Image Brand Compliance

Review image files (PNG/JPEG) against brand guidelines. Score across 8 visual compliance dimensions, identify violations with exact locations, and apply fixes when the user accepts them.

For HTML emails, SMS copy, or text-based content, use the brand-compliance skill instead. This skill focuses on pixel-level visual analysis of image files.

When to Use This Skill

Use this skill when:

• "Review this Instagram ad image against our brand guidelines"
• "Check if this hero image follows brand standards"
• "Score this AI-generated ad for brand compliance"
• "Fix the brand violations on this PNG"
• "Is this social media image on-brand?"
• Validating image-gen or image-editor output before publishing

Use brand-compliance instead when:

• Reviewing HTML email templates, SMS copy, or text content
• Checking legal/CAN-SPAM/TCPA compliance on email or SMS
• Scoring content that is primarily text-based

Inputs Required

1. Image — file path to a PNG or JPEG (e.g., /tmp/instagram-ad.png)

2. Brand Guidelines — one of:

   • File path to markdown or PDF guidelines (e.g., ./brand-guidelines.md)
   • Inline text pasted into the prompt
   • Auto-detect: look for brand-guidelines.md in the current working directory

If either input is missing, ask for it before proceeding. If no brand guidelines are available, prompt the user to create one using brand-compliance/references/brand-guidelines.md as a template.

Workflow

Step 1 — Parse Brand Guidelines

Read the provided brand guidelines and extract image-relevant rules:

• Color palette: Primary, secondary, CTA, background hex codes and usage context rules
• Logo: Variants (dark/light), placement zones, minimum size, clear space, allowed backgrounds
• Typography: Font families, weights, sizes for text overlays
• Imagery style: Photography mood, subject framing, lighting, prohibited aesthetics
• Messaging lexicon: Approved/prohibited terms, taglines, brand signature
• Legal marks: Required disclosures (FTC #ad, copyright, trademarks)
• Accessibility: Contrast requirements, minimum font sizes
• Channel specs: Target dimensions, aspect ratios, file size limits

Note which sections are unconfigured — only score against configured sections.

Step 2 — Analyze the Image

Read the image file visually. Assess each compliance dimension against the extracted guidelines:

• Colors: Sample dominant and accent colors, compare to brand palette hex codes
• Logo: Check presence, variant selection, placement, clear space, size, distortion
• Typography: Identify font families and sizes on text overlays, compare to brand spec
• Imagery mood: Evaluate subject matter, lighting, composition, emotional tone
• Copy: Read all text overlays, check against approved/prohibited terminology
• Legal: Check for required marks (FTC disclosure, copyright, trademark symbols)
• Accessibility: Estimate text-over-background contrast ratios, assess readability
• Channel fit: Check image dimensions and aspect ratio against target channel

Identify specific violations with exact location (e.g., "CTA button area uses #FF3333 instead of brand CTA #CC262C").

Step 3 — Score and Report

Score across 8 dimensions (0–5 each, 40 points max). Only score dimensions where guidelines are configured.

Compliance tiers:

• 38–40: Fully Compliant — Ready to publish
• 32–37: Mostly Compliant — Minor fixes needed
• 24–31: Partially Compliant — Significant issues
• 0–23: Non-Compliant — Major revision required

For detailed scoring criteria per dimension, see references/image-scoring-rubric.md.

Step 4 — Render HTML Compliance Dashboard

Generate an HTML compliance dashboard and save it as brand-compliance-report-{asset-name}.html in the working directory.

Image embedding: When including the reviewed image or any reference images in the HTML dashboard, always embed them as base64 data URIs (data:image/png;base64,...) directly in the src attribute. Never use local file paths — the viewer cannot resolve them. Use Python to encode:

import base64
with open("image.png", "rb") as f:
    src = "data:image/png;base64," + base64.b64encode(f.read()).decode()

Dashboard structure:

1. Header — Brand name, asset filename, overall score (large, color-coded by tier), tier label
2. Dimension breakdown — 8 rows, each with score badge (0–5) and labeled progress bar
3. Violations list — collapsible <details>, grouped by dimension, each with: what was found, where in the image, exact fix recommendation
4. Recommendations — collapsible <details>, prioritized Critical > High > Medium > Low, each with projected score impact
5. Action prompt — "Type 'apply fixes' to implement all critical and high-priority changes"

Open with mcp__work__open_file if available. If the MCP tool is unavailable, print the file path for the user to open manually.

Also summarize violations in plain text in the chat so the user can act without switching to the dashboard.

Step 5 — Apply Fixes (on user acceptance)

When the user says "apply fixes", "yes", "make the changes", or accepts specific fixes:

Prerequisites check: Before running any Pillow scripts, verify the dependency:

python3 -c "import PIL" 2>/dev/null || echo "MISSING"

If missing, instruct the user to run pip install Pillow before proceeding.

Fix methods by violation type:

For Python/Pillow edits: write a self-contained script to a temp file, run it with Bash, save the corrected image alongside the original (e.g., {name}-fixed.png).

After all fixes are applied, re-run the full scoring analysis on the updated image and produce a new compliance dashboard showing:

• New overall score and tier
• Per-dimension delta (e.g., "Color Palette: 2 -> 5")
• Remaining violations (if any)
• Confirmation of what was changed

Output Format

Plain-text summary (always show in chat)

Brand Compliance Score: [X]/40 — [Tier]

Violations ([N] found):
1. [Dimension] — [what] at [location]
   Fix: [exact change]
2. ...

Type 'apply fixes' to implement all changes.

After fixes (re-score)

Updated Score: [X]/40 — [Tier]  (was [Y]/40)

Changes applied:
- [Fix 1]
- [Fix 2]

Remaining: [N violations] — [list or "None — fully compliant"]

Edge Cases

• Unconfigured guideline sections: Note which dimensions were skipped and why. Do not penalize for rules that are not defined.
• Multiple images: If the user provides 2+ images, score each independently then note cross-image consistency issues.
• Partial acceptance: If user says "apply only the color fixes" or "skip the logo", apply only the accepted changes and re-score accordingly.
• No brand guidelines available: Prompt the user to provide them. Point to brand-compliance/references/brand-guidelines.md as a template to follow.
• Pillow not installed: Detect before attempting fixes. Provide install instructions. Do not attempt Python image edits without confirming the dependency.
• Image too large or corrupted: Report the issue and ask the user to provide a valid image file.

Best Practices

• Provide brand guidelines with exact hex codes, not just color names — "Burnt Ochre #B35D33" enables precise matching
• Include logo files (dark and light variants) in the guidelines so fixes can composite them automatically
• Specify the target channel (Instagram feed, story, display ad) so channel specifications can be scored
• Run image-brand-compliance after using the image-gen or image-editor skills to catch issues before publishing
• For HTML emails or text content, use brand-compliance instead — it has channel-specific scoring for email/SMS/Instagram copy

Related Skills

• brand-compliance — Review HTML, email, SMS, and text content against brand guidelines (8-dimension, 0–40 scoring). Use for non-image content.
• image-gen — Generate ad images with AI + text/logo compositing. Run image-brand-compliance on the output to validate.
• image-editor — Interactive drag-and-drop editor for image overlays. Use to manually adjust before re-scoring.
• multi-channel-ad-ideation — Orchestrate full campaign ideation across channels. Brand guidelines are applied during ideation; use this skill for post-production validation.

Example

For a complete worked example showing an image review, scoring, violation identification, fix application, and re-scoring, see examples/image-review-example.md.