How to Use AI to Generate Stock Photo Keywords (Step-by-Step Tutorial)
Djaka Pradana - CEO Meita.ai · 
You’ve got 500 photos from a weekend shoot. They’re edited, color-corrected, and ready for upload. There’s just one problem: each one needs a title, description, and 30-50 keywords before any stock platform will accept it.
At 5 minutes per image, that’s 42 hours of keywording. Almost a full work week — just typing metadata.
AI keywording tools solve this. Here’s exactly how to use them, step by step.
The Problem with Manual Keywording
Before we dive into the tutorial, let’s be clear about why manual keywording doesn’t scale:
- 5 minutes per image is the average for experienced keyworkers
- Quality drops after the first 50-100 images (fatigue, repetition)
- Inconsistent keywords across similar images hurt search ranking
- Opportunity cost — every hour keywording is an hour not shooting
AI doesn’t get tired, doesn’t get inconsistent, and processes images in seconds.
Step 1: Choose Your AI Model
Most AI keywording tools use one of these vision models:
| Model | Speed | Quality | Cost |
|---|---|---|---|
| Google Gemini 2.0 Flash | Very fast | Excellent | Free tier available |
| Google Gemini 2.5 Pro | Fast | Best | ~$0.01/image |
| OpenAI GPT-4 Vision | Moderate | Excellent | ~$0.02/image |
| OpenRouter (various) | Varies | Varies | Varies |
Our recommendation: Start with Gemini 2.0 Flash. It’s free (with an API key from Google AI Studio), fast, and produces surprisingly good results for stock photography. Upgrade to Gemini Pro or GPT-4 Vision only if you need higher accuracy for complex or niche content.
Getting a Free Gemini API Key
- Go to Google AI Studio
- Sign in with your Google account
- Click “Get API Key” → “Create API Key”
- Copy the key — you’ll paste it into your keywording tool
That’s it. You now have free access to one of the most capable vision AI models available.
Step 2: Import Your Photos
Open your AI keywording tool and import your images. In Meita.ai, you can:
- Drag and drop a folder of hundreds of images
- Click “Add Item” to select files individually
- Supported formats: JPG, PNG, BMP, SVG, EPS, MP4, AVI, MOV, and more
The tool will display thumbnails of all imported images in a grid, ready for processing.
Pro tip: Organize your photos into themed folders before importing (e.g., “Beach Lifestyle”, “Office People”, “Food Flat Lays”). This helps you review results more efficiently and apply custom prompts per batch.
Step 3: Configure Your Settings
Before generating, set up a few things:
Select Your AI Model
Choose the model you set up in Step 1. If you have multiple API keys, some tools (like Meita.ai) can load-balance across them for faster processing.
Set Your Target Platform
Different platforms have different requirements:
- Adobe Stock: 50 keywords max, 70-character title
- Shutterstock: 50 keywords max, 200-character description
- Freepik: 50 keywords max
- General/EXIF: Embedded directly into image files
Customize Your Prompt (Optional)
Most AI tools let you customize the generation prompt. This is powerful for niche content:
- Food photography: “Focus on ingredients, cuisine type, cooking method, and food styling terms”
- Travel photography: “Include destination name, landmarks, cultural elements, and travel themes”
- Business/corporate: “Emphasize professional settings, teamwork, leadership, and modern workplace concepts”
Step 4: Generate Metadata
Hit the generate button. Here’s what happens behind the scenes:
- Your image is compressed and sent to the AI model
- The AI analyzes visual content — objects, people, scenes, colors, mood
- It generates a title, description, and keywords optimized for stock platforms
- Results are returned and displayed next to each image
Speed benchmarks (Meita.ai with Gemini Flash, 8 parallel workers):
| Batch Size | Time | Per Image |
|---|---|---|
| 10 images | ~8 seconds | 0.8s |
| 100 images | ~60 seconds | 0.6s |
| 500 images | ~4 minutes | 0.5s |
| 1,000 images | ~8 minutes | 0.5s |
Compare that to manual keywording at 5 minutes per image: 1,000 images would take 83 hours manually vs. 8 minutes with AI.
Step 5: Review and Edit
AI-generated metadata is good, but not perfect. Always review before uploading:
What to Check
Titles: Are they accurate and natural-sounding? Fix any that are awkward or too generic.
Keywords: Look for:
- Missing obvious keywords the AI missed
- Incorrect keywords (AI sometimes misidentifies objects)
- Brand names or celebrity names that slipped through
- Duplicate or near-duplicate keywords
Categories: Make sure the assigned category matches the content.
How Much Editing Is Needed?
In our experience, 80-90% of AI results need zero or minimal editing. The remaining 10-20% might need:
- 1-2 keywords added or removed
- A title tweaked for clarity
- A wrong category corrected
This “review and tweak” process takes about 15-30 seconds per image — compared to 5 minutes for full manual keywording.
Step 6: Export
Once you’re satisfied with the metadata, export it for your target platform:
CSV Export
Generate a platform-formatted CSV file:
- Adobe Stock CSV — filename, title, keywords, category
- Shutterstock CSV — filename, description, keywords, categories, flags
- Freepik CSV — filename, title, keywords
- And more (Canva, Dreamstime, Vecteezy, Envato, Pond5)
EXIF/IPTC Embedding
Alternatively, write the metadata directly into your image files. This means the metadata travels with the file — upload to any platform and the title, description, and keywords are already there.
When to use CSV vs. EXIF:
- CSV — best for bulk uploads to a specific platform
- EXIF — best when uploading to multiple platforms or when you want metadata permanently attached to files
Are AI Keywords Good Enough?
This is the most common question from contributors trying AI keywording for the first time. The honest answer: yes, with a quick review.
AI-generated keywords are:
- More consistent than manual keywords (no fatigue effect)
- More comprehensive — AI catches details humans overlook
- Properly ordered — most relevant keywords first
- Platform-compliant — no brand names, proper character limits
The main weakness is niche or culturally specific content where the AI might miss context. A photo of a specific regional dish might get “food, plate, meal” but miss “pad thai, Thai cuisine, street food.” Custom prompts help here.
Bottom line: AI keywording + 30 seconds of human review per image produces results as good or better than 5 minutes of manual keywording.
Cost Breakdown
| Method | Cost per Image | Time per Image | 1,000 Images Cost | 1,000 Images Time |
|---|---|---|---|---|
| Manual | $0 (your time) | 5 min | $0 | 83 hours |
| ChatGPT ($20/mo) | ~$0.02 | 2 min | $20 | 33 hours |
| AI Tool (own API key) | ~$0.005 | 0.5 sec + 30 sec review | $5 | 8.5 hours |
| AI Tool (free Gemini) | $0 | 0.5 sec + 30 sec review | $0 | 8.5 hours |
With a free Gemini API key and Meita.ai, the cost of keywording 1,000 images is literally $0 and under 9 hours (mostly review time).
Get Started
- Download Meita.ai (free, Windows/macOS/Linux)
- Get a free Gemini API key
- Import your photos and generate metadata
- Export your first CSV and upload to your stock platform
You’ll wonder why you ever keyworded manually.