<aside> 🛠

Pick the Right Object from SAM2/SAM3 Segmentation

After SAM2 or SAM3 segments a scene, every object comes back as a generic label (object_0, object_1, …). Object Picker opens an interactive modal of the segmentation overlay so you can click the objects you actually want, then builds a clean binary B&W mask for downstream steps. No more guessing which index is which.

</aside>

What it does

Object Picker is an interactive node that sits directly after a SAM2 or SAM3 segmentation node. It receives the colored segmentation overlay (IMAGE) and the per-object metadata (TEXT) from SAM3, opens a modal where each segmented region is displayed with its description, and lets you click to include or exclude objects. Once you confirm your selection, Object Picker outputs a single binary (black/white) mask covering exactly the chosen objects, ready to feed into masking, inpainting, blending, color, or any other downstream tool in the flow.

Problem it solves

• Decode anonymous SAM2 / SAM3 indices - SAM3 returns objects as object_0, object_1, etc. Object Picker shows you what each index actually is so you can pick correctly
• Visual selection over guesswork - Click the object on the image instead of typing index numbers blindly
• Multi-object selection in one step - Pick several objects at once (e.g., '0,2,4') and merge them into a single mask
• Include or invert in one toggle - Choose between masking only the picked objects, or masking everything else
• Precise hand-off to downstream tools - Produces a clean binary mask compatible with all CNAPS Studio masking, inpainting, and blending tools
• Skip manual brushing - No need to redraw masks the segmentation model already gave you
• Reduce trial-and-error - Confirm the right target visually before running the full flow

Input/Output

<aside>

• Input: SAM2 / SAM3 output (Image + Text)
  • IMAGE - Colored segmentation overlay produced by SAM2 or SAM3
    • Source: SAM2 or SAM3 (Scene Segmentation) node — upstream
    • Format: RGB image with each object rendered in a distinct color
  • TEXT - Per-object metadata describing each segmented object

    • Source: SAM2 or SAM3 - upstream

    • Format: Object index + description (e.g., object_0: "person on the left", object_1: "red car")

    • Purpose: Powers the picker UI so you can see what each colored region represents

      

      

</aside>

<aside>

• Output: Binary mask d

  • Format: Black/white binary mask (PNG)
  • Resolution: Same as input image
  • Content: White pixels for selected objects, black for everything else (or inverted, depending on mode)

  

  

</aside>

Features

• Objects to include — set via the picker UI
  • How it works: Click "2 selected" (or current count) on the node to open the segmentation overlay modal. Each colored region is paired with its SAM2 / SAM3 description. Click regions to add or remove them from the selection.
  • Format: Comma-separated indices (e.g., '0,2,4') - this is filled in automatically as you click in the modal
  • Purpose: Defines exactly which segmented objects participate in the output mask

• Mode — select or invert

  Option	Behavior	Use Case
  select	Mask covers ONLY the picked objects	"Mask just the person and the dog"
  invert	Mask covers everything EXCEPT the picked objects	"Keep the subject untouched, process the background"

How to Use with Segmentation