Vision models without a dataset

Build a working image model in minutes.

Tell ijk what you want to recognize or segment. Test the model live, improve it through conversation, and download it when you are happy with it. You can start without training images or labels.

Build your model Watch the 30-second walkthrough

A 30-second look at building a segmentation model.

01 / START HERE

Start with an idea, not a dataset.

You do not need to collect and label images before you can try an idea. Describe the task in plain English and get a model you can test right away. Use your camera or your own images to see how it performs on the things that matter to you.

CLASSIFICATION

Know what you are looking at.

Build a model that tells you which of your classes an image matches.

SEGMENTATION

See exactly where it is.

Build a model that marks what matters in an image, pixel by pixel.

02 / HOW IT WORKS

From description to working model.

Describe the task, see what the model does, and keep improving it. Your projects, conversations, and saved versions stay together in your workspace.

01Describe itTell the agent what you want to recognize or segment.
02Test itSee predictions live or try the model on your images.
03Refine itAdjust the task through conversation and save versions as you improve it.
04Take it with youDownload an ONNX model and example code. ONNX is a portable format that runs with tools on many platforms.

Build your model

03 / WHAT'S NEXT

More ways to build when data is scarce.

We are making it easier to build and test useful models with less data.

Improve with a few examples

Adapt a model to your task with a small number of images.

Generate data when you need it

Create examples for training and testing when real data is scarce.

Go beyond images

Build text and audio models through the same conversational workflow.

04 / GET STARTED

Build your
first model.

Start with a description. Test the result. Download a model you can use.