Pretrained CLIP · ViT-B/32

What does CLIP see?

Give it an image and a few descriptions. Change something. Then follow the numbers behind the match.

Model not loaded

Run on

First download: about 600 MB on WebGPU, or 170 MB on CPU. Cached for later visits.

Runs on your device. Uploaded images and descriptions stay in this browser.

Start with a question

What to look for

1. Choose an image

Choose from the sample gallery

Select an image to use it.

Or upload an image

PNG, JPEG or WebP · up to 10 MB · processed locally

One description per line

Use 1–12 descriptions. CLIP scores the words you supply; it does not write a caption.

Your prediction (optional)

Choose your inputs while the model loads.