With Amazon Lookout for Vision, you pay only for what you use. There is no upfront commitment or minimum fee. After you create a project and upload baseline images, Amazon Lookout for Vision lets you detect anomalies by training a customized defect detection machine learning (ML) model on your manufacturing lines.
You can choose to detect anomalies either by using Amazon Lookout for Vision’s cloud APIs or on the edge by deploying the trained model to a hardware device of your choice. Amazon Lookout for Vision models can be deployed on any NVIDIA Jetson edge appliance or x86 compute platform running Linux with an NVIDIA GPU accelerator. We use Amazon SageMaker Neo to compile our models to be edge compatible. See the list of supported devices here. You can use AWS IoT Greengrass to deploy, and manage your edge compatible customized models on your fleet of devices. You will pay separately for the AWS IoT Greengrass service.
There are three components that determine your bill: charges for training the model (training hours), charges for detecting anomalies on the cloud (cloud inference hours), and/or charges for detecting anomalies on the edge (edge inference units).
Training Hours
You pay for the number of hours it takes to train your defect detection model. Amazon Lookout for Vision can provision multiple compute resources in parallel to quickly train your model. For example, Lookout for Vision might provision three compute resources to run simultaneously. Each resource, using three different algorithms, runs for 30 minutes. You then pick the best algorithm for your model. This means that you will have a trained model ready for use within 30 minutes, but the number of hours billed will be 1.5 hours (30 minutes x 3 resources).
Cloud Inference Hours
You pay for the number of hours you detect anomalies or defects with your customized model. To find defects using your model, Lookout for Vision provisions a minimum of one compute resource, known as an inference unit. A single cloud inference unit represents 1 hour of processing and can support up to 5 Transaction Per Second (TPS). You can increase or decrease your defect detection capacity depending on the demands of your manufacturing line by requesting additional inference units. You are billed per inference unit * inference hours that you use your defect detection model. For example, if you are using 2 inference units and you use the model for 8 hours (start at 9 AM and stop at 5 PM), you would be billed 8 * 2 = 16 inference hours. If you do not explicitly stop using your model, you will continue to be charged even if you are not actively detecting anomalies using the model.
Edge Inference Units
You will be charged a monthly fee per Lookout for Vision edge inference unit used. One edge inference unit is the compute used by the Lookout for Vision model deployed on your edge appliance to process up to 120 inspections per minute from connected camera streams. You are billed monthly based on the number of edge inference units per edge appliance * number of edge appliances that are using the defect detection model. Lookout for Vision charges a minimum of one edge inference unit per month from one connected camera stream. For example, a Lookout for Vision model deployed on an edge appliance to detect defective silicon wafers from a production line that manufactures 100 silicon wafers per minute consumes one edge inference unit per month. The monthly charge in this example is 1 edge inference unit * 1 edge appliance = 1 edge inference unit.
We’ve seen that typically one active camera stream can support a maximum of 120 inspections per minute and would use one edge inference unit in one month. For example, if you have two active camera streams connected to an edge appliance running a Lookout for Vision anomaly detection model, you would be billed for 2 edge inference units for that month. Now, if you have 5 edge appliances, each using 2 edge inference units, you would be billed for 5*2 = 10 edge inference units for that month. Regardless of whether you’re actively using Lookout for Vision on your edge appliance for 4 hours a day or 24 hours a day, as long as the query rate in any given month is below 120 inspections/minute on that edge appliance you will be charged for one edge inference unit.
AWS Free Tier
As part of the AWS Free Tier, you can get started with Amazon Lookout for Vision for free. The Free Tier lasts 3 months and includes 10 free training hours per month, up to 4 free cloud inference hours per month, and up to 5 free edge inference units per month.
Pricing tables
Amazon Lookout for Vision Training and Inference usage is billed on minute increments, with a minimum of one minute.
* Amazon Lookout for Vision Training and Cloud Inference usage is billed on minute increments, with a minimum of 1 minute
**Amazon Lookout for Vision edge inference is billed in edge inference unit increments, with a minimum of 1 edge inference unit per month per edge appliance.
