AWS Rekognition is a managed computer vision service that provides image and video analysis powered by deep learning. With pricing based on the number of images processed, minutes of video analyzed, and faces stored, Rekognition offers pay-as-you-go access to capabilities that would cost millions to build in-house.
TL;DR: Image analysis starts at $0.001 per image (first 1M/month) with volume discounts. Video analysis costs $0.10 per minute. Custom Labels adds $4.00 per inference hour plus $1.00 per training hour. A free tier provides 5,000 images per month for 12 months.
Image Analysis Pricing
| Monthly Volume | Price per Image |
|---|---|
| First 1 million images | $0.001 |
| Next 9 million images | $0.0008 |
| Next 90 million images | $0.0006 |
| Over 100 million images | $0.0004 |
These per-image prices apply to the following Image API operations:
- Label Detection - Identify objects, scenes, activities, and concepts
- Content Moderation - Detect unsafe or inappropriate content
- Text Detection - Extract text from images (signs, documents, license plates)
- Celebrity Recognition - Identify well-known individuals
- Image Properties - Analyze color, quality, and sharpness
Each API call for one image counts as one image processed. Calling multiple APIs on the same image counts as multiple processed images.
Face Detection and Search Pricing
| Operation | Price |
|---|---|
| Face detection and comparison | Same as Image Analysis tiers ($0.001 first 1M) |
| Face indexing (adding to collection) | Same as Image Analysis tiers |
| Face search (searching a collection) | Same as Image Analysis tiers |
| Face metadata storage | $0.01 per 1,000 face metadata stored per month |
Face collections let you index faces and then search against them. Indexing a face and searching for a match are each billed as one image processed. Storing face metadata costs $0.01 per 1,000 faces per month, which is $0.00001 per face.
For a collection of 1 million stored faces, the monthly storage cost is $10.00. Searching 100,000 images against this collection would cost an additional $100 at the first-tier rate.
Video Analysis Pricing
| Operation | Price per Minute |
|---|---|
| Label detection | $0.10 |
| Content moderation | $0.10 |
| Face detection | $0.10 |
| Face search | $0.10 |
| Person tracking | $0.10 |
| Celebrity recognition | $0.10 |
| Text detection | $0.10 |
Video analysis is billed per minute of video processed, with a minimum charge of 1 minute per video. Running multiple analysis types on the same video incurs separate charges for each type. A 10-minute video analyzed for both labels and content moderation would cost $2.00 (10 min x $0.10 x 2 operations).
Streaming video analysis for real-time face detection uses Kinesis Video Streams and is priced at the same $0.10 per minute rate.
Custom Labels Pricing
| Component | Price |
|---|---|
| Training | $1.00 per hour |
| Inference | $4.00 per hour |
Custom Labels lets you train models to detect custom objects and scenes specific to your business. Training time depends on dataset size and complexity but typically ranges from 1 to 24 hours. Each running inference endpoint costs $4.00 per hour.
Custom Labels Cost Example
| Phase | Duration | Cost |
|---|---|---|
| Training (initial) | 4 hours | $4.00 |
| Inference endpoint (24/7) | 730 hours/month | $2,920/month |
| Inference endpoint (8 hrs/day) | 240 hours/month | $960/month |
Running a Custom Labels inference endpoint 24/7 costs approximately $2,920 per month. For workloads that do not require constant availability, scheduling endpoints to run only during business hours reduces costs to about $960 per month.
Free Tier
| Feature | Free Allowance | Duration |
|---|---|---|
| Image analysis | 5,000 images/month | 12 months |
| Video analysis | 1,000 minutes/month | 12 months |
| Face metadata storage | 1,000 faces/month | 12 months |
| Custom Labels training | 10 hours | 12 months (one-time) |
The Rekognition free tier gives you 12 months to evaluate the service. The 5,000 free images per month are sufficient for building and testing prototype applications before scaling to production.
Real-World Cost Examples
| Use Case | Monthly Volume | Monthly Cost |
|---|---|---|
| Content moderation startup | 500,000 images | $500 |
| E-commerce product tagging | 2M images (labels) | $1,800 |
| Security camera face search | 50K images + 10K faces stored | $50.10 |
| Media video analysis | 1,000 min video (2 APIs) | $200 |
| Manufacturing defect detection | Custom Labels, 8 hrs/day | $964 |
Cost Optimization Tips
1. Batch Your API Calls Efficiently
Each API call on an image is billed separately. If you need both label detection and content moderation, both will be charged. Design your pipeline to call only the APIs you need for each image rather than running all available operations.
2. Pre-Filter Images Before Processing
Use simple heuristics (file size, dimensions, or basic image hash deduplication) to filter out irrelevant images before sending them to Rekognition. Removing duplicates and low-quality images can reduce processed volume by 20-40%.
3. Manage Custom Labels Inference Endpoints
Never leave Custom Labels endpoints running when they are not needed. Use AWS Lambda scheduled events to start endpoints before business hours and stop them after, saving up to 67% compared to 24/7 operation.
4. Use Volume Pricing Tiers
At high volumes, Rekognition pricing drops significantly. Processing 10 million images costs $0.0008 per image (tiers 2+), a 20% reduction from the base rate. Consolidate image processing from multiple applications to reach higher tiers faster.
5. Choose the Right API for Your Use Case
Text detection with Rekognition costs $0.001 per image. For document-heavy text extraction, AWS Textract may provide better accuracy at a different price point. Evaluate whether Rekognition or a specialized service fits your needs and budget.
Related Guides
FAQ
Is Rekognition charged per API call or per image?
Rekognition is charged per image per API call. If you call both DetectLabels and DetectModerationLabels on the same image, that counts as two image operations. Each operation is billed at the per-image rate for your monthly volume tier.
How is video analysis billed for short clips?
Video analysis has a minimum charge of 1 minute per video. A 15-second clip is billed as 1 minute ($0.10). For very short clips, consider extracting frames and using image APIs instead, which would cost $0.001 per frame.
Can I reduce Custom Labels costs for intermittent workloads?
Yes. Start and stop Custom Labels endpoints programmatically based on demand. An endpoint takes 2-5 minutes to start, so schedule it slightly before your processing window begins. This approach works well for batch workflows that run during specific time windows.
Lower Your Rekognition Costs with Wring
Wring helps you access AWS credits and volume discounts to lower your Rekognition costs. Through group buying power, Wring negotiates better rates so you pay less per image processed.
