Amazon Rekognition pricing

With Amazon Rekognition there are 4 different types of usage, each with their own pricing details.  Learn more about the following:

• Amazon Rekognition Image pricing 
• Amazon Rekognition Video pricing
• Amazon Rekognition Custom Labels pricing
• Amazon Rekognition Face Liveness pricing
  
• Amazon Rekognition Custom Moderation pricing

Amazon Rekognition Image makes it easy to add image analysis to your applications using proven, highly scalable, deep learning technology that requires no machine learning expertise to use. With Amazon Rekognition Image, you only pay for what you use. There is no up-front commitment or minimum fee. There are two types of costs with Amazon Rekognition Image: the cost for image analysis and the cost for face metadata storage.

Image analysis: Amazon Rekognition Image charges you each time you analyze an image using our APIs. Running multiple APIs against a single image counts as processing multiple images. Usage is billed based on a tiered pricing model tied to volume of images processed per month. The majority of Amazon Rekognition Image APIs are categorized into two groups, Group 1 and Group 2, which have different pricing.

Group1: AssociateFaces, CompareFaces, DisassociateFaces, IndexFaces, SearchFacesbyImage, SearchFaces, SearchUsersByImage, SearchUsers.
Group 2: DetectFaces, DetectModerationLabels, DetectLabels, DetectText, RecognizeCelebrities, DetectProtectiveEquipment APIs.

Other than the APIs listed in Group 1 and Group 2 above, Amazon Rekognition Image also supports Image Properties, which is priced separately from Group 1 and Group 2.

Face metadata storage: To enable face and user search, you will need to store a repository of face metadata objects (face vectors and user vectors) against which Amazon Rekognition can search for matches. Storage charges are applied monthly, and are pro-rated for partial months.

As part of the AWS Free Tier, you can get started with Amazon Rekognition Image for free. The free tier period lasts 12 months from the date of account creation.

Image analysis: During the free tier period you can analyze 1,000 images per month for free each, in Group 1 and Group 2 APIs. Free tier is not offered for Image Properties.

Face metadata storage: During the free tier period, you can store 1,000 face vector objects and 1,000 user vector objects per month for free.

Amazon Rekognition Video supports both stored video analysis and real time streaming video events. With Amazon Rekognition Video you pay only for what you use. There are no resources to provision, no upfront costs, or minimum fees.

Amazon Rekognition streaming video events processes video from new or existing Kinesis Video Streams. Rekognition starts processing the Kinesis Video stream only when you send us the notification to start video analysis and can analyze up to 120 seconds of video per event. You pay only for the amount of video processed by Amazon Rekognition. Note: You will pay separately for the Amazon Kinesis Video Streams service.

Amazon Rekognition Stored Video Analysis charges you for videos analyzed from Amazon S3. When running multiple API calls against the same section of the video, you will get charged separately for each API.

Face Metadata Storage: To enable face search, you will need to store a repository of face metadata against which Amazon Rekognition can search for matches. Storage charges are applied monthly, and are pro-rated for partial months.

As part of the AWS Free Tier, you can get started with Amazon Rekognition Video at no cost. The Free Tier lasts 12 months from the date of account creation and includes 60 free minutes of video analysis per month. Amazon Rekognition Video free tier covers Label Detection, Content Moderation, Face Detection, Face Search, Celebrity Recognition, Text Detection and Person Pathing.

With Amazon Rekognition Custom Labels, you can identify the objects and scenes in images that are specific to your business needs. When using Rekognition Custom Labels, there are two types of costs.

Training hours
There is a cost for each hour of training required to build a custom model with Amazon Rekognition Custom Labels. Amazon Rekognition Custom Labels may run multiple compute resources in parallel to train your model more quickly. This means that the number of hours billed may be more than the number of actual elapsed hours it takes to train the model. The number of training hours needed to train your model depends on many factors such as number of images and number of labels in the training set, and types of machine learning algorithms used to train your model. We have typically seen that 90% of models need less than 24 training hours to train. Models that take more than 72 training hours will be automatically terminated. You will not be charged if your training is automatically terminated.

For example, let’s assume your training started at 4:00pm and ended at 8:30pm, and Amazon Rekognition Custom Labels used two resources in parallel to train your model more quickly. Your total training hours billed would be 9 hours (4.5 hours elapsed time x 2 resources).

Inference hours
There is a cost for each hour your trained custom model is available to process images. The number of images you can process in a single hour depends on many factors, such as the size of the images processed and the complexity of the custom model. With Amazon Rekognition Custom Labels, you can run multiple compute resources in parallel to process your images more quickly. This means that the number of hours billed may be more than the actual elapsed hours it takes to run your trained custom model.

For example, let’s assume you started your inference with your custom model at 2:00pm and ended at 5:00pm, and you chose to provision two resources in parallel to process your images. Your total inference hours billed would be 6 hours (3 hours elapsed time x 2 resources).

If you expect to process images in batches (e.g. once a day or week, or at scheduled times during the day), you should provision your custom model at scheduled times, process all your images, and then deprovision your resources. If you do not deprovision your resources, you will continue to be charged even if no images are processed.

As part of the AWS Free Tier, you can get started with Amazon Rekognition Custom Labels for free. The Free Tier lasts 12 months from the date of account creation and includes 2 free training hours per month and 1 free inference hour per month.

Amazon Rekognition Face Liveness verifies that only real users, not bad actors using spoofs, can access your services. Face Liveness analyzes a short selfie video to detect spoofs presented to the camera, such as printed photos, digital photos, digital videos, or 3D masks, as well as spoofs that bypass the camera, such as pre-recorded or deepfake videos. Face Liveness is a fully managed feature that can be easily added to your React web, native iOS, and native Android applications running on most devices with a front-facing camera. No infrastructure management, hardware-specific implementation, or machine learning expertise is required. The feature automatically scales up or down in response to demand, and you only pay for the face liveness checks you perform.

With Amazon Rekognition Custom Moderation, you can enhance the accuracy of the moderation deep learning model. You can train a custom adapter with as few as 20 annotated images and achieve enhanced accuracy for image analysis tasks related to your specific use case. No infrastructure management, hardware-specific implementation, or machine learning expertise is required.

To create and use an adapter you must:

1. Train your adapter with annotated images
2. Use your adapter by providing the unique AdapterID to the DetectModerationLabels API for inference

Training cost
There is a cost for each minute of training required to build an adapter with Amazon Rekognition Custom Moderation. Amazon Rekognition Custom Moderation may run multiple compute resources in parallel to train your adapter more quickly. This means that the number of minutes billed may be more than the number of actual elapsed minutes it takes to train the adapter. The number of training minutes needed to train your adapter depends on many factors such as number of images and number of labels in the training set. We have typically seen that 90% of adapters need less than 60 training minutes to train. You will not be charged if your training is automatically terminated.

For example, let’s assume your training started at 4:00pm and ended at 4:35pm, and Amazon Rekognition Custom Moderation used a single resource to train your adapter. Your total training minutes billed would be 35 minutes (35 minutes elapsed time x 1 resource).

As part of the AWS Free Tier, you can get started with Amazon Rekognition Custom Moderation for free. The Free Tier lasts 12 months from the date of account creation and includes 2 free training hours per month to train a custom moderation adapter or a custom labels model.

Inference cost
You will be charged each time you analyze an image using the custom moderation adapter. There is no up-front commitment or minimum fee. Usage is billed based on a tiered pricing model tied to volume of images processed per month.