Pricing examples

Example 1: Finding weld defects in automobile engine assembly
Let’s say you are an automobile manufacturer and want to identify weld defects in all the engines assembled in your factory using Lookout for Vision. For simplicity, assume for eight hours a day, every minute, one engine unit is inspected for weld defects before it is sent to the next stage of your assembly line. The first bill at the end of the month (22 days, assuming no weekend usage) using Amazon Lookout for Vision in US East (N. Virginia) would be calculated as follows:Monthly training charges
Assume it takes three training hours to train your model. On the first day you use 3 training hours for which you leverage the free tier. Your monthly training charges would be as detailed in the table below:Usage type Hours used Price Free tier usage Number of billable hours Charge for the month Training 3 hours $2.00/hour
3 hours
0 0 Monthly cloud inference charges
Assume you run inference on the cloud daily (8 hours/day * 22 days) for 176 inference hours. On the first day you use 4 inference hours for which you can leverage the free tier. For the remaining time, you would incur charges as detailed in the table below.Usage type Hours used Price Free tier usage Number of billable hours Charge for the month Cloud Inference 176 $4.00/hour
4 hours
176  4 = 172 hours 172 * $4.00/hour = $688.00
Total Charges: $688
Monthly edge inference charges
Now, let’s say you want to run inference onpremises. Let’s assume that you have 20 unique camera streams that are being processed through the Lookout for Vision models deployed on your edge appliances. For simplicity, let’s assume that these 20 unique camera streams will use 20 edge inference units.
Usage type Edge Inference Units Price Free tier usage Number of billable inference units Charge for the month Edge Inference 20 $100/edge inference unit /month
5 edge inference units per month
205 = 15 edge inference units 15*$100 = $1,500
Total monthly charge
For the month, your bill will be $2,188, a total that includes $0 for training, $688.00 for cloud inference, $1,500 for edge inference, and a complete use of the free tier.
Usage Type (per month) Charge for the month Training + Cloud Inference Hours + Edge Inference Units Total Charge = $0 + $688 + $1,500 = $2,188 
Example 2: Finding missing components in printed circuit boards (PCBs)
Let’s say you are a PCB manufacturer, and you want to check for PCBs with missing components before they are sent to the next processing stage. For simplicity, assume you need two cloud inference units to process the number of PCBs manufactured in an 8hour workday. Assume you’ve already used all your free tier usage for this month. The bill at the end of using Amazon Lookout for Vision for 22 days in the month (assuming you are not running the lines during the weekends that excludes eight days in the month) in US East (N. Virginia), would be calculated as follows:Monthly training charges
Assume you retrained your model for two training hours to improve your model performance. The monthly training charges would be as detailed below.
Usage type Hours used Price Number of billable hours Charge for the month Training 2 hours $2.00/hour
2 hours 2*$2.00/ hour = $4.00 Monthly cloud inference charges
Assume you use cloud inference hours daily (8 hours/day x 22 days x 2 inference units) for 352 inference hours. For the 22 working days of the month, you would incur charges as detailed in the table below.
Usage type Hours used Price Number of billable hours Charge for the month Cloud Inference tier 1 (First 300 hours) 300 hours $4.00/hour
300 hours 300 * $4.00/hour = $1,200.00
Cloud Inference tier 2 (Next 3000 hours) 52 hours $3.60/hour
52 hours 52*$3.60/hour = $187.20 Total Charges = $1,387.20 Total monthly charge
For the month, your bill will be $1,391.20, a total that includes $4 for training, $1,200.00 for cloud inference hours (tier 1), and $187.20 for cloud inference hours (tier 2).
Usage Type (per month) Charge for the month Training + Cloud Inference Hours Total Charge = $4 + $1,200+ $187.20 = $1,391.20 
Example 3: Finding surface defects in silicon wafers
Let’s say you manufacture silicon wafers on three production lines. You want to check for wafers with spots and discolorations before they are sent to the next processing stage. You want to use this model onpremises to run inference on the edge so you can check for spots and discolorations in real time. Assume you have already used all your free tier usage for this month. The bill at the end of using Amazon Lookout for Vision for the month in US East (N. Virginia), would be calculated as follows:Monthly training charges
Assume you retrained each model per production line for two training hours to improve your model performance (2 hours x 3 production lines) for six training hours.
