LLM SEO Generator: A WordPress Plugin for Automatically Generating Image SEO Metadata Using AI Visual Capabilities

LLM SEO Generator is an open-source WordPress plugin that uses the visual capabilities of multimodal large language models to automatically generate SEO-friendly titles, alt texts, descriptions, and captions for images. It supports local Ollama deployment and cloud APIs (Claude, OpenAI), balancing privacy and performance. It addresses the issues of time-consuming manual image metadata processing, missed SEO opportunities, and violations of WCAG accessibility standards.

In content operation and website management, image SEO optimization and accessibility compliance are often overlooked. Manually writing metadata for each image is time-consuming and labor-intensive; for large volumes of images (e.g., real estate, e-commerce, photo libraries), manual processing is almost impossible. This leads to default filenames, blank alt attributes, which affect search engine rankings, are unfriendly to visually impaired users, and violate WCAG standards.

Vision-Driven Analysis

Uses visual models to deeply identify image subjects, extract features, test uniqueness, control length, filter clichés, and generate specific descriptions (e.g., "White coastal house with blue metal roof and palm tree landscape").

Metadata Generation Modes

Supports titles (5-6 words), alt texts (100-125 characters), captions (1-2 sentences), and detailed descriptions (2-3 sentences). It offers two modes: "Fill Missing Only" and "Regenerate All".

Batch Processing

Intelligently filters images with missing metadata, processes them in batches (10 images per batch), tracks progress in real time, and automatically converts WebP/PNG to JPEG for Ollama compatibility.

Multi-Provider Architecture

• Ollama: Local deployment, zero cost, strong privacy. Recommended models like llava, 10 images/minute;
• Claude: Highest visual quality, recommended claude-3-5-sonnet;
• OpenAI GPT-4 Vision: Balances quality and cost, recommended gpt-4o.

Prompt Engineering

Follows the "Subject + Key Visual Features" formula, with a built-in library of visual elements (e.g., architectural styles, colors, materials) to assist in descriptions.

Image Format Processing

Automatically converts PNG/WebP to JPEG with 90% quality, and automatically cleans up temporary files.

Extensibility Design

Provides an abstract base class LLM_SEO_Abstract_Provider, allowing developers to easily add new LLM providers.

• Real Estate: Batch generate property image descriptions, add SEO keywords, ensure accessibility compliance;
• E-commerce: Generate product image descriptions to improve image search visibility;
• Photography Portfolios: Quickly generate professional captions;
• Content Management Systems: Supplement metadata for historical images to enhance website search performance.

Privacy

• Local Ollama: Data never leaves the server, suitable for sensitive images;
• Cloud APIs: Images are sent to service providers, so privacy policies should be reviewed.

Security

Data cleaning, WordPress nonce verification, permission checks, and API keys are not hard-coded.

Performance

3-5 seconds per image, 10 images/minute in batch, memory usage ~2GB, recommended PHP memory ≥256MB.

1. Prioritize local Ollama to save costs and protect privacy;
2. Set website-related keywords to enhance metadata relevance;
3. Test single images first to confirm quality before batch processing;
4. Use the "Fill Missing Only" mode for the first time to avoid overwriting existing metadata;
5. Process during off-peak hours and monitor server resources.

LLM SEO Generator transforms multimodal AI visual capabilities into a practical tool, enhancing SEO effectiveness and accessibility experiences, providing an efficient and flexible solution for image-intensive websites. As multimodal AI develops, such automated tools will help humans free themselves from repetitive work and focus on creative tasks.