Limits

Common limits for Vision, and those limits specific to Document AI and Image analysis models.

Image Analysis Models

Limits for Image Analysis Models
Limit Limit Value
Accepted File Formats JPEG and PNG
Maximum File Size
Single request
Up to 5 MB per image
Batch request
Up to 5 MB per image
Maximum Image Count
Single request
Single image
Batch request
Each job can have 2,000 images or fewer, or 500 KB or less in the body of the request

Stored Video Analysis Models

Limits for Stored Video Analysis Models
Limit Limit Value
Accepted File Formats MOV, MP4, H264, MKV, and WEBM
Maximum File Size Up to 20 GB and 10 hours duration.

Document AI Models

Important

The AnalyzeDocument and DocumentJob capabilities in Vision are moving to a new service, Document Understanding. The following features are impacted:
  • Table detection
  • Document classification
  • Receipt key-value extraction
  • Document OCR
These features are available in Vision until January 1, 2024. After then, they're available only in Document Understanding.
Limits for Document AI Models
Limit Limit Value
Accepted File Formats JPEG, PNG, PDF, and TIFF
Maximum File Size
Single request
5 MB
Batch request
500 MB per document
Maximum Document Count
Single request
Five pages
Batch request
  • 2,000 pages per document
  • Each job can have 2,000 documents or fewer, or 500 KB or less in the body of the request

Common

Limits Common to All Models
Limit Limit Value
Minimum Resolution 32 x 32 pixels
Maximum Resolution 10,000 x 10,000 pixels
Text Alignment Text can be text aligned horizontally within the document. Vertical text alignment within the document is not supported.
Languages OCR supports English.
Character Size The minimum height for text to be detected is 15 pixels. At 150 DPI, the height is the same as eight-point font.
Character Type Handwritten character and printed character recognition are supported.
Characters
  • a - z
  • A - Z
  • 0–9
  • % } + ~ ^ = * ; ÷ _ ( € < , " @ # ¤ £ ¢ / § ? ] . ¥ > ₹ - ® ` ! © & $ \ ' { \\ [ | ) :
Maximum Number of Projects in a Tenancy 25
Maximum Number of Models in a Tenancy 25
Maximum Number of Asynchronous Jobs in Pending State Allowed in a Tenancy 200
Minimum Number of Labeled Images for each Label for a Custom Model 10
Maximum Number of Pretrained Synchronous Transactions Allowed per Second per Tenancy
  • 10 transactions per second for US East (Ashburn), US West (Phoenix), and UK South (London) regions
  • Other regions: Five transactions per second for all other region