Limits

Consultation Service

If you have questions about the Alibaba Cloud Visual Intelligence API, contact us through our DingTalk support group. Search for group number 23109592 to join. The group is open to all enterprise users, developers, and service providers.

Data limits

• Data source: For the best performance, provide input data from Alibaba Cloud OSS. Use an internal network address within the same region. Do not use cross-region access.

• Image resolution: The dimensions of an input image cannot exceed 5000 × 5000 pixels, and the file size cannot exceed 9.5 MB.

  Note

  • If this limit conflicts with a specific API limit, the API's size limit prevails.

  • If an image exceeds the maximum resolution, you can scale it to a smaller size. For more information, see Image scaling.

• For API-specific limits, see the documentation for that API.

• Processing time: The processing time of certain APIs, such as the invisible watermark addition API, is proportional to the size of the input image. The larger the image, the longer the processing time. To get an accurate estimate for your specific parameters, test the API in OpenAPI Explorer.

Data outputs

• Output method: All APIs that output data provide results through a temporary URL. This URL is valid for 30 minutes and expires afterward. For long-term or permanent retention, download the file from the URL within 30 minutes and save it to your own OSS bucket or another storage location.

• Data retention: In accordance with the user security and privacy agreement, the Visual Intelligence API and its features do not save any input or output data, including images you upload and their processing results. You are responsible for managing all related data, especially the output results. You cannot view the history of images you have previously uploaded or processed through the platform.

Call limits

• The following Visual Intelligence API features are available only to enterprise users who have verified their identity: fingerprint extraction, video face fusion, template-based video face fusion, image face fusion, public figure recognition, and celebrity recognition. After you enable these features, you must submit a permission request in the console to use them. For all other features, you only need to enable the service, which is free of charge.

  Note

  Users who enable General Video Face Fusion, Template-based Video Face Fusion, and Image Face Fusion on or after July 17, 2024, can call these features directly without submitting a request.

  If you enabled these services before July 17, 2024, and need to use them, contact us through our DingTalk support group (23109592) to be added to the whitelist.

• The Visual Intelligence API offers the following billing methods: pay-as-you-go, single-category subscription resource plans, general subscription resource plans, and subscription QPS. Charges are deducted in the following order: Subscription QPS > Single-category subscription resource plan > General subscription resource plan > Pay-as-you-go.

  Note

  For more information about billable services, see Billable items and Billing methods.

• QPS limits:

  • For pay-as-you-go features, the default QPS limit is 1 QPS for the OCR category and 2 QPS for all other categories.

  • For features that use a subscription resource plan, the default QPS limit is 5 QPS. After you purchase a resource plan, the new QPS limit takes effect at 08:00 on the following day.

  • For subscription QPS, your QPS limit is based on the amount that you request at the time of purchase.

  For some features, you can increase the QPS limit by making a purchase or by contacting us through our DingTalk group (23109592).

  Note

  Only an Alibaba Cloud account can perform the confirmation, not a RAM user or a role.

• Embedded device limits: The official SDK depends on runtime environments such as the JVM or Python interpreter, as well as third-party libraries that cannot be ported to the limited flash memory and RAM of microcontrollers (MCUs) such as ESP32. The built-in SSL/TLS stack on ESP32 also consumes significant resources when handling complex signature authentication logic. Therefore, calling the API directly or implementing signature authentication on MCUs such as ESP32 is not supported. You can use Alibaba Cloud Function Compute (FC) as a relay to process requests and forward the results to the ESP32 device.

• Region limits: The service is currently available only in the China (Shanghai) and China (Shenzhen) regions. Other regions will be supported in the future. International accounts cannot be used to enable or purchase the service. Although cross-region calls are supported (for example, calling a service endpoint in China (Shanghai) from China (Beijing)), the physical distance and network transmission increase latency and reduce response speed, particularly for features such as face and human body recognition. We recommend that you deploy your business in the same region as the service endpoint (China (Shanghai)) to minimize latency.

  Note

  The license plate recognition feature is available in the China (Shenzhen) region.

• Access limits: The service is deployed in the Chinese mainland. Calls from servers outside the Chinese mainland may experience network instability and timeouts. Therefore, calls from Hong Kong (China), Macao (China), Taiwan (China), and other regions outside the Chinese mainland are not supported.

FAQ

Q: How do I stop using the Visual Intelligence API or stop being billed?

A: Enabling the service is free of charge. You are only billed when you make API calls. If you stop making calls, no fees are incurred. To completely revoke API access and prevent accidental calls, you can delete the AccessKey that is used to call the service. Also, make sure that no calls are being made from the debugging page.

Q: Does the Visual Intelligence API support the mcap data format?

A: The mcap data format is not currently supported. APIs typically accept image data in URL format. Convert your data to a supported format, or provide it through an OSS URL.