Automate Image Alt Text with n8n + Ollama (AI Accessibility Workflow)

Published March 24, 2026 · 12 min read · By Workflow Forge

Missing or poor alt text is the #1 accessibility violation on the web. WebAIM’s 2025 study found that 54% of all website images lack adequate alternative text. For businesses, that means legal risk under the ADA, EAA (European Accessibility Act), and WCAG 2.2 guidelines — and it means millions of screen reader users can’t interact with your content.

Fixing this manually is painful. A mid-size e-commerce store with 10,000 product images would take weeks of human effort to audit and describe. Cloud vision APIs (Google Vision, Azure, AWS Rekognition) charge $1–4 per 1,000 images and send all your product photos to third-party servers.

With n8n and Ollama, you can generate WCAG-compliant alt text at scale on your own hardware — for free, with full privacy. This tutorial builds a workflow that:

1. Scans your website, CMS, or image directory for images missing alt text
2. Sends each image to a local vision model (LLaVA) running on Ollama
3. Generates context-aware, descriptive alt text following WCAG 2.2 guidelines
4. Validates the output for length, quality, and relevance
5. Updates your CMS, database, or exports a CSV for bulk import

Why Local AI for Alt Text Generation?

Prerequisites

• n8n — self-hosted instance (Docker or npm)
• Ollama — running locally with a vision model: llava:13b (best quality) or llava:7b (faster)
• Image source — website URL, CMS API (WordPress, Shopify), or local directory
• ~8GB RAM for LLaVA 7B, ~16GB for LLaVA 13B

Pull the vision model:

ollama pull llava:13b
# Or for faster processing with less RAM:
ollama pull llava:7b

Architecture Overview

Image Source (Website crawl / CMS API / Directory scan)
    ↓
[Find Images] — Identify images with missing or empty alt text
    ↓
[Download Image] — Fetch image binary for analysis
    ↓
[Analyze Image] — LLaVA vision model generates description
    ↓
[Format Alt Text] — Apply WCAG guidelines (length, context, decorative check)
    ↓
[Validate] — Quality check: not too long, not too generic, relevant
    ↓
[Output] — Update CMS, write to database, or export CSV

WCAG 2.2 Alt Text Guidelines

Before building the workflow, understand what makes good alt text according to W3C guidelines:

• Be specific and concise — describe the image content in 125 characters or fewer (screen readers may truncate longer text)
• Convey function, not just appearance — “Submit order button” not “green rectangle”
• Skip “image of” or “photo of” — screen readers already announce it as an image
• Mark decorative images — use alt="" for purely decorative images (borders, spacers, backgrounds)
• Include relevant context — a product image alt text should include product name and key features
• Don’t repeat surrounding text — if the caption already says “Red Nike Air Max”, the alt text should add different detail

Step-by-Step Build

GET https://yoursite.com/wp-json/wp/v2/media?per_page=100&mime_type=image/jpeg

// Filter in n8n Function node:
const needsAlt = items.filter(item =>
  !item.json.alt_text || item.json.alt_text.trim() === ''
);
return needsAlt;

GET https://yourstore.myshopify.com/admin/api/2024-01/products.json?fields=id,title,images

// Extract images missing alt text:
const images = [];
for (const product of items) {
  for (const img of product.json.images) {
    if (!img.alt || img.alt.trim() === '') {
      images.push({
        json: {
          product_id: product.json.id,
          product_title: product.json.title,
          image_id: img.id,
          image_url: img.src,
          current_alt: img.alt || ''
        }
      });
    }
  }
}
return images;

// Use HTTP Request to fetch page HTML, then parse with Function node:
const cheerio = require('cheerio');  // n8n has cheerio built in
const $ = cheerio.load(items[0].json.data);
const missingAlt = [];

$('img').each((i, el) => {
  const alt = $(el).attr('alt');
  const src = $(el).attr('src');
  if (!alt || alt.trim() === '') {
    missingAlt.push({
      json: { src, page_url: items[0].json.url, index: i }
    });
  }
});
return missingAlt;