Usage type Hours used Price Number of billable hours Charge for the month Training 6 hours $2.00/hour
6 hours 6*$2.00/ hour = $12.00 Monthly edge inference charges
You have 1500 unique camera streams, each monitoring a production line that produces a maximum of 120 silicon wafers per minute. Each connected camera stream is sending a maximum of 120 images per minute to the Lookout for Vision model for inspection, thus consuming one edge inference unit. You have 3 cameras connected to one edge appliance, which means you have a total of 500 edge appliances (1500/3). For the month, you will be using 3 edge inference units * 500 edge appliances = 1500 edge inference units. For the month, you incur charges as detailed in the table below.
Usage type Inference Units Price Free Tier Usage Number of billable edge inference units Charge for the month Edge Inference Tier 1 (First 500 edge inference units) 500 $100/edge inference unit/month
0 500 500 *$100 = $50,000
Edge Inference Tier 2 (Next 2000 edge inference units 1000 $80/edge inference unit/month
0 1000 1000*$80 = $80,000 Total Charge = $50,000 + $80,000 = $130,000
Total monthly charge
For the month, your bill will be $130,012, a total that includes $12 for training, $50,000 for edge inference units (Tier 1), and $80,000 for edge inference units (tier 2).
Usage Type (per month) Charge for the month Training + Cloud Inference Hours + Edge Inference Units Total Charge = $12 + $50,000+ $80,000 = $130,012 
Example 4: The total cost of ownership using Lookout for Vision compared to other existing Machine Vision based solutions
Now, let’s look at the total cost of ownership using Lookout for Vision to detect anomalies. Let’s assume you have one production line that manufactures 100 pizzas per minute. You have an industrial camera connected to an NVIDIA Jetson GPU edge appliance. You deploy your pizza defect defection model onto this edge appliance to detect defective pizzas on your production line.Hardware costs per camera
This includes the cost of the industrial camera (including the camera installation and commissioning) and the edge appliance (if applicable)
Components Details Using Lookout for Vision on edge for anomaly detection Other Machine Vision anomaly detection solution providers Cost of an industrial camera (including installation and commissioning) Camera + installation costs $2,250
$10,000 Cost of the edge appliance – per camera One edge appliance supports 3 cameras ($1,600/3) $534
0 Total $2,784 $10,000 Model training and update costs per camera
Let’s assume that using Lookout for Vision you train the model for 20 hours and update the model once every quarter (initial trained model + 3 additional model updates per year). For simplicity, let’s assume you’ve finished your free tier for training and inference.
Components Details Using Lookout for Vision on edge for anomaly detection Other Machine Vision anomaly detection solution providers Training 20 hours for training @ $2.00 per hour $40
0 (included in camera costs) Additional model updates (3) 3 * $40 $120
$5,000 Total $160 $5,000 Edge Inference costs per camera
One connected camera feed to detect defective pizzas being manufactured at 100 pizzas per minute will use one Lookout for Vision edge inference unit per month.
Components Details Using Lookout for Vision on edge for anomaly detection Other Machine Vision anomaly detection solution providers Edge Inference Units (Tier 1) 1 edge inference unit per month ($100 * 12) $1,200
0 (included in camera costs) AWS IoT Greengrass for model deployment and management $0.16 per appliance per month $1.92
Total $1,201.92 Total cost for Year 1 per camera
If we sum up the hardware, training, and inference costs the total cost of ownership using Lookout for Vision compared to other existing Machine Vision based solution is as below:
Components Using Lookout for Vision on edge for anomaly detection Other Machine Vision anomaly detection solution providers Hardware costs $2,784
$10,000 Training and model updates $160
$5,000 Inference $1,201.92 Total cost of ownership $4,145.92 $15,000
Additional pricing resources
Easily calculate your monthly costs with AWS
Contact AWS specialists to get a personalized quote
Get started building with Amazon Lookout for Vision in the AWS Management Console.