 variants)
  - A1: Number-led (different metric)
  - A2: Number-led (different framing)
  - A3: Question-based hook
  - A4: Pain-point hook

GROUP B: Body Testing (2 variants)
  - B1: Feature-focused body
  - B2: Data-driven body

GROUP C: CTA Testing (2 variants)
  - C1: Higher-commitment CTA
  - C2: Curiosity-driven CTA

A/B Test Pairing

Pair variants for testing so each pair isolates one variable:

Test Priority

Prioritize testing based on which element has the most variance in existing performance data:

1. Hooks first — highest impact on CTR. If top ads have wildly different hook styles, test hooks. 2. Body second — impacts conversion rate. Test if CTR is stable but CVR varies. 3. CTA last — smallest effect size. Test when hook and body are optimized.

Analyzing Existing Ad Performance

When the user provides performance data from existing ads:

1. Correlate metrics with copy elements — which hook types get highest CTR? Which body styles drive CVR? 2. Identify winning patterns — not just the best ad, but what pattern the top 3-5 ads share 3. Find the gap — what hasn't been tested yet? If all top ads use number-led hooks, a question hook is worth testing 4. Preserve winners, vary strategically — new variants should keep the winning pattern and change one thing

Output Format

When delivering ad copy variants, always include:

1. Variants grouped by test variable (hook, body, CTA) 2. Annotation showing which winning element each variant preserves 3. Recommended test priority 4. Platform-specific formatting (character limits respected) 5. A/B test pairings showing which variants to test against each other

Brand Extraction

Extract brand identity from a website for use in ad creative generation.

Single-Call Workflow

firecrawl_extract_branding extracts branding data, downloads the logo, and captures a screenshot in one call:

brand = firecrawl_extract_branding(url="https://example.com")

Return Shape

brand.logo.file_id        -- File ID of the saved logo (pass to image gen)
brand.logo.path           -- Storage path of the saved logo
brand.logo.description    -- "Brand logo for Example"

brand.screenshot.file_id  -- File ID of the saved screenshot (pass to image gen)
brand.screenshot.path     -- Storage path of the saved screenshot
brand.screenshot.description -- AI description of the page

brand.branding            -- Dict with full brand data:
  .colors                 -- primary, accent, background, text (hex codes)
  .typography             -- fonts, font families, sizes
  .spacing                -- base units, border radius
  .components             -- button styles, input styles
  .personality            -- brand tone and voice
  .designSystem           -- overall design system summary
  .confidence             -- extraction confidence scores

brand.file                -- Saved branding JSON file (NOT an image)

Critical: brand.file is the JSON data file. Never pass it to image gen tools. Only brand.logo.file_id and brand.screenshot.file_id are images.

Using Brand Assets in Image Generation

Always pass both logo and screenshot as reference images. SVG logos are automatically converted to PNG by the extraction tool, so always include the logo regardless of the original format.

reference_images = [brand.logo.file_id, brand.screenshot.file_id]

The logo provides brand mark consistency. The screenshot shows the actual product UI so the model can derive both the visual style and generate realistic product shots.

Default creative pattern (SaaS / product ads)

By default, compose: logo + headline + realistic product screenshot. Describe the product screenshot based on brand.screenshot.description, not guessed.

openai_image_edit(
    requests=[{
        "prompt": (
            "Social media ad creative for [company name]. "
            "Top: the [company] logo. "
            "Headline: '[headline]'. "
            "Below: a clean, realistic product screenshot of [describe actual UI from brand.screenshot.description]. "
            "Match the brand style from the references."
        ),
        "reference_images": reference_images
    }],
    size="1024x1024",
    quality="high"
)

This is the recommended default. If the user asks for a different creative direction (lifestyle, illustration, abstract, etc.), follow their lead — the references still provide brand consistency.

Using Brand Data for Ad Copy

The extracted brand data is primarily useful for copy, not image prompts.

Copy Tone from Brand Personality

Use brand.branding["personality"] to set the voice for ad copy:

• Professional → authoritative, data-driven, clear value props
• Playful → conversational, emoji-friendly, informal CTAs
• Technical → precise language, feature-focused, specs matter
• Luxury → aspirational, exclusive language, understated claims

Colors in Copy (not image prompts)

Brand colors inform copy context (e.g., knowing the brand is a fintech vs. a kids' toy brand), but do NOT paste hex codes into image prompts — the reference images already contain the actual colors.