// HTTP Request node settings:
Method: GET
URL: {{ $json.image_url }}
Response Format: File
Output Field: image_data

// HTTP Request node to Ollama API:
POST http://localhost:11434/api/generate

{
  "model": "llava:13b",
  "prompt": "You are an accessibility expert writing alt text for screen readers. Describe this image following WCAG 2.2 guidelines:\n\n1. Be specific and concise (under 125 characters)\n2. Describe the content and function, not just appearance\n3. Do NOT start with 'image of', 'photo of', or 'picture of'\n4. If this is a decorative/background image with no informational content, respond with exactly: DECORATIVE\n5. Include relevant details: colors, text visible in image, actions depicted\n\nContext: This image is from a {{ $json.page_context || 'website' }}.\nProduct name (if applicable): {{ $json.product_title || 'N/A' }}\n\nWrite ONLY the alt text, nothing else.",
  "images": ["{{ $binary.image_data.base64 }}"],
  "stream": false,
  "options": {
    "temperature": 0.3,
    "num_predict": 80
  }
}

// Function node: choose prompt based on context
const context = $json.page_context || 'general';

const prompts = {
  'product': `Describe this product image for an online store listing.
Include: product type, color, material, key features visible.
Format: "[Product feature] [color/material] [product type]"
Example: "Stainless steel insulated water bottle with flip-top lid, 32oz, matte black finish"
Keep under 125 characters.`,

  'blog': `Describe this image for a blog article.
Focus on what the image communicates in the context of the article.
Be descriptive but concise. Under 125 characters.`,

  'icon': `This is a UI icon or button image.
Describe its function, not its appearance.
Example: "Search", "Close menu", "Add to cart"
If purely decorative, respond: DECORATIVE`,

  'chart': `Describe this chart or graph for someone who cannot see it.
Include: chart type, what is being measured, key trends or values.
Example: "Bar chart showing monthly revenue from Jan to Dec 2025, trending upward from $10K to $45K"`,

  'general': `Describe this image concisely for a screen reader user.
Be specific about content. Under 125 characters.
Do not start with "image of" or "photo of".`
};

return [{ json: { ...items[0].json, prompt: prompts[context] || prompts.general } }];

// Function node: validate and clean alt text
const altText = $json.generated_alt.trim();
const result = { ...items[0].json };

// Check if decorative
if (altText === 'DECORATIVE' || altText.toLowerCase().includes('decorative')) {
  result.final_alt = '';
  result.is_decorative = true;
  return [{ json: result }];
}

let cleaned = altText;

// Remove "image of", "photo of" prefixes
cleaned = cleaned.replace(/^(an? )?(image|photo|picture|photograph|screenshot|icon) (of|showing|depicting) /i, '');

// Capitalize first letter
cleaned = cleaned.charAt(0).toUpperCase() + cleaned.slice(1);

// Remove trailing period (screen readers add their own pause)
cleaned = cleaned.replace(/\.$/, '');

// Enforce max length (125 chars for screen reader compatibility)
if (cleaned.length > 125) {
  cleaned = cleaned.substring(0, 122) + '...';
}

// Quality flags
result.final_alt = cleaned;
result.is_decorative = false;
result.char_count = cleaned.length;
result.quality_warnings = [];

if (cleaned.length < 10) result.quality_warnings.push('too_short');
if (cleaned.length > 125) result.quality_warnings.push('too_long');
if (/^(a |the |an )/i.test(cleaned)) result.quality_warnings.push('starts_with_article');
if (/image|photo|picture/i.test(cleaned)) result.quality_warnings.push('contains_image_word');

return [{ json: result }];

// HTTP Request node:
POST https://yoursite.com/wp-json/wp/v2/media/{{ $json.image_id }}

{
  "alt_text": "{{ $json.final_alt }}"
}

// HTTP Request node:
PUT https://yourstore.myshopify.com/admin/api/2024-01/products/{{ $json.product_id }}/images/{{ $json.image_id }}.json

{
  "image": {
    "id": {{ $json.image_id }},
    "alt": "{{ $json.final_alt }}"
  }
}

