Google Cloud Vision API
cloud.google.com
- Free plan
- Windows
- macOS
- Linux
- Web
- API
What Google Cloud Vision API is
Google Cloud Vision API gives developers image-analysis features they can add to applications through REST or RPC. It supports image labeling, face and landmark detection, OCR, explicit-content tagging, object localization, and web detection. Text Detection reads text in images, while Document Text Detection handles dense text, handwriting, and PDF or TIFF files. Object Localization returns labels and bounding boxes for multiple objects. Web Detection can return related web entities, matching image URLs and pages, visually similar images, and a best-guess label. Face Detection finds faces and facial landmarks and provides likelihood ratings, but does not identify specific people. Google documents workflows with Cloud Storage, Cloud Functions, Pub/Sub, and Cloud Translation API, and recommends ML Kit for Firebase for Android and iOS development using its vision services. Limits include 20 MB per image, 10 MB per JSON request, 16 images per synchronous annotate request, and 2,000 per asynchronous batch request. The first 1,000 units per feature each month are free; charges vary by feature and usage tier.
Who it suits
It suits developers looking to add prebuilt image-analysis capabilities to applications. Mobile developers should note Google's recommendation of ML Kit for Firebase for Android and iOS.
Good to know, and what to watch
- Access is available through REST and RPC.
- Document Text Detection supports handwriting and PDF or TIFF files.
- Object Localization returns labels and bounding boxes.
- The first 1,000 units per feature each month are free.
- Face Detection does not identify specific people.
- Images are limited to 20 MB each.
- Synchronous requests are limited to 16 images.
- Asynchronous batches are limited to 2,000 images.
Plans
Every plan and priceCloud Vision API pay-as-you-go
First 1,000 units
Units above 5,000,000
Units 1,001 to 5,000,000
Free plan
Getting started
- Open the Google Cloud Vision website
- Choose the API access method, REST or RPC
- Connect the API with your application on a supported platform
- Select the image features your workflow needs
- Use documented integrations such as Cloud Storage or Cloud Functions
- Start with the free plan and move to paid usage from $1.50 when needed
Limits to check
The documented limits are 20 MB per image, 10 MB per JSON request, 16 images per synchronous annotate request, and 2,000 images per asynchronous batch request. The first 1,000 units per feature each month are free; further use is billed by feature and usage tier.
The Club's review
Overview
Google Cloud Vision API is a cloud service that gives developers prebuilt image-analysis features through REST or RPC. It is best suited to teams adding several vision tasks to applications without building each capability themselves. Its broad feature set is useful, but per-feature billing and the absence of individual face recognition are important trade-offs.
Key features
Image labeling, object localization, face and landmark detection, OCR, explicit-content tagging, and Web Detection cover a varied set of application needs. Object Localization returns labels and bounding boxes for multiple recognized objects, which is valuable when an application needs to locate items rather than merely assign an image a category.
Text Detection recognizes text in images; Document Text Detection handles dense text and handwriting and supports PDF and TIFF files. That distinction makes the API more suitable for document-heavy workflows than a basic image-text reader. Web Detection can return related entities, matching image URLs, pages containing matching images, visually similar images, and a best-guess label, giving image-based products several kinds of context.
Face Detection locates faces and facial landmarks and provides likelihood ratings, but it does not identify individuals. It is therefore unsuitable for identity recognition. Explicit-content tagging supports image moderation, but the API does not provide text moderation.
Google documents workflows with Cloud Storage, Cloud Functions, Pub/Sub, and Cloud Translation API. Payload limits matter for larger jobs: images may be up to 20 MB, JSON requests up to 10 MB, synchronous annotate requests can contain 16 images, and asynchronous batch requests can contain 2,000.
For Android and iOS development, Google recommends ML Kit for Firebase, which offers SDKs using Cloud Vision services as well as on-device vision APIs and custom-model inference. Cloud Vision API itself is presented as a cloud deployment option, so mobile teams should consider that recommendation rather than assume this API is the preferred mobile integration.
Pricing
The Cloud Vision API pay-as-you-go plan charges by image and feature: each feature applied to an image is a billable unit, with monthly usage tiers. The first 1,000 units per feature each month are free. Label Detection costs $1.50 per 1,000 units for 1,001–5,000,000, then $1.00 thereafter. This model is useful for trying several capabilities at modest volume, but applications that apply multiple features to each image should account for each feature as a separate billable unit.
The free allowance resets by feature each month, rather than covering unlimited use. The supplied pricing details do not give complete rates for every feature, so developers should not extrapolate the Label Detection rates to OCR or other operations.
Platforms
Cloud Vision API is available through REST and RPC, with platform support for API, Linux, macOS, web, and Windows. Google states that customers own their data and that it processes data under customer agreements; its compliance center provides certifications, attestations, and audit reports. Standard Support offers unlimited individual access to technical and billing support through multiple channels.
Who it's for
This is a strong fit for developers who want multiple prebuilt image-analysis capabilities in an application, especially where OCR, object location, image labeling, or image moderation can share an integration. It is less suitable when individual face identification, text moderation, or a mobile-first on-device implementation is central to the project.
Pros and cons
Pros
- Several image-analysis features are available through one API, reducing the need to assemble separate services for common vision tasks.
- Document Text Detection supports dense text, handwriting, PDF, and TIFF, which extends OCR use beyond ordinary images.