Cached Results

firecrawl_extract_branding caches results by domain. Subsequent calls for the same domain return cached data without re-fetching. Use refresh=True to force a fresh extraction:

brand = firecrawl_extract_branding(url="https://example.com", refresh=True)

Google Ads Creatives

Specs and guidelines for generating Google Ads creative assets.

Responsive Search Ads (RSAs)

RSAs are Google's primary text ad format. You provide multiple headlines and descriptions; Google assembles combinations and optimizes.

Asset Requirements

Headline Rules

• Max 30 characters including spaces
• No exclamation marks in headlines (Google policy)
• No excessive capitalization (e.g., "FREE SHIPPING" is rejected)
• Include the target keyword in at least 2-3 headlines
• Each headline should be able to stand alone (Google may show any combination)
• Don't repeat the same message across headlines — each should offer a different angle

Headline Strategy (15 headlines)

Distribute across these categories:

Description Rules

• Max 90 characters including spaces
• Expand on what headlines introduce — don't repeat them
• Include a clear value proposition
• End with a CTA when possible
• First description is shown most often — make it strongest

Description Strategy (4 descriptions)

1. Primary value prop — the core reason to click (shown most often) 2. Feature + benefit — specific capability and its outcome 3. Social proof / credibility — results, ratings, customer count 4. CTA + offer — what they get and how to get it

Pin Positions

You can pin specific headlines / descriptions to positions:

• Pin sparingly — pinning reduces Google's optimization ability
• Pin your brand to Headline 1 if brand recognition matters
• Pin your CTA to Headline 3 to ensure it always appears
• Never pin all positions — defeats the purpose of RSAs

Display Ads

Responsive Display Ads

Google's responsive display format assembles your assets into ads across the Display Network.

Image Guidelines

• File size: max 5MB
• No text overlays covering more than 20% of the image
• High contrast between subject and background
• Product or service should be clearly visible
• Avoid borders or excessive whitespace

Generating Display Ad Images

Every creative needs three layers: brand (logo), copy (headline), and product (realistic screenshot). Use brand.screenshot.description to describe the actual product UI.

Square (1:1):

openai_image_edit(
    requests=[{
        "prompt": (
            "Display ad creative for [company name]. "
            "Logo at top. Headline: '[headline]'. "
            "Below: realistic product screenshot of [describe actual UI from brand.screenshot.description]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    size="1024x1024",
    quality="high"
)

Landscape (1.91:1):

openai_image_edit(
    requests=[{
        "prompt": (
            "Wide display ad creative for [company name]. "
            "Logo at top left. Headline: '[headline]'. "
            "Right side: realistic product screenshot of [describe actual UI]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    size="1536x1024",
    quality="high"
)

Performance Max

Performance Max campaigns use all Google surfaces. Asset requirements are a superset of RSA + Display.

Additional Assets for PMax

Asset Group Strategy

• Each asset group should target a distinct audience theme
• Provide maximum assets per group for best optimization
• Use high-quality images that represent the product / service
• Include both lifestyle and product-focused images

Google Ads Copy Rules

Policies to Follow

• No misleading claims or exaggeration
• No exclamation marks in headlines
• No ALL CAPS (except standard acronyms like "AI", "CRM")
• No phone numbers in ad text
• No trademarked terms without authorization
• Prices and discounts must be accurate and current
• Landing page must match the ad's offer

Quality Score Factors

Ad copy directly impacts Quality Score:

1. Relevance — copy must match the target keyword intent 2. Expected CTR — compelling copy improves predicted click-through 3. Landing page experience — ad promise must be fulfilled on the page

Write copy that closely matches what users are searching for and what the landing page delivers.

Meta Ads Creatives

Specs and guidelines for generating Meta (Facebook / Instagram) ad creative assets.

Text Specs

Primary Text

The main body text that appears above the image / video.

The first 125 characters are critical — everything after is truncated behind "See more". Front-load the hook and value proposition.

Headline

Appears below the image, next to the CTA button.

Keep headlines punchy. They compete with the CTA button for attention.

