Paint over what you want gone and the model repaints what should have been behind it.
Clone-stamping an object out of a photo works until the background has structure — brickwork, a horizon, a face half-hidden behind the thing you are deleting. A generative model does not copy nearby pixels; it reconstructs what the scene plausibly looked like without the object, including perspective and shadows.
ImageLabs runs the erase pass on Nano Banana Pro, so removals hold up on textured backgrounds where traditional healing brushes smear.
Upload the photo: JPG, PNG, WebP or HEIC. Higher-resolution inputs give the model more context to rebuild from.
Brush over the object: Cover the object and a little of its shadow or reflection — those give it away more often than the object itself.
Describe what should be there: Optional, but 'empty pavement' or 'plain sky' steers the fill when the background is ambiguous.
Shadows and reflections too: Include them in the mask and they disappear with the object instead of hovering where it used to be.
Keeps the original resolution: Output comes back at the size you uploaded or the next step up, never smaller.
Iterate for 5 credits: The prompt stays in the box, so a second pass is a tweak rather than a retype.
Can I remove a person from a photo for free? Signed-in accounts get 10 free credits every day, and new users can try Pro free for 24 hours — cancel anytime.
Does it work on busy backgrounds? Yes — that is where a generative fill beats a healing brush, because it reconstructs structure rather than blending neighbouring pixels.
Will the rest of the photo change? No. Only the masked region is repainted; everything outside it is untouched.
What about large objects? Large removals work best when the background is simple or partly visible elsewhere in the frame.