- The first 1,000 units per feature each month are free, allowing limited use across different features before paid tiers apply.
- Google documents integration workflows with Cloud Storage, Cloud Functions, Pub/Sub, and Cloud Translation API.
Cons
- Each feature applied to an image is a billable unit, so multi-step analysis increases usage charges.
- Face Detection provides likelihood ratings but cannot recognize specific people.
- There is no text-moderation capability, limiting its use for products that need both image and text moderation.
- Image, request, and batch-size limits may constrain large or unusually sized submissions.
Alternatives
For teams comparing image-recognition or moderation products, see AI Image Recognition Software and AI Content Moderation Software.
- MediaPipe is a better fit when free, on-device ML solutions and a framework are the priority.
- Roboflow offers a free tier with 10 credits monthly, enough to train about 30 models or run 80,000 inferences, making it an option for teams whose work centers on model training or inference.
- SentiSight.ai is an alternative for teams considering an offline model license with a single-time custom quote based on model speed.
- DeepDetect offers an open-source single-machine option for CPU or GPU use, for teams seeking a self-hosted route.
- Imagga API may suit a small project that can work within 100 monthly API requests and its basic tagging, categorization, cropping, and color solutions.
- Nyckel includes 100 invokes per month and up to five functions in its free plan, an alternative for low-volume API use.
- PaddleOCR is an alternative for document parsing, with an official API free tier capped at 20,000 pages daily.
- LandingLens is another option for teams evaluating image-analysis platforms.
Verdict
Choose Google Cloud Vision API if you are a developer who needs several prebuilt image-analysis features behind one cloud API and can manage per-feature usage charges. Its breadth and monthly free allowance make it practical to start with, but look elsewhere if you need identity recognition, text moderation, or an on-device mobile-first approach.
The verdict
Cloud Vision API offers several image-analysis features through developer interfaces, with a free monthly allowance for each feature. Check the request and image limits, and remember that face detection is not identity recognition.
Questions people ask
Does Google Cloud Vision API have a free plan?
Yes. The pricing model is freemium, with a free plan and paid usage from $1.50.
Which platforms can use the API?
It is listed for API, Linux, macOS, web, and Windows environments.
Can it recognize a specific person from a face?
No. Face Detection locates faces and landmarks and provides likelihood ratings, but specific individual facial recognition is not supported.
What OCR options are available?
Text Detection reads text in images. Document Text Detection handles dense text, handwriting, and PDF or TIFF files.
What integrations are documented?
Google documents workflows with Cloud Storage, Cloud Functions, Pub/Sub, and Cloud Translation API.
What the maker publishes
- Access
- The API is available through REST and RPC. source
- API access
- Programmatic access is offered through client libraries and REST, with RPC also documented as an API reference. source
- Authentication
- Google Cloud IAM and Application Default Credentials are supported, with service-account and workload-identity options described for different environments. source
- Client languages
- Official quickstarts and samples cover C#, Go, Java, Node.js, PHP, Python and Ruby. source
- Company founding
- Gartner lists Google's year founded as 1998. source
- Company headquarters
- LinkedIn lists Google's headquarters in Mountain View, California. source
- Company size
- LinkedIn lists Google as a public software-development company with 10,001+ employees. source
- Crop hints
- Crop-hint detection can return salient-region polygons and supports up to 16 requested aspect ratios. source
- Documentation depth
- Documentation includes setup guides, quickstarts, OCR and handwriting guides, REST/RPC references, quotas, languages, samples and release notes. source
- Encryption
- Google Cloud encrypts data at rest by default; asynchronous Vision processing uses ephemeral encryption keys and supports CMEK compliance. source
- Face detection limit
- Face Detection locates faces and facial landmarks and provides likelihood ratings, but specific individual facial recognition is not supported. source
- File limits
- Uploaded image files should not exceed 20 MB and base64 JSON requests are limited to 10 MB. source
- Image features
- Features include image labeling, face and landmark detection, OCR, and explicit-content tagging. source
- Image labeling
- Label detection returns generalized image labels with descriptions, confidence scores and topicality ratings. source
- Image properties
- Image-properties analysis reports dominant RGBA colors, confidence and pixel fractions. source
- Image search
- Web Detection can return related web entities, matching image URLs, pages containing matching images, visually similar images, and a best-guess label. source
- Integrations
- Google documentation shows Vision API workflows with Cloud Storage, Cloud Functions, Pub/Sub, and Cloud Translation API. source
- Integrations
- Google documentation demonstrates workflows connecting Vision with Cloud Storage, Cloud Functions, Pub/Sub, Translation, Text-to-Speech and DLP. source
- Intended users
- The API is presented for developers seeking quick integration of prebuilt vision features into applications. source
- Landmark detection
- Landmark detection returns a landmark name, confidence score, bounding box and coordinates. source
- Limit: console and onboarding
- Gartner reviewers include a complaint that the console is unclear and the quickstart is slow. source
- Limit: cost predictability
- G2 reviewers repeatedly report that costs rise quickly or become difficult to predict when multiple features and high image volumes are used. source
- Limit: customization cost
- A TrustRadius reviewer says highly specialized fine-tuning may be better handled with local open-source libraries to avoid API costs. source
- Limit: image quality and customization
- G2 reviewers report weaker results on blurry or complex images and request more model customization and debugging controls. source
Checked against the maker's pages on 21 Oct 2026.