Remove objects from photos without upload
Paint over whatever spoils the photo – passers-by in the background, a sign, a power line, a stain – and the AI fills the area to match its surroundings. The model (LaMa) is downloaded once and then runs on your device: your photo is not uploaded anywhere. The rest of the image keeps its full resolution.
- Paint over the whole object – including its shadow and reflection.
- Rather mark a little too much than too little.
- If something is left: just paint over it again.
LaMa (Apache 2.0) · one-time ≈ 62 MB ✓ stored
Pick a photo on the left and paint over everything that should disappear.
How it works
Choose a photo
Drop a photo on the field, pick it, or paste it from the clipboard. The first time, the browser downloads the model (about 62 MB) and keeps it.
Paint over distractions
Use the brush to cover everything that should disappear – preferably a little generously and including the shadow. Zoom in for precise work.
Remove & save
One click on “Remove” fills the marked areas. If something is left, just paint over it again. Then download or copy the image.
Frequently asked questions
Is my photo uploaded?+
No. The photo is edited in your browser and does not leave your device. The only download is the AI model (from Hugging Face), once; the runtime is served by mytoolio itself.
What can be removed well?+
Small to medium-sized things in front of a fairly even background: people in the distance, signs, cables, bins, stains, scratches or a date stamp. Sky, grass, water, walls and tarmac are usually filled without a trace.
Why does the area sometimes look washed out or odd?+
The AI cannot know what was behind the object – it invents a plausible continuation. That works less well for large areas, fine patterns, text, faces, or when the object takes up most of the image. The model also processes every area at 512 × 512 pixels. “Sharpen large areas” sharpens bigger areas with an upscaling model and adds the grain of the surroundings. Tip: remove large objects in several steps.
How long does it take?+
With a current graphics card (WebGPU, e.g. Chrome or Edge) about one second per marked area. Without WebGPU the processor does the work – 20 to 30 seconds per area. Several separate markings are filled one after the other.
Is the image quality preserved?+
Yes. Only the marked areas are changed; everything else stays the same pixel for pixel. Photos up to 4096 pixels on the long side keep their size, larger ones are reduced to that. The file is saved in its original format; metadata such as the location (EXIF/GPS) is not included in the new file.
Does it work on a phone and offline?+
Yes. On a phone you paint with one finger and zoom with two. Once the model has been downloaded, the tool also works without an internet connection; “Delete stored models” frees the space again.
Can I just tap an object instead of painting over it?+
Image completion with LaMa (Resolution-robust Large Mask Inpainting, Apache 2.0 licence) on onnxruntime-web – entirely on your device. The filled areas are invented by the AI and do not show what was actually behind the object. Please only edit photos you have the necessary rights to.