Vertex AI Vision pricing
This document explains Vertex AI Vision pricing details.
If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.
Billing framework and discounting schedule
| Feature | Sub-feature | PAYG pricing | Monthly pricing |
|---|---|---|---|
| Streams | Data Ingested | $0.0085 / GB | NA (same as PAYG) |
| Data Consumed | $0.0085 / GB | NA (same as PAYG) | |
| Pre-trained AI Models and AutoML | Person / Vehicle Count | $0.10 per min | $10 per stream per month |
| Person / Face Blur | $0.10 per min | $10 per stream per month | |
| PPE Detection | $0.10 per min | $10 per stream per month | |
| General Object Detection | $0.10 per min | $10 per stream per month | |
| Occupancy Analytics Suite | NA | $20 per stream per month | |
| AutoML (detection) for Streams | $0.20 per min | $20 per stream per month | |
| Product recognizer1 | $0.025 per 1k images | NA | |
| Tag recognizer1 | $0.025 per 1k images | NA | |
| Visual Inspection AI2 - Anomaly | N/A | $100 per camera stream per solution per month | |
| Visual Inspection AI2 - Assembly | N/A | $100 per camera stream per solution per month | |
| Visual Inspection AI2 - Cosmetic | N/A | $100 per camera stream per solution per month | |
| Custom AI | Training - Visual Inspection AI | $2 / node hour | NA |
| Vertex Custom Models | Vertex node price | NA | |
| Vision Warehouse | Video storage for streams | $0.020 per GB-mo | NA (same as PAYG) |
| Asset storage for images | $0.020 per GB-mo | NA (same as PAYG) | |
| Index node hour for batch videos and images | $3 / node hour | NA (same as PAYG) | |
| Search request for batch videos and images | $3 per 1k requests | NA (same as PAYG) |
1 Each image can contain either one product or one tag
2 Visual Inspection AI is currently in GA (restricted) - Contact Google sales for more info.
Billing examples
Example 1 (with Model)
Under a PAYG model, customers will only pay for components they use. Let's use a simple vehicle counting app for illustrating the pricing.
The components used to build this Car counting app are Streams (Ingress) and Model (Vehicle Counting). User will connect one stream to the vehicle counting processor. At this point no charges will be incurred. Once a user starts streaming video to the app, charges will be incurred.
App Cost = Streaming costs + Analytic costs
Streaming costs = Stream ingressed (GB) * Stream price / GB ($0.0085/GB)
- Total Stream ingressed (GB): Clip size per minute * minutes streamed / 1024
- Clip size: 60 MB / minutes, Minutes of video streamed for processing: 100
- Total Stream ingressed (GB) = 60 MB / Minute * 100 minutes / 1024 MB= 58.6 GB (rounded)
Streaming costs = Stream ingressed (GB) * Stream price / GB
Streaming Costs = 5.86 GB * $0.0085 / GB = $0.049
Analytic Costs = Minutes of Video processed for streaming * Vehicle Counting Processor price / min
Analytic Costs = 100 minutes * $0.10 / min = $10
App Cost = Streaming costs + Analytic costs
Therefore, App Cost = $0.049 + $10 = $10.049
Example 2 (with Warehouse)
Under a PAYG model, customers will only pay for components they use. Let's use a simple warehouse app for illustrating the pricing.
Users will connect one stream to the warehouse and at some point will also download clips that they have searched for in the warehouse. Once a user starts streaming video to the warehouse, charges will be incurred as follows assuming the video is stored for two months.
Total Cost = Month 1 + Month 2
Month1 Cost = Streaming costs (Ingress) + Warehouse costs + Streaming costs (Egress)
Streaming costs (Ingress) = Stream ingressed (GB) * Stream price / GB ($0.0085/GB)
Video stream ingressed (GB) : 10 GB
Streaming costs = Stream ingressed (GB) * Stream price / GB
Streaming Costs = 10 GB * $0.0085 / GB = $0.085
Warehouse Costs = Stored Video (GB) * Storage costs (per month)
Warehouse Costs = 10 GB * $0.020 / GB-mo = *$0.2 / month
Streaming costs (Egress) = Video Clips egressed (GB) * Stream price / GB ($0.0085/GB)
Video clip egressed: 10 MB or equivalently 0.01 GB
Streaming costs (Egress) = Video Clips egressed (GB) * Stream price / GB
Streaming Costs = 0.01 GB * $0.0085 / GB = $0.000085
Month 1 Cost = Streaming costs (Ingress) + Warehouse costs + Streaming costs (Egress)
Month 1 Cost = $0.085 + $0.2 + $0.000085 = $0.285085
Similarly, Month 2 Cost: Assuming no further streaming (ingress or egress) of video and clips Month 2 Cost = Streaming costs (Ingress) + Warehouse costs + Streaming costs (Egress)
Month 2 Cost = $0 + $0.2 + $0 = $0.2
Total Cost = Month 1 cost + Month 2 cost
Total Cost = $0.285085 + $0.2 = $0.485085
Vision Warehouse billing examples for batch videos and images
Vision Warehouse for batch videos and images has a different pricing model than for streaming videos. The pricing consists of:
- Storage cost for images charged as $0.02 per GB, per month.
- Search query cost charged as $3 per 1k request.
- Index node hour cost for each virtual machine used to build and host index shards.
The node hour cost is mainly determined by the number of images and the total duration of batch videos.
It consists of a one-time asset analysis cost (charged by
AnalyzeAsset and AnalyzeCorpus)
to build the index data and a continuous index serving cost (charged by
the serving time starting at DeployIndex and ending at
UndeployIndex).
The one-time index build cost can be estimated as follows:
- Images index build cost in node hours: 0.043 node-hour per 1k images.
- Batch video index build cost in node hours: 0.292 node-hour per video hour.
To estimate the 30-day index serving cost, first divide your index data into shards. The number of index serving nodes is total index shards times total index replicas.
- One index shard can hold up to 40 million images or 1,200 video hours.
- A deployed index has two replicas per shard.
- Images index serving cost: (# images / # images-per-shard) * # replicas-per-shard * hourly cost * 720 hours.
- Batch video index serving cost: (# video-hours / # video-hours-per-shard) * # replicas-per-shard * hourly cost * 720 hours.
Here are some examples of Vision Warehouse cost given different asset data sizes.
All examples are based on pricing in us-central1.
| Data size | Image storage | Index shards | Serving nodes | Search QPS | One-time index build cost | 30-day storage cost | 30-day index serving cost | 30-day search cost |
|---|---|---|---|---|---|---|---|---|
| 40 million images | 18.6 TB* | 1 | 2 | 1 | $5,160 | $381 | $4,320 | $7,776 |
| 1 billion images | 466 TB* | 25 | 50 | 10 | $129,000 | $9,537 | $108,000 | $77,760 |
| 1,200 video hours | N/A | 1 | 2 | 1 | $1,051 | $0 | $4,320 | $7,776 |
| 30,000 video hours | N/A | 25 | 50 | 10 | $26,280 | $0 | $108,000 | $77,760 |
* Assume the average image size is 500 KB.
What's next
- Read the Vertex AI Vision documentation.
- Get started with Vertex AI Vision.
- Try the Pricing calculator.
- Learn about Vertex AI Vision solutions and use cases.