1. Choose the smallest useful mask
Open Remove Watermark and add a JPG, PNG, or WebP image. Rectangle is a good default for wide labels. Circle, Square, Diamond, and 4-Point Star can keep the selection tighter when the visible mark has a matching footprint. Text creates a solid glyph-shaped mask for visible words, numbers, or phrases.
A tighter mask usually leaves more useful surrounding pixels available to the repair algorithm.
2. Move, resize, or rotate instead of redrawing
Click or tap an existing mask to select it. Drag inside to move it, drag a square corner handle to resize it, or drag the round handle outside the selection to rotate it. The rotation handle automatically uses the side with room when a mask sits close to an image edge.
Text-mask bounds are fitted to the typed content, so a single digit or short word does not carry a large invisible box that prevents corner placement.
3. Let the batch choose a starting mode, then override it if needed
When uploaded images share an aspect ratio, the editor starts with One mask for all. This is useful when a repeated watermark stays at the same relative position, angle, and approximate size. Mixed aspect ratios start with Edit individually, because one normalized mask is less predictable across different layouts.
The recommendation is not a lock. You can switch modes yourself. A practical workflow is to create one shared mask first, then switch to individual editing and customize only the exceptions.
4. Repair locally and inspect the After preview
The source images stay in the browser. The worker repairs selected visible pixels from nearby unmasked pixels, then the editor encodes the result as the source format, PNG, JPG, or WebP. A single result downloads directly; multiple successful results can be packaged into a ZIP.
Look closely at edges, repeated textures, faces, hair, letters, and architecture. If the result is not convincing, make the mask tighter or adjust its position, size, or rotation on that image and run the repair again.
5. Know the limits before using a large batch
The current editor accepts up to 50 images, with a 10 MB limit per source image and a 200 MB total source limit. Small visible marks over smooth or repeated backgrounds are generally easier than large marks covering unique detail. The tool edits visible pixels only; it does not claim to remove invisible provenance, Content Credentials, authenticity metadata, or SynthID-style signals.
Watermark removal is destructive image editing. Save the unedited source when licensing, attribution, provenance, or archival history matters.