Click-to-Segment: The Fastest Way to Get the Exact Mask You Want

4/25/2026

Text prompts are great when you know the name of the object. But what if you want a specific part of it — or need to exclude something right next to it? Click-to-segment lets you point at exactly what you want, and the model follows your finger.

Try it now →

What is click-to-segment?

Instead of typing "person" and hoping the model picks the right one, you click on the image itself:

  • Include points (green) tell the model "this area belongs to the object."
  • Exclude points (red) tell it "ignore this nearby area."

Combine them and you can isolate a single shoe from a pile, separate a person from the bike they're holding, or grab just the logo on a busy product shot — all without writing a single word.

When to use it

ScenarioText promptClick-to-segment
"The largest cat"Works wellOverkill
"The cat on the left, not the one on the right"TrickyOne green + one red point
"Just the shoe, not the shadow"HardOne green on shoe, one red on shadow
"The logo, not the background text"Very hardA few precise points
Object has no obvious nameImpossibleStill works

Use it whenever precision matters more than speed, or when describing the object in words is harder than pointing at it.

How to use it

  1. Upload an image to the Online Segment or 3D Generation page.
  2. Switch to the Visual tab.
  3. Choose Include or Exclude, then click on the image.
  4. Drop as many points as you need — there is no penalty for extra points.
  5. Hit Visual Segment. The model returns the mask in about a second.
  6. Download the transparent PNG, merge it with other objects, or send it straight to 3D generation.

The markers you see on the image are the same icons shown in the picker, so you always know what you placed.

Tips for better results

  • Start with one include point on the center of the object. If the mask bleeds into nearby things, add an exclude point on the border.
  • Negative points are powerful. A single red point on a confusing background patch often fixes more than three extra green points.
  • Combine with other modes. Use click-to-segment to isolate the object, then browse the mask and export.

Also available via API

The same workflow is exposed as a public API endpoint at /v1/pvs:

curl -X POST https://api.sam3.ai/v1/pvs \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image": "base64_encoded_image_string...",
    "points": [
      { "x": 412, "y": 318, "positive": true },
      { "x": 540, "y": 290, "positive": false }
    ]
  }'
  • Same Bearer API Key auth as /v1/segment.
  • 1 credit per call, regardless of point count.
  • Returns mask polygons in image-pixel coordinates.

Full spec in the API reference.

What's next

Click-to-Segment: The Fastest Way to Get the Exact Mask You Want | SegmentLens Blog | SegmentLens - SAM 3 Powered Intelligent Image Platform