Build the deployment pipeline for your model
This feature is coming soon.


Deployments Pipeline builder is a tool that can be used to configure the inputs and outputs for your model deployment.

Deployment Types

After naming your deployment, the first step is to select your deployment type.


Process data in either scheduled or ad hoc batches. Used for medical image interpretation, recorded video/image capture, satellite imagery, and one-off deployment testing. Supports image and video input types.


Always-on processing capabilities with low latency. Needed for AutoLabel, API integrations with mobile app, etc. Supports image input types.

Deployment Input Types

Image Input

This input type is designed for pipelines that will run on still images.

Video Input

This input type is designed for pipelines that will run on video files. Select the desired Frame Rate for videos to be processed.

Input Source Types

You can choose to sync your images from Google Cloud Storage or Amazon S3 buckets.

Pipeline Blocks

Pipeline blocks are data processors that can be used to define the processing steps in the pipeline. Each block can be individually enabled and configured based on your use case. Individual costs may vary between pipeline blocks.
Users can configure each block to build their deployment pipeline.
Your current pipeline
Pipeline blocks are configurable and can be reordered to define the processing steps.

Image Transformation

Transformation blocks are typically used at the beginning of a pipeline to transform data before sending it to a model.
Image transformation blocks create permanent changes to images in the pipeline. Use with caution.

Image Crop

Crop the image to the specified height and width. A center crop or fixed crop is supported.

Image Resize

Resize the image to the specified height and width. Images can be transformed with a fixed resize or longest size resizing to retain previous detections.


Prediction blocks create inferences, detections, or predictions on images in the pipeline. Predictions are added as annotations on the metadata of an image. These blocks can also be used as inputs in other blocks. Ex. Add a prediction to all assets where motion is detected.
Single Model
Add a custom model to your pipeline.
Tiled Model
Tile images to detect small objects.

General object detection

Detect common objects with a pre-trained model containing over 90 available classes.

Object tracking

Track objects that move through the a series of video frames.

Output Destination Types

You can output your results to a Google Cloud Storage or Amazon S3 bucket, a Google Cloud Pub/Sub, or a webhook.