How do I lift objects out of an image into separate editable layers?

Detect, pick, lift

• Run Detect objects on the selected image or the current page — the model returns up to a dozen distinct objects with labels and bounding boxes.
• The preview draws a labeled, color-coded box per object; click a box to include or exclude it.
• Apply, and each chosen object is cropped out and placed as its own image layer at the exact position it occupied — then drag it anywhere.

The clean plate problem

Moving an object usually exposes the problem underneath: the object still sitting in the base image. The "Reconstruct base" option runs a second pass that removes the chosen objects from the source and repaints the background, so the lifted layers move over a plate that looks like they were never there. Skip it when you only need the crops.

Layers are just annotations

Each lifted object is an ordinary image annotation: drag it, resize it, restyle it, re-edit its colors in the overlay editor, or delete it — the layer model is the annotation model, so nothing about the result is locked to the tool that made it.

Honest requirements and limits

Detection and the optional reconstruct pass send the image to the model — it needs a Gemini API key configured in Settings, and the image leaves the device for those calls. Detection lists at most twelve objects and can miss ambiguous ones; boxes you exclude simply are not lifted.

Doing it in PDFDepth

Right panel → Image Editing Mode → Magic Layer, on a selected image or the page. Detect, untick anything you do not want, decide whether the base needs reconstructing, and apply — the objects land as draggable layers on the sheet.