Skip to content
AI.info

The Pulse

Google Cloud details AI metadata extraction for storage objects

Google Cloud has detailed a preview service that extracts metadata from images and PDFs stored in Cloud Storage. The company says upcoming Gemini support will extend the service to audio, video and other tasks.

Google Cloud details AI metadata extraction for storage objects

Google Cloud customers can use a preview service to extract metadata from stored images and PDFs and attach it to the files for use by AI agents. The company's blog post details object enrichments, which applies pretrained models to Cloud Storage objects and saves their output alongside them as object context.

Cloud Storage release notes list the preview as available since Oct. 13. It can be accessed through the Cloud Storage console or API.

Models and costs

The preview offers five first-party models for images and PDFs. The image models can detect objects and faces, label images and assess explicit content; the PDF model classifies documents into business and industry categories such as finance, news and health.

Google says enriching 10 million images or PDFs costs $700 under published pricing. The company also claims its image annotations offer 75% cost savings compared with similar services from other cloud providers. Charges also apply for model inference, metadata storage, data access operations and networking, according to the product documentation.

Hyperscience said in the post that it achieved a more than 50% lift in document classification accuracy while reducing classification inference costs by 2–3 times. That customer account compares the results with an existing baseline.

Metadata stays with files

Customers select buckets and a processing location, choose models and grant the service permission to access the buckets. The service processes eligible objects asynchronously, handles model inference and writes the results back to object context, with retries and throttling built in.

The metadata is stored as structured key-value pairs alongside the object. The context travels with the object during copy, move and replication operations and is removed when the object is deleted. Administrators can use identity and access management permissions to control who can read or modify the context.

Applications can filter object listings by those key-value pairs or read context alongside objects. For analysis, customers can configure the metadata to flow into BigQuery through Storage Insights datasets, where it can be queried with SQL and joined with structured business data.

Google identifies creative asset search, content moderation and training-data selection as potential uses. It also describes pairing object context with the Cloud Storage Model Context Protocol server to give agents access to stored data and its context.

Limits and upcoming Gemini support

Images and PDFs must be 20 MiB or smaller. Objects with missing or invalid content-type metadata are not processed, even if their underlying data is a valid image or PDF. The feature also does not support objects encrypted with customer-supplied encryption keys, Rapid buckets or buckets with hierarchical namespace enabled.

Google plans to offer Gemini through object enrichments for tasks including summarization, entity extraction, transcription and translation across text, PDFs, images, audio and video. The company asks customers to contact it to learn more and request access to that upcoming functionality; it is not part of the capabilities described as currently available in the preview.

Sources

  1. Original source · Blog post Google Cloud cloud.google.com

Other pages read

  1. Google Cloud Documentation · page 1 docs.cloud.google.com
  2. Google Cloud Documentation · Release notes docs.cloud.google.com

Companies in this story

Related stories

Explore

More articles