Innovation & Tech

Google Cloud AI Metadata Preview Costs $1,400 for 10 Million Images

Google Cloud’s object enrichments preview automates image and PDF metadata in Cloud Storage, with a $1,400 worked example for 10 million files and strict 20 MiB limits.

By Nathan Brooks

5 min read

Google Cloud Previews AI Metadata Extraction in Cloud Storage
Google Cloud Previews AI Metadata Extraction in Cloud Storagecookieone / pixabay

What's News

  • Google Cloud introduced object enrichments in preview on November 19, 2026.
  • Each image model costs $700 for 10 million images; using both produces $1,400 in one-time inference charges, and Google says annotations cost 75% less than similar rival services.
  • The same example adds $55.20 per month; object contexts list at $3 per million contexts per month.
  • The preview caps objects at 20 MiB, caps PDF classification at 100 million objects per customer, and excludes Rapid and hierarchical-namespace buckets.
  • Vivint says object enrichments and BigQuery cut AI development time from six months to one month.

By Nathan Brooks

Google Cloud’s November 19, 2026 AI metadata preview puts object detection and image labeling at $1,400 in one-time inference charges for 10 million images.

The $1,400 figure covers both models, or $700 each, in Google Cloud’s worked example. It adds $55.20 in recurring monthly charges for object contexts, Storage Intelligence and dataset storage. The object enrichments service moves AI inference orchestration into Cloud Storage, letting agents query prepared metadata instead of reprocessing stored frames.

Google Cloud takes over the pipeline

Google Cloud said customers can enable the fully managed feature through the Cloud Storage console or API. The service asynchronously scans eligible objects, runs selected models, parses their responses and writes results to each object. Built-in retries and throttling replace the queues, workers, scheduling and custom pipeline code customers would otherwise build around Cloud Run or Google Kubernetes Engine.

Google described an operations system that searches thousands of smart-camera or vehicle frames for package delivery or roadway hazards. Running inference on every prompt would consume tokens and add minutes of delay, Google said. Preparing metadata in advance shifts that work into asynchronous storage-side processing, while keeping each result attached to its object instead of a separately synchronized store.

Early customers report faster workflows

Google Cloud published early customer results with the announcement. L’Oreal principal data and AI architect Antoine Castex said the service labeled the company’s creative-asset repository “100 times faster at a fraction of existing pipeline costs.” Hyperscience vice president of partnerships Mayur Pillay reported a document-classification accuracy lift exceeding 50%, with inference costs two to three times lower.

Vivint vice president of product AI Brandon Bunker said object enrichments and BigQuery reduced AI development time “from six months to one month.” The result gives the preview a concrete schedule comparison, rather than a generic productivity claim.

Five models write queryable metadata

Google Cloud lists five first-party models, including the following:

  • Object detection returns categories and rectangular boundaries.
  • Image labeling returns categories.
  • Face detection adds facial landmarks and expression likelihoods.
  • PDF classification assigns document categories with scores.

The outputs include business and industry categories such as finance, games, news and health. Google Cloud attaches the annotations to source objects rather than a separately synchronized store.

Google Cloud stores annotations as object contexts, a Cloud Storage metadata construct containing a key, string value and timestamps. An optional JSON payload can hold up to 128 KiB of detailed results.

Applications can filter object listings by context keys and values, while Identity and Access Management permissions control who can add, change or remove them. Contexts also transfer automatically during copy, rewrite, compose, move and restore operations.

Storage Insights datasets can export metadata across millions or billions of objects to BigQuery, Google Cloud says. Data teams can query the object_contexts_catalog_view with standard SQL and join labels or document classifications with structured business tables. Google says the setup supports asset discovery, training-data curation and multimodal analytics without a separate ETL pipeline.

Pricing rewards volume, but location can duplicate work

Google says its image annotations cost 75% less than similar services from other cloud providers. Its published rate is $0.25 per 1,000 objects for each image model’s first 1 million objects each month, then $0.05 per 1,000 above that threshold. That produces the $700 per-model charge in the 10-million-image example.

Google Cloud’s example also charges $55.20 per month for object contexts, Storage Intelligence and dataset storage. Object contexts alone cost $3 per million contexts per month. Customers also owe data-access and networking charges, and object reads can trigger storage-class transitions in buckets using Autoclass.

Google Cloud makes PDF classification free until general availability later in 2026. Public preview usage is capped at 100 million objects per customer, and configuring the same bucket and model in more than one processing location repeats both the work and the charge.

The preview has hard boundaries

The preview accepts image and PDF objects no larger than 20 MiB, Google Cloud says. Each object needs a valid supported Content-Type value identifying its file type.

Google Cloud supports BMP, GIF, ICO, JPEG, PNG, TIFF and WEBP images. Processing runs in the us, eu and asia locations, which customers can choose for data sovereignty and proximity to the bucket’s storage region.

Google Cloud bars object enrichments on Rapid buckets and hierarchical-namespace buckets. It does not support customer-supplied encryption keys and requires exact bucket names rather than wildcards or regular expressions.

Google Cloud sets no explicit workload-size limit, but capacity constraints can delay processing. Cloud Monitoring tracks pending objects, model-execution counts and processing status.

Gemini remains outside the preview

Google lists native Gemini enrichment as upcoming rather than included in the preview. It intends the feature for multimodal reasoning across text, PDFs, images, audio and video, with summarization, entity extraction, transcription and translation among the prospective uses. Google directs interested customers to request access separately.

What does this change for cloud teams?

For teams building agents over camera, vehicle and document repositories, object enrichments changes the unit of work from repeated inference calls to metadata queries. Google Cloud says the approach cuts token consumption and multi-minute prompt delays, while keeping results attached to objects through copy, move and restore operations.

The trade-off remains explicit: Google manages orchestration, but customers pay for each model run, stored context and data read. Workload volume, bucket placement and processing location therefore matter as much as model choice, because reads add charges and multi-location configurations repeat inference work.

Before general availability later in 2026, teams can test whether the managed pipeline delivers enough savings to justify those charges. PDF classification remains free during public preview, but Gemini enrichment requires separate access and several bucket configurations remain excluded.

Google’s next proof point is Gemini availability and broader bucket compatibility. Until then, workload placement, model choice and bucket eligibility will determine whether the managed workflow pays off.

Source: unite.ai

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

2,473 articles

Related articles

« Previous article