Skip to content

Visynx Python SDK

Visynx is a computer vision platform: you label images, train object detection and segmentation models, and deploy them, all in one place. visynx is the official Visynx Python SDK. Give it an API key and a model name and you get predictions back in Python, run on Visynx's GPUs or on your own machine or NVIDIA Jetson. It is for developers and researchers who trained a model in Visynx and want to use it from code: in a script, a backend service, a notebook, or a device on the edge.

Train a model in Visynx, grab an API key, and you are three lines from predictions — detection, segmentation, pose, classification, and semantic segmentation, all through one call.

Features

  • Hosted inference in three lines: predict() takes a path, a folder, an open file, a PIL image or a numpy array, and runs on Visynx's GPUs.
  • Every model type, one result: object detection, instance and semantic segmentation, pose (keypoints) and classification all return the same Result.
  • Extras on the hosted GPUs: SAM 2 instance polygons (segment=True) and monocular depth (depth=True), with nothing extra to install.
  • Run it yourself: download native, ONNX (FP32/FP16) or quantized edge exports and run them offline with visynx[local] (Professional and Enterprise plans).
  • NVIDIA Jetson: TensorRT engines in FP16 or INT8, CSI and USB cameras, and one-command device pairing.
  • Typed errors: one exception per failure, each quoting the request ID, with 429s retried automatically.
  • A light install: the core package depends only on requests and Pillow.

Get started

pip install visynx
  • Installation: how to install visynx with pip, and the extras.
  • Quickstart: your first prediction, local models, NVIDIA Jetson and errors.
  • API Reference: every public class and function.
  • Changelog: what changed in each release.