// Spreadsheet File node:
// Columns: image_url, current_alt, generated_alt, is_decorative, char_count, quality_warnings
// Output: alt-text-audit-2026-03-24.csv

Batch Processing for Large Sites

Processing 10,000+ images requires careful batching to avoid overwhelming Ollama:

// Function node: batch controller
const BATCH_SIZE = 20;
const DELAY_MS = 1000; // 1 second between batches

const allImages = items;
const batches = [];

for (let i = 0; i < allImages.length; i += BATCH_SIZE) {
  batches.push(allImages.slice(i, i + BATCH_SIZE));
}

// Process batches with the Loop Over Items node
// Set batch size to 20, add a Wait node (1s) between batches
// LLaVA processes ~2-5 images/second on CPU, 10-20/second on GPU

Scheduled Audits: Keep Alt Text Current

New images are added constantly. Set up a Cron trigger to run the workflow weekly:

// Cron node: Run every Sunday at 2 AM
// Only process images added/modified since last run

// Function node: filter by date
const lastRun = $workflow.staticData.lastRun || '2000-01-01';
const newImages = items.filter(img =>
  new Date(img.json.date_created) > new Date(lastRun)
);

// Save timestamp for next run
$workflow.staticData.lastRun = new Date().toISOString();
return newImages;

Compliance Reporting

Generate an accessibility audit report after each run:

// Function node: generate report
const total = items.length;
const decorative = items.filter(i => i.json.is_decorative).length;
const generated = items.filter(i => !i.json.is_decorative && i.json.final_alt).length;
const warnings = items.filter(i => i.json.quality_warnings?.length > 0).length;

const report = {
  date: new Date().toISOString().split('T')[0],
  total_images_scanned: total,
  decorative_images: decorative,
  alt_text_generated: generated,
  needs_review: warnings,
  compliance_rate: ((generated + decorative) / total * 100).toFixed(1) + '%'
};

return [{ json: report }];

Send the report via email, Slack, or save to a dashboard. Track your compliance rate over time to demonstrate WCAG progress to stakeholders.

Real-World Use Cases

• E-commerce stores — Generate alt text for product catalogs. Shopify, WooCommerce, Magento, BigCommerce. Improves SEO (Google indexes alt text) and accessibility compliance.
• News and media sites — Describe editorial photos, infographics, and embedded charts for screen reader users.
• Documentation platforms — Generate descriptions for technical screenshots, architecture diagrams, and UI mockups.
• Education — Make course materials accessible by describing slides, diagrams, and educational images.
• Healthcare portals — Describe medical images locally without sending patient data to cloud APIs (HIPAA compliance).
• Government websites — Meet Section 508 and WCAG 2.2 AA requirements for all published images.

SEO Benefits of Alt Text

Alt text isn’t just for accessibility — it directly impacts search rankings:

• Google Image Search — alt text is the primary signal for image search ranking. Proper alt text gets your images into Google Image results, driving additional traffic.
• Page relevance — alt text helps search engines understand page content, improving overall page ranking for target keywords.
• Featured snippets — pages with proper image descriptions are more likely to appear in featured snippets and rich results.

Troubleshooting

LLaVA not generating good descriptions

• Use llava:13b instead of llava:7b for better quality
• Ensure temperature is 0.3 or lower for factual descriptions
• Add domain context to the prompt (“this is a product image for an electronics store”)

Ollama connection issues from n8n

• If n8n is in Docker, use http://host.docker.internal:11434 instead of localhost
• Set OLLAMA_HOST=0.0.0.0 in Ollama’s environment to allow network access
• Check firewall rules: sudo ufw allow 11434/tcp

Processing too slow

• Resize images before sending to Ollama (768px max dimension is sufficient for alt text)
• Use llava:7b for 2–3x faster processing at slightly lower quality
• Process in parallel batches if you have a GPU with enough VRAM

Related Tutorials

• Generate AI Product Descriptions with n8n + Ollama
• Automate Content Moderation with n8n + Ollama
• Automate Data Enrichment with n8n + Ollama
• Complete Guide: Self-Hosted AI with n8n + Ollama