Remove objects from photos
Paint over a passer-by, a watermark, a date stamp or any unwanted object, and MI-GAN, an AI model that runs in your browser, fills the area with matching background.
Loading…
Everything stays in this browser until you save it; nothing is uploaded.
About this tool
Remove unwanted objects from a photo: paint over a passer-by, a sign, a stain, a watermark or a date stamp, and the area is filled with background that matches its surroundings. It works with JPG, PNG and WebP photos, and with AVIF, BMP and still GIF images, of up to 25 megapixels.
The filling is done by MI-GAN, an AI inpainting model from Picsart AI Research, running in your browser, so the photo is never uploaded. The model, 28 MB, downloads the first time you remove something and is then kept by the browser; after that each removal takes about a second on the processor, whatever the size of the photo. Only the area you marked changes, and every other pixel stays exactly as it was.
How to use it
- Choose a photo or drop it on the page. Convert HEIC photos to JPG first, for example with “Convert image format (HEIC to JPG, PNG, WebP)”, and make photos over 25 megapixels smaller with “Resize images”.
- Set the “Brush size” and paint over everything that should go, covering it completely, edges included. “Clear marks” wipes the marks so you can start again.
- Select “Remove marked area”. The first time, the model is downloaded; the area is then filled in about a second.
- If anything is left over, paint over it and remove it again. “Undo” goes back up to 5 steps, and pressing and holding “Hold to compare” shows the photo as you first opened it.
- Select “Download”: JPG, PNG and WebP photos keep their format, other images are saved as PNG, and the file name ends in “-cleaned”.
Common questions
Is my photo uploaded?
No. Only the model is downloaded from this site; the photo is opened and edited in your browser and stays on your device.
What can it remove cleanly?
Small objects against simple backgrounds work best, such as a person on a beach, a bin on grass or a spot on a wall or in the sky. Large areas, faces and busy patterns may come out smudged or with repeated texture, because the model fills the area from the surrounding picture and works at 512 × 512 pixels internally.
How do I remove a watermark or a date stamp?
Paint over it completely, every letter and its outline, then select “Remove marked area”. Watermarks over plain areas come out cleanest; over detailed parts of the picture the fill may look blurred, so check with “Hold to compare” and repeat on anything left over. Remove watermarks only from pictures you are allowed to edit.
Does the rest of the photo change?
No. Only the marked area is filled; every other pixel stays exactly as it was, and transparency is kept. Saving as JPG or WebP compresses the picture again at quality 0.92, while PNG keeps every pixel as it is. EXIF details such as location are not kept.
Which model is it, and may I use it commercially?
MI-GAN by Picsart AI Research, presented at ICCV 2023. Its code and weights are released under the MIT licence, which allows commercial use. Its licence and source are listed in this site’s third-party notices.
Related tools
Add text watermark
Draw a text watermark on every image with a chosen position, size and opacity. Uses a local system font only.
QR code and barcode reader
Read QR codes and barcodes from up to 20 PNG, JPEG or WebP images, including several codes in one image.
Make an animated GIF
Combine images into one animated GIF in the order given, with size, frame delay, loop count and palette size. Images of other sizes are scaled to fit or fill.
Split a GIF into frames
Export every frame of an animated GIF, or every Nth frame, as a full PNG, WebP or JPEG image composited as a viewer shows it.
Adjust image brightness, contrast and saturation
Change the brightness, contrast, saturation, hue, warmth and sharpness of photos, or make them black and white (grayscale), sepia or negative. The same settings apply to every image in the batch.
Blur faces in photos
Find every face in your photos and cover it with pixels (mosaic), a blur or a black shape, to hide who is in a picture before you share it. Faces are found by a small model that runs in your browser.