Description

Appears below the headline on some placements (not always shown).

Don't rely on the description for critical info — it's often hidden.

Image Specs by Placement

Feed Ads (Facebook + Instagram)

4:5 takes more vertical space in the feed, increasing visibility. Use 1:1 for broad compatibility.

Story / Reel Ads (Facebook + Instagram)

Leave ~250px clear at top (profile bar) and bottom (CTA bar).

Right Column (Facebook Desktop)

Audience Network

Carousel Ads

Carousel ads display up to 10 scrollable cards, each with its own image, headline, description, and link.

Card Specs

Carousel Strategies

• Product showcase — each card features a different product
• Feature walkthrough — each card explains one feature / benefit
• Story arc — cards tell a sequential story that builds to a CTA
• Social proof series — each card shows a different testimonial
• Before / after — alternate between problem and solution cards

Generating Carousel Images

Generate all cards in a single batch. Each card should have the brand logo, a short headline for that card's feature, and a relevant product screenshot.

openai_image_edit(
    requests=[
        {"prompt": "Card 1 for [company]. Logo at top. Headline: '[feature 1]'. Below: product screenshot showing [feature 1 UI]. Match brand style.", "reference_images": [logo_id, screenshot_id]},
        {"prompt": "Card 2 for [company]. Logo at top. Headline: '[feature 2]'. Below: product screenshot showing [feature 2 UI]. Same style as card 1.", "reference_images": [logo_id, screenshot_id]},
        {"prompt": "Card 3 for [company]. Logo at top. Headline: '[feature 3]'. Below: product screenshot showing [feature 3 UI]. Same style as card 1.", "reference_images": [logo_id, screenshot_id]},
    ],
    size="1024x1024",
    quality="high"
)

Creative Best Practices

Image Creative

• Show the product in use — context beats isolation
• One clear focal point — don't overcrowd the image
• Contrast with the feed — bright images stand out against grey / white feeds
• Brand colors in the image — not just in copy, in the visual itself
• Faces perform well — human faces increase engagement
• Avoid stock photo aesthetics — authenticity outperforms polish

Copy + Creative Alignment

The image and copy must tell the same story:

• If the hook mentions a number, the image should reinforce it
• If the body talks about ease, the image should feel simple and clean
• CTA in copy should match the CTA button selected

Platform Differences

Facebook Feed:

• Longer primary text can work (users scroll slower)
• Link description is shown more often
• Right column placements need simple, clear images

Instagram Feed:

• Visual quality matters more than copy length
• 4:5 aspect ratio takes maximum screen real estate
• Hashtags in primary text can extend reach

Instagram Stories:

• Full-screen vertical format
• First 3 seconds determine whether users swipe past
• Swipe-up / link CTA must be visually prompted
• Keep text minimal — the visual does the work

Generating Meta Ad Images

Every creative needs three layers: brand (logo), copy (headline + subline), and product (realistic screenshot of the actual app). Use brand.screenshot.description to understand what the product UI looks like.

Feed Creative (1:1)

openai_image_edit(
    requests=[{
        "prompt": (
            "Social media ad creative for [company name]. "
            "Top: the [company] logo. "
            "Headline: '[headline from ad copy or site hero text]'. "
            "Below: a clean, realistic product screenshot of [describe actual UI from brand.screenshot.description]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    size="1024x1024",
    quality="high"
)

Story Creative (9:16)

openai_image_edit(
    requests=[{
        "prompt": (
            "Vertical story ad creative for [company name]. "
            "Top: the [company] logo. "
            "Headline: '[headline]'. "
            "Below: a realistic product screenshot of [describe actual UI]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    size="1024x1536",
    quality="high"
)

Text-Heavy Creative (headlines in image)

When the creative needs large, readable text baked into the image, use Nano Banana for better text rendering:

nano_banana_image_generation(
    model="pro",
    requests=[{
        "prompt": (
            "Ad creative for [company name]. "
            "Logo at top. Large headline: '[headline]'. "
            "Below: realistic product screenshot of [describe actual UI]. "
            "Match the brand style from the references."
        ),
        "reference_images": [brand.logo.file_id, brand.screenshot.file_id]
    }],
    aspect_ratio="1:1",
    image_size="2K"
)