Removedor de Objetos
Pinte sobre qualquer objeto indesejado e remova-o de forma limpa com IA. Créditos diários grátis.
Drop an image here, or click to browse
PNG, JPEG, or WebP. We'll auto-detect the object to remove — or brush over it yourself.
Brush over what you want gone, keep the rest of the photo exactly as it was
AI inpainting that reconstructs the background behind an object you paint out. Only the masked pixels change, everything else is preserved. Auto-detect handles overlay marks on its own, and the result opens in a before/after slider so you can check the edges before you download.
How it works
Upload the photo
Drop or pick a PNG, JPEG, or WebP up to 12 MB. Confirm you own the image or have the rights to edit it, which is required before every job.
Auto-detect or brush
Auto mode finds overlay-style marks on its own. Brush mode gives you a size-adjustable brush with undo and clear, so you paint exactly the region you want gone.
Compare, then download
The result opens in a drag-to-compare before/after slider. Check the edges, then download the cleaned PNG. Not right? Undo and brush a tighter mask.
What is actually in the tool
Pointer-precise brushing
Brush size from a fine tip up to 120 pixels, drawn with pointer events so mouse, trackpad, touch, and stylus all work. Undo drops the last stroke, clear resets the mask without losing the image.
Auto-detect mode
Grounding DINO, an open-vocabulary detector, is prompted for watermarks, copyright marks, logo overlays, and signatures. Detected boxes are dilated and merged into a mask automatically, so a typical corner watermark needs no brushing at all.
An over-detection guard that protects the photo
A detection larger than 25 percent of the image, or a total coverage above 35 percent, is thrown out and reported as not detected. Detectors do box entire subjects, and inpainting that region would return a smeared photo instead of a clean one. Refusing is the correct answer.
Mask-bounded inpainting
LaMa regenerates only the masked pixels. Everything outside the mask is preserved exactly, so there is no global re-encode, no colour shift across the frame, and no quality loss on the 95 percent of the photo you did not touch.
Before/after slider on the result
The cleaned image is shown under a draggable divider against the original. Verifying the edges of the patch takes a second, which matters because inpainting failures are usually visible only at the boundary.
Refund on failure
Credits are charged up front and returned automatically if the model fails, times out, or the detector finds nothing usable. You are not billed for an attempt that gave you no image.
Alpha handled correctly
A PNG with transparency is flattened onto white before detection, because transparent pixels reaching the model as black is what turns a result into a black rectangle. Opaque photos pass straight through untouched.
No image retention
The cleaned image is returned inline in the response, not stored. The job row we keep for compliance holds metadata only: status, model, credits charged, affirmation, timestamp, IP. No original, no output, no mask.
What it does well, and where it fails
Inpainting does not recover what was behind an object. It invents something plausible and blends it to the mask edge. That distinction predicts every good and bad result you will get, so it is worth stating before you spend a credit.
What it does well
Objects that sit on a continuable background: a stranger on a beach, a bin on grass, a road sign against sky, a date stamp in a corner, a logo over a flat area, dust and blemishes. Small to medium regions with a mask that hugs the object come back clean enough to publish.
What it struggles with
Large objects, anything occupying a big share of the frame, objects overlapping faces, and objects in front of text, tiles, brickwork, or straight architectural lines. The model invents a plausible background, so structure that must line up across the patch often does not.
What auto mode is and is not
Auto-detect is aimed at overlay marks: watermarks, logos, signatures. It is not a general "find the object I dislike" button. For an arbitrary object in a scene, brush it. That is not a limitation of the brush path, it is where the brush path is genuinely better.
What it does not do
No generative fill of new content, no object replacement, no background swap, no upscaling, no batch queue, no video. One image at a time, remove only. The result is a PNG at your uploaded resolution.
Credits, limits, refunds
| Item | Value | Detail |
|---|---|---|
| Cost per removal | 50 credits | Flat, regardless of image size or how much you brushed. |
| Free daily grant | 200 credits | Refreshed each UTC day on a free account, so 4 removals a day at no cost. |
| Failed job | 0 credits | Charged up front, refunded automatically when the removal fails or nothing is detected. |
| Upload limit | 12 MB | PNG, JPEG, or WebP. Anything larger is rejected before any credit is spent. |
Sign-in is required so credits and the per-job ownership affirmation attach to an account. There is no premium gate on the tool itself: a free account gets the same model, the same modes, and the same output as a paid one, just fewer removals per day.
What people use it for
Clear a stranger out of a holiday photo
Brush over the person in the background. Sand, sea, and sky are exactly the kind of background the model continues convincingly.
Remove a camera date stamp
Old cameras burned an orange date into the corner. Brush it, or let auto mode catch it as an overlay mark.
Clean a product shot
Cables, tape, clips, and stand feet that crept into the frame on a plain backdrop come out cleanly, and the product itself is untouched because the mask never covers it.
Tidy a property or listing photo
Bins, hoses, and stray signage on the driveway. Keep it honest: cosmetic clutter, not structural defects a buyer needs to see.
Remove your own watermark from your own file
You watermarked an export, lost the original, and need a clean copy. Auto mode is built for exactly this shape of mark.
Delete a distraction before a crop would ruin the composition
Sometimes cropping the object out costs you the shot. Removing it keeps the framing you wanted.
Walkthroughs
Remove an object with the brush
- 1Upload the photo and tick the ownership confirmation.
- 2Switch the mode toggle to Brush and set the brush size to roughly the width of the object.
- 3Paint over the object. Cover it fully, including its shadow and any halo, but do not paint half the photo: a tight mask reconstructs better than a generous one.
