Eraslino › Guides
Photo eraser guides, tested on real photos
Every guide below draws a mark (text, a watermark, a glare spot, a shadow, an object) onto a public-domain photo, erases it with Eraslino, and measures how far the result is from the untouched original. Each example opens in the eraser with one click, so you can repeat the test in your own browser.
- How to remove text from an image: three backgrounds tested
Captions on sky, wood and brick, erased with both fill methods and scored against the untouched photos.
- How do watermark removers work? What they actually do to the pixels
Inpainting replaces, it doesn't lift: a 45% watermark erased two ways and reversed mathematically.
- How to remove glare from a photo: when erasing works, by spot size
Small, medium and large hot spots on wood: which fill to use and where erasing stops working.
- AI object remover test: how big an object can you erase?
The same ball at four sizes on a lake: where the AI fill stops looking like the real scene.
- How to remove a shadow from a photo (and why erasing it is usually wrong)
A hard shadow on wood: erasing vs brightening by the measured ratio, and four common mistakes.
- How to remove a date stamp or your own watermark from a photo
One orange date stamp over water and over pebbles: when to erase and when to crop.
How we measure
Each test starts from a public-domain or CC0 photo from Wikimedia Commons, scaled to 1280 px wide. We draw the mark (text, watermark, glare, shadow, object or date stamp) onto it with a script, so the photo without the mark is known exactly. The marked photo is what the example link opens in the eraser, with the selection already painted.
We then run the real Eraslino page in headless Chrome on an Apple M4 Pro, once with Smart fill (MI-GAN) and once with Quick patch (Telea), and read the result back from the canvas.
- Error: the mean absolute difference between the result and the unmarked photo over the selected pixels, averaged over red, green and blue, on the 0–255 scale. 0 means identical. "Left in" is the same score for the marked photo, i.e. doing nothing.
- Detail: the average difference between each pixel and its 3 × 3 neighbourhood (a measure of fine texture) inside the selection, kept 4 px away from its edge, divided by the same value for the unmarked photo. 1.00 means as much fine texture as the original; values near 0 mean a smooth fill.
- Time: the time the page itself reports for the fill, with the model already downloaded and cached.
Neither number is a verdict on its own: a smooth smear can score a low error, and a sharp, believable texture in the wrong place a high one. The figures in each guide are shown next to the images for that reason.