- Does my photo get uploaded?
- No. The model that finds the subject runs on your own computer, inside this browser tab, and the cut-out is written into the pixels there too. The only thing that ever downloads is the model itself, once, from our own server. Your image never goes the other way. You can check: open your browser's network panel after the first use and drop a photo in, and you won't see a single request. Most cut-out services work the other way round. They upload the photo to their servers, run the model there, and send back the result, which is also why they can meter it. For a product shot that doesn't matter much. For a photo of a person, a document, or anything you'd rather keep, it does.
- Is the download really full resolution?
- Yes, and it's the point of the tool. The model looks at a small copy of your photo, 320 × 320 pixels, because that's the size it was trained on. What comes out is a map of which parts are subject and which are background. That map is scaled up smoothly to your photo's real size and written into the alpha channel of your original pixels, which are never resized. So the subject in your download has every pixel it had in your photo. It hasn't been shrunk to the model's size and blown back up. The line under the preview shows the exact dimensions and file size you'll get before you download.
- What kind of photos work best?
- One clear subject that stands out from what's behind it: a person against a wall, a product on a table, a logo on a plain box, a pet on a floor. The model is picking the thing a person would look at first, so the more obvious that is, the cleaner the edge. It has a harder time with fine hair against a busy background, glass and other see-through things, and scenes where there's no single subject at all, like a landscape or a crowd. Check the edges with Before / After before you use the result somewhere that matters. If the photo has no clear subject, say a blank wall or a sky, the tool tells you so instead of handing back an empty PNG.
- Why does the first use take longer?
- Because the model and the engine that runs it have to reach your browser once, about 18 MB on most machines and around 30 MB where the browser supports WebGPU, which runs the model on your graphics card. The progress line shows the real megabytes as they arrive, so you're never looking at a spinner and guessing. Both files are stored in your browser's cache and checked against a fingerprint before they're used, so a download that broke halfway is thrown away and fetched again, never run. From the second photo on, and on every later visit, the tool goes straight to the cut-out.
- Can I put the subject on a colour instead of transparency?
- Yes. Switch Background to White or Colour and the result changes instantly, because the tool keeps the cut-out it already made and only redraws what's behind it. The model doesn't run again. The soft edge where the subject meets the background is blended into the new colour rather than left as a hard line. That's what stops the grey or black fringe you get when a transparent cut-out is dropped onto a colour later in another app. If your original already had transparent areas, they stay see-through on a transparent background and take the new colour on a solid one.
- Why can't it open my iPhone photo?
- Many iPhone photos are HEIC files, even when they're named .jpg, and most browsers can't decode HEIC. The tool reads the first bytes of the file rather than trusting the name, spots HEIC, and tells you instead of failing with a vague error. Run the photo through HEIC to JPG first (also in your browser, also no upload) and bring the JPG back here. Phone photos that are already JPG work directly, and the tool respects the rotation your phone saved, so a portrait shot doesn't come out sideways.
- Why isn't there an account or a credit balance?
- Because cutting the background out of one photo isn't worth a subscription. The usual pattern here is a free preview at a fraction of your photo's size, then credits, a plan or an account for the real file, and sometimes a watermark on top. That model exists because those services pay for a server to run every image. This one runs on the computer you already own, so there's nothing to charge per image. No sign-up, no watermark, no resolution cap below the 12000-pixel limit a browser tab can handle, and no count of how many photos you run through it.
