Image Extractor
Split a sheet of pictures into one file each — without any crop carrying a piece of the item next to it. Everything happens in your browser, so nothing is uploaded.
Why Use Our Image Extractor?
No Neighbours in Your Crop
A crop is a rectangle and a picture of a thing is not, so on a loosely arranged sheet the rectangles overlap and each crop quietly carries part of the item beside it. Instead of choosing a better rectangle — there isn’t one — every pixel is assigned to whichever item it actually belongs to, and the rest is cleared.
Finds the Pieces Itself
No grid to set up and no boxes to drag. Each separate picture is found from the sheet’s own transparency, and specks are ignored. Feed it one sheet or a folder of them; one slider decides how close two parts have to be to count as the same thing, and it applies to the whole batch.
Join What Was Split
Some things arrive in parts no rule can reliably join — a drawn sun whose rays float clear of its disc, where the gap inside the sun is wider than the gap between two different products next to it. Tick those pieces and merge them; they go back at the exact offsets they were cut from, so the result is the original region, not an approximation.
Named in Reading Order
Pieces come out numbered left to right and top to bottom, with rows worked out from the items’ own heights so a slightly uneven row still reads as one row. Numbers are zero-padded, so the files sort correctly everywhere.
Nothing Leaves Your Machine
There is no upload step and no server doing the work. The sheet is read from your disk, split in your browser, and the PNGs are written there too — which also means a sheet of a hundred items costs nothing and waits on nothing.
Drop your sheets here
One sheet or a folder of them — every picture on each comes back as its own file, with the neighbours taken out
PNG, WebP, GIF and JPG. Nothing is uploaded anywhere.
One Sheet In, Many Files Out
From a grid of pictures to a folder of them
Add Your Sheets
Drop in one sheet or a whole folder of them. Each is read straight off your disk — there is no upload step, so nothing waits on a network and nothing is sent anywhere.
Every Piece Is Found
The tool finds each separate picture on the sheet and puts them in reading order, left to right and top to bottom. It errs towards keeping things apart, so a pen and its loose cap arrive as two pieces rather than risk fusing two different items into one file — one slider joins them back up when you want that.
Neighbours Are Removed
Where two items sit close enough that their rectangles overlap, every pixel is assigned to whichever item it actually belongs to, and the others are cleared. The page tells you how much each crop was carrying.
Correct Anything It Got Wrong
A drawn sun whose rays float clear of its disc arrives as a dozen pieces: drag a box round them on the preview and merge. Two things drawn a pixel apart arrive as one crop: pick it and split. Then take any single piece, one sheet’s worth, or every sheet at once as a ZIP of numbered transparent PNGs.
Questions About Extracting Images
What it does, and where it stops
Why do ordinary crops pick up bits of the picture next door?
Because a crop is a rectangle and a picture of a thing is not. Lay a few items out loosely and their rectangles overlap — on the two sheets this tool was measured against, eleven of twenty pieces on one and fifteen of sixteen on the other had part of a neighbour sitting inside the rectangle they were cut from — the crowded sheet worst of all, which is exactly when it matters. No choice of rectangle avoids that. The only fix is to decide, pixel by pixel, which item each pixel belongs to, and clear the rest. That is what this does.
Are my images uploaded anywhere?
No. The whole thing runs inside your browser — finding the pieces, separating them and writing the PNGs. The page loads the site’s usual analytics, as every page here does, but your picture is not in any of that and never appears in a request.
What kind of sheets does it work on?
Best on sheets that already carry transparency, which is what image generators produce when asked for a set of objects — these need no guesswork at all, because the file already says where each picture ends. It also handles a sheet of items on one flat colour. What it cannot do is separate items from a photographic background; for that there is nothing to measure, and it will tell you it found only one piece rather than pretend otherwise.
How do I pick the pieces to merge?
On the preview, every picture found is outlined and numbered. Drag a box round the parts of one thing — a sun and its rays — and everything the box encloses is picked; click a single outline to take just that one; ctrl-click (⌘ on a Mac) adds or removes one; shift-click takes a run; Escape clears. The box takes only what it encloses completely rather than everything it brushes past, because a neighbour caught by accident would be silently baked into the merged file. The same picking works on the thumbnails below.
It split one item into several. Can I fix that?
Two ways, and the second is the reliable one. Raising the grouping distance joins parts that sit closer together than it — a jar and its lid, a pencil and its cap. But no distance is right for every sheet: measured across three, one needs two pixels and another needs six, and each setting breaks the other. A drawn sun is the plain case — the gap between its disc and its rays is wider than the gap between two entirely different products beside it, so nothing measured from distance alone can tell those apart. So you can also just tick the pieces that are really one thing and merge them. They are laid back at the exact offsets they were cut from, which reproduces the original region precisely rather than approximating it, and the numbering closes up behind them.
Two different things came out in one file. Can I separate them?
Yes — tick that piece and press split. It runs the same analysis again on that piece alone with no grouping distance, so whatever is not truly joined comes apart; a brush resting a pixel above a slice of log becomes two files. Lowering the grouping distance would part them too, but that setting applies to the whole sheet and pulls other things apart that belong together, so splitting the one piece is the safer correction. Splitting something that really is one connected object changes nothing and says so.
Can I undo a merge or a split?
Yes — one button returns the sheet to everything that was found, and it only appears when there is something to undo. Changing a setting re-runs the sheet and clears your changes too, since the pieces are renumbered and a grouping made against the old numbers would no longer mean anything.
Can I do a batch of sheets at once?
Yes. Drop in as many as you like and they are worked through one after another, with the queue showing which one it is on. They are done in turn rather than all together on purpose: this is arithmetic with no waiting in it, so running them at once would not finish any sooner — it would only hold several decoded sheets in memory at the same time. Changing a setting re-runs the whole batch, after a short pause so a slider drag does not set it going on every step.
What are the pieces named?
After the sheet they came from, numbered in reading order — left to right, top to bottom, with rows worked out from the items’ own heights so a row of slightly misaligned things still reads as one row. Every file is padded to the same number of digits so they sort correctly everywhere, and a batch of several sheets is foldered by sheet so two of them numbering from one cannot be muddled.
A heap of sawdust lost the flecks around it. Why?
It should not any more. A heap is not one shape — it is a mound with specks thrown around it, and those specks come out as their own tiny components. They used to fall under the minimum size and be discarded, which took the scatter out of the picture and left the crop cutting through specks that plainly belonged to it. A fragment is now given to the piece it is lying on, chosen by which piece its pixels are actually nearest to rather than whose rectangle it happens to fall inside.
Do the soft edges survive?
Yes, and keeping them is a deliberate part of how ownership is worked out. The faint half-transparent fringe around an object is too weak to define where that object starts, so it is assigned to whichever item is nearest rather than being thrown away. Discarding it would trade every smooth edge on the sheet for a jagged one.
Why PNG?
Because every piece cut from a sheet carries transparency, and PNG is the format that keeps it everywhere a file might be opened. A piece pulled out of a sheet is usually on its way somewhere else to be used, which is the wrong moment to flatten it onto a background.
Does it cost anything or need an account?
Neither. There is no server doing the work, so there is nothing to meter — the hundredth sheet costs exactly what the first did.