When to use this tool
- Objects that spoil a photo
- Watermarks and logos on your own images
- Text and labels you no longer need
- Repair work on old photos
AI image tool
Name what should go: the model removes it and rebuilds the surface underneath, keeping the rest of the frame as it was.
JPG, PNG and WebP · up to 10 MiB · 1024 × 1024 PNG output, about a minute


Yes, as long as you own the image: describe the watermark or the line of text as the thing to remove. Dense or semi-transparent text over a busy background is the hardest case.
The model rebuilds the entire frame rather than patching pixels, and it works at a fixed 1024 px square. A larger original is therefore re-rendered at that size.
About a minute on the shared GPUs behind this site. The task is queued if the hardware is busy, and a confirmed failure is refunded automatically.
Only one here: the file is capped at 10 MiB, with no pixel ceiling — the reference image is scaled down inside the service before the frame is re-rendered at 1024 × 1024. (The other tools in the studio also cap the picture at 16 megapixels.)