- 4Remove, then drag the before/after divider across the patched area to check the edges before downloading.
Let auto mode handle a corner watermark
- 1Upload the image. Auto is the default mode, so there is nothing to switch.
- 2Press remove. Detection and inpainting run in one pass, typically a few seconds.
- 3If it comes back saying nothing was detected, the mark was either too faint for the detector or too large to be safe to remove, and your credits are already back.
- 4Switch to Brush, paint over the mark, and run it again.
Fix a result that smeared
- 1Smearing almost always means the mask was too large or the background was too structured.
- 2Start again from the original and brush a tighter mask that follows the object outline rather than a rectangle around it.
- 3For an object in front of a straight line or repeating pattern, remove it in two smaller passes rather than one large one.
- 4If the object covers a large share of the frame, accept that inpainting is the wrong tool and crop or reshoot instead.
Keep the untouched parts provably untouched
- 1Run the removal, then hold the before/after slider at the edge of the patch.
- 2Everything outside the mask is preserved pixel for pixel by the model, so only the patched region should differ.
- 3Download the result. It is delivered as a PNG, so saving it does not add a fresh round of JPEG artifacts on top of the patch.
- 4If you need a smaller file for the web afterwards, compress the finished PNG separately.
What runs when you press remove
1. Detect (auto mode only)
Grounding DINO, an open-vocabulary detector, is given the image plus a text query for watermarks, copyright marks, logo overlays, and signatures. It returns boxes with confidence scores. In brush mode this step is skipped entirely: your strokes are the mask.
2. Build the mask
Boxes below the confidence floor are dropped. Survivors are dilated slightly, merged where they overlap, clamped to the image bounds, and rasterised into a black PNG with white rectangles. Too much white and the whole job is refused rather than run.
3. Inpaint
LaMa takes the image and the mask and regenerates only the white region. It is resolution-robust and mask-bounded, which is why the rest of the frame survives unchanged. The cleaned PNG is returned to your browser inline.
Privacy, retention, and rights
This one is not browser-side, and we will not pretend otherwise
Inpainting needs a GPU, so your image is sent to our server and on to Replicate, where the model runs. Plenty of our tools are 100 percent local. This one cannot be, and a tool that claimed otherwise would be lying to you.
What we do control is retention. The cleaned image comes back inline in the API response and is never written to storage. The job record we keep for compliance contains status, the model used, credits charged, the ownership affirmation, a timestamp, and the request IP. It contains no image, no mask, and no output.
The ownership checkbox is not decoration. Every job requires you to confirm you own the image or have the rights to edit it, and the API refuses the request without it. Removing copyright or attribution marks from content you do not own is prohibited by our acceptable use policy. Use this on your own photos and on work you are licensed to edit.
Who it's built for
Photographers
Clear the litter, the stray tourist, the light stand at the edge of the frame. Mask-bounded editing means the rest of your file is byte-identical to what you shot.
Sellers and marketplace listers
Clean product and listing photos without a desktop editor. Cosmetic clutter only: nothing a buyer needs to see should be inpainted away.
Content and social teams
Pull a date stamp, a competitor logo on your own asset, or a distracting object out of a hero image without opening a full editing suite.
Anyone with one annoying object
No subscription, no install, no learning curve. Upload, brush, download. Four a day free is enough for occasional cleanup.
Questions people ask
What can the object remover actually remove?
How does auto mode know what to remove?
Why does auto mode sometimes say nothing was detected?
Does it change the rest of my photo?
What does it cost?
Do I need an account?
Does my image get uploaded?
What file types and sizes are supported?
Can I remove a watermark from a stock photo I found online?
When will the result look wrong?
Try it now
Remove an object, keep the rest of the photo
Free daily credits, no paid plan, refunded if it fails. Upload an image you own and brush over what should not be there.
Start removingA página Removedor de Objetos é construída, revisada e mantida pela equipe Molixa. Usamos a ferramenta que disponibilizamos e atualizamos a documentação quando o comportamento muda.
Ferramentas de Imagem relacionadas
Compressor de Imagens
Multi-encoder lado a lado, slider de comparação visual, lote zip, sem limite de 20/mês.
Image to Text
OCR in your browser. Paste a screenshot, get text. 16 languages, no upload, no signup.
Conversor de Imagem para Base64
6 formatos de snippet, upload em lote, avisos de tamanho, apenas no navegador.
Gerador de Favicon
10 tamanhos PNG + ICO + manifest + trecho HTML, tudo em um ZIP.
Conversor de SVG para PNG
DPI personalizado, 7 predefinições de resolução, lote + ZIP, PNG/JPG/WebP.
Redimensionador de Imagens
37 predefinições, 4 modos, exportação em lote ZIP, apenas no navegador.
Ferramentas Populares
Detector de Conteúdo IA
Verifique se um texto foi escrito por IA com um mapa de calor frase por frase.
Resumidor de Vídeos do YouTube
Transforme qualquer vídeo do YouTube em anotações claras com capítulos, citações, chat e flashcards.
Resumidor de PDF
Resumidor de PDF com citações de páginas, comparação entre documentos e modelos por domínio.
Reescritor de Texto com IA
Parafraseie em 10 modos com visualização de diferenças, congelamento de palavras e treinamento de voz da marca.
Resolvedor de Matemática com IA
Resolvedor de matemática gratuito com soluções passo a passo, upload de foto e 4 modos de aprendizado.
Formatador e Validador de JSON
Formate, valide, visualize em árvore, gere schema, tipos e diff. 100% no navegador.