[{"data":1,"prerenderedAt":172},["ShallowReactive",2],{"tool-content-en-image-object-remover":3},{"id":4,"title":5,"body":6,"description":162,"extension":163,"locale":164,"meta":165,"navigation":166,"path":167,"seo":168,"stem":169,"toolId":170,"__hash__":171},"tools\u002Ftools\u002Fen\u002Fimage-object-remover.md","AI Object Remover - Erase Passers-by",{"type":7,"value":8,"toc":143},"minimark",[9,14,18,22,50,54,72,76,79,82,86,89,93,96,100,105,108,112,115,119,122,126,129,133,136,140],[10,11,13],"h2",{"id":12},"what-does-this-object-remover-do","What does this object remover do?",[15,16,17],"p",{},"It deletes things from a photo and reconstructs what should have been behind them. Paint over a passer-by who wandered into your shot, or click an object to have it outlined automatically, then press Erase. Two models run on your device: MI-GAN reconstructs the background, and Segment Anything traces object outlines when you click one. Your photo is never uploaded.",[10,19,21],{"id":20},"key-features","Key features",[23,24,25,29,32,35,38,41,44,47],"ul",{},[26,27,28],"li",{},"Brush selection: paint over whatever should disappear, with an adjustable brush and an unpaint tool for fixing a slip.",[26,30,31],{},"Pick object: click something and a Segment Anything model traces its outline. Shift+click (or right-click) marks a spot that is not part of it.",[26,33,34],{},"Grow selection: expands the selection outwards before erasing, which is what keeps an object's soft edge and contact shadow from being left behind.",[26,36,37],{},"Iterative: every erase applies to the current result, so you can clear several things one after another.",[26,39,40],{},"Undo any erase with the button or Ctrl\u002FCmd+Z.",[26,42,43],{},"Compare with the original at any point.",[26,45,46],{},"Export to PNG, JPEG, or WebP with adjustable quality.",[26,48,49],{},"Fully local: the image is decoded, edited, and re-encoded in your browser tab.",[10,51,53],{"id":52},"how-to-use-it","How to use it",[55,56,57,60,63,66,69],"ol",{},[26,58,59],{},"Drop a photo onto the upload area or click to browse.",[26,61,62],{},"Choose how to select: Brush to paint over the area, or Pick object to click it.",[26,64,65],{},"Adjust \"grow selection\" if the result leaves a faint outline behind.",[26,67,68],{},"Press Erase. The model downloads on the first use and is reused from cache afterwards.",[26,70,71],{},"Repeat for anything else you want gone, then export.",[10,73,75],{"id":74},"why-small-strokes-beat-one-big-selection","Why small strokes beat one big selection",[15,77,78],{},"The model reconstructs the masked area from what surrounds it, so the more context it can see around your selection, the better it guesses. A selection that swallows half the frame leaves almost nothing to reason from, and the result turns mushy. Clearing a group of people works better as several passes, one person at a time, than as a single sweep over all of them. The model's authors give the same advice.",[15,80,81],{},"For the same reason a busy, highly structured background — brickwork with strong perspective, text, faces — is harder than grass, sky, sand, or a plain wall. Straight lines that run through the erased area are where a reconstruction is most likely to look wrong.",[10,83,85],{"id":84},"grow-selection-and-why-it-matters","Grow selection, and why it matters",[15,87,88],{},"An object's visible boundary is not where it ends. There is usually a halo of anti-aliased pixels, plus a contact shadow on the ground. If the selection stops at the crisp edge, those leftovers stay in the image, and the model treats them as real context to blend with — reconstructing a faint ghost of what you tried to delete. Growing the selection by a few pixels sweeps them into the erased region. Growth is measured as a percentage of the shorter edge, so the same setting behaves the same on a phone snap and a 40-megapixel photo.",[10,90,92],{"id":91},"privacy-and-security","Privacy and security",[15,94,95],{},"Every pixel is decoded, edited, and re-encoded in your browser using the Canvas API and ONNX Runtime Web. No image data is sent anywhere. Only the model files are downloaded, once, and then served from the browser cache.",[10,97,99],{"id":98},"faq","FAQ",[101,102,104],"h3",{"id":103},"is-my-photo-uploaded-to-a-server","Is my photo uploaded to a server?",[15,106,107],{},"No. Both models run in your browser through ONNX Runtime Web, and the only network requests are for the model files themselves. The photo stays in the tab.",[101,109,111],{"id":110},"how-big-is-the-download","How big is the download?",[15,113,114],{},"About 27 MB for the MI-GAN inpainting model, downloaded on your first erase. If you use \"pick object\", Segment Anything adds roughly 13 MB more. Both are cached afterwards.",[101,116,118],{"id":117},"how-long-does-one-erase-take","How long does one erase take?",[15,120,121],{},"A couple of seconds, and it barely depends on the size of your photo: the pipeline crops around your selection and always works at 512x512 internally, so a 40-megapixel photo costs about as much as a small one. Using \"pick object\" adds a one-time few-second pass per image to analyse it.",[101,123,125],{"id":124},"can-it-remove-a-watermark-or-text","Can it remove a watermark or text?",[15,127,128],{},"It can, for small marks over ordinary backgrounds. Text over detailed content tends to leave traces, and the result is a reconstruction rather than a recovery of what was hidden — the original pixels are gone, not restored.",[101,130,132],{"id":131},"why-does-the-result-look-blurry-or-invented","Why does the result look blurry or invented?",[15,134,135],{},"The model is filling in content that does not exist in the file. Large selections, strong perspective lines, and repeating patterns are the usual causes. Undo, select less, and go again in smaller passes.",[101,137,139],{"id":138},"is-there-a-size-limit","Is there a size limit?",[15,141,142],{},"Processing is local, so the practical limit is your device memory. Photos over 40 megapixels trigger a warning because each erase then moves noticeably more data around.",{"title":144,"searchDepth":145,"depth":145,"links":146},"",2,[147,148,149,150,151,152,153],{"id":12,"depth":145,"text":13},{"id":20,"depth":145,"text":21},{"id":52,"depth":145,"text":53},{"id":74,"depth":145,"text":75},{"id":84,"depth":145,"text":85},{"id":91,"depth":145,"text":92},{"id":98,"depth":145,"text":99,"children":154},[155,157,158,159,160,161],{"id":103,"depth":156,"text":104},3,{"id":110,"depth":156,"text":111},{"id":117,"depth":156,"text":118},{"id":124,"depth":156,"text":125},{"id":131,"depth":156,"text":132},{"id":138,"depth":156,"text":139},"Free browser-only object remover. Brush over a passer-by or click an object, and an on-device model rebuilds the background. No upload, no watermark.","md","en",{},true,"\u002Ftools\u002Fen\u002Fimage-object-remover",{"title":5,"description":162},"tools\u002Fen\u002Fimage-object-remover","image-object-remover","F4bim_quOu1AXrcrjKavC4V7IBIqHCITdtzXQX363ac",1792391418304]