Account

What new tool do you want?

Image Tools

Remove Objects, Watermarks & Scratches from Photos

Paint over unwanted things in a photo — scratches, dust, blemishes, date stamps, text or small objects — and fill them in from the surroundings. Undo anytime; runs on your device.

How to use

  1. 1Upload a photo.
  2. 2Paint over what you want removed (adjust the brush size).
  3. 3Press Remove; repeat or undo, then download.

Frequently asked questions

What works best?

Small or thin things on simple backgrounds: scratches and dust on old photos, blemishes, date stamps, text, wires and small objects against sky, walls or grass.

Why is a large object replaced with a blurry patch?

The fill is built from the surrounding colors. Large areas in front of detailed textures (like a crowd) are hard to reconstruct without AI.

Can I remove watermarks from images I don't own?

Only remove marks from images you have the rights to use — watermarks protect creators' work.

How the fill is actually generated: diffusion, not AI generation

The algorithm behind the brush is a multi-scale diffusion fill — it shrinks the image into a pyramid of progressively smaller versions, fills the painted-over area at the coarsest, smallest scale by spreading in color from its known surrounding pixels, then works back up through each larger scale refining that fill with local smoothing, and finally adds matching grain so the patch doesn't look artificially smooth or 'plastic' next to real photo noise. It's the same family of technique long used in photo-retouching software's content-aware fill, not a generative AI model synthesizing new content — which is exactly why it excels at simple, low-detail backgrounds and struggles once the surrounding area has real structure to reconstruct.

Why small, deliberate strokes beat one big pass

Because the fill estimates each erased pixel from its neighbors working inward, a thin scratch or a small date stamp has plenty of nearby real pixels to draw from and fills in convincingly. A single huge brushed area removes most of its own useful neighboring context at once, forcing the algorithm to extrapolate across a much larger gap with less to go on — which is why the tool's own guidance recommends several small passes rather than one large one, and why undo and incremental edits exist as first-class actions rather than a single irreversible operation.

Example: cleaning up an old scanned photo

Say you've scanned a decades-old print that has a few dust specks, a faint scratch running across the sky, and a date stamp burned into the bottom-right corner by the original camera. Set a small brush size, carefully trace just the scratch line and dust specks first and press Remove — against a plain sky these disappear almost perfectly since the diffusion fill has clean, simple surrounding color to work from. Then paint over just the date-stamp digits and remove them separately; treating it as its own smaller pass gets a cleaner result than trying to erase the scratch and the date stamp in one big combined stroke.

Where this hands off to something else

Painting over a person standing in front of a detailed, busy background — a crowd, a patterned rug, dense foliage — asks the algorithm to invent texture it has no real way to infer, and the result is typically a soft, blurred patch rather than a convincing reconstruction; that's a fundamentally harder problem that needs a generative AI model trained on millions of images, not a local diffusion fill. And because watermarks exist specifically to protect a creator's ownership of their work, this tool is meant for cleaning up your own photos — scratches, dust, blemishes, stray objects — not for stripping attribution off images you don't have rights to use.

Related tools

More: Image Tools