Image Tools

What the image tools are for

Most visits start with a small, dull job. A photo is too big for a form, so Image Resizer brings it down. A page loads slowly, so Image Compressor cuts the file size. A site wants a PNG and you have a JPEG, so Image Converter swaps the format. Image Cropper and Circle Crop trim a picture to the shape a profile or a slide needs.

Other tools go further. Image Studio is a layered editor for real retouching. Collage Maker puts several photos on one sheet. Screenshot Beautifier makes a plain screen grab tidy enough to share. Colour Picker & Palette helps you pin down an exact colour and build a set around it. Red-Eye Remover, Pixelate an Image and Film Effects each change one thing and nothing else.

What changes when the work happens in the browser

Image files are large, and that is the point. Sending a big photo to a server and waiting for it to come back often takes longer than the edit itself. Here the file never moves. Your browser already knows how to decode and draw pictures, so a resize or a crop happens at once, on your own machine. There is no queue and no size cap.

The honest limit is your own device. A very large scan, or a batch of many photos at once, uses real memory, and a modest phone may struggle where a laptop will not. The browser can also only open formats it already understands. If it cannot decode a file, no tool here can either. Closing other tabs helps more than anything else.

Before you start

Order matters. Crop first, then resize, then compress. Each step throws away pixels, and you want the last step working on the picture you are actually keeping. Keep the original file somewhere safe. A JPEG is compressed again every time it is saved, so a photo that has been through several rounds looks a little worse each time.

Choose by the kind of picture. Photographs suit JPEG. Flat colour, logos and line art stay sharper as PNG, and a true vector belongs in SVG, which SVG Optimiser & Converter can slim down. Enlarging cannot invent detail that was never captured, so start from the largest version you have. If the image is going in front of an audience, Colour Blindness Simulator shows how it reads to people who see colour differently.

Three encoders your browser does not ship

Browsers read more picture formats than they write. Yours will open an AVIF file without complaint, but no browser ships an AVIF writer behind the canvas. Ask one for an AVIF and it quietly hands back a PNG instead of an error. That is why Image Converter checks the type of the file that came back, rather than trusting the type it asked for.

Where the browser cannot write a format, the converter uses a real encoder built to WebAssembly, taken from the jSquash build of the Squoosh codecs. AVIF comes from libavif and libaom and weighs 3.4 MB. MozJPEG is 250 KB. Oxipng is 165 KB. All three are files on this site, and each one is fetched the first time it is needed, never when the page opens. Save a WebP and you download none of them. The code comes to your picture; the picture still goes nowhere.

Two of those are not new formats at all, just better encoders for old ones. For a JPEG both encoders run, your browser's and MozJPEG's, and you get whichever file came out smaller. The quality number means different things to the two of them, so this is settled by measuring rather than by believing. PNG is not re-encoded: oxipng rewrites the packing and every pixel stays as it was. AVIF is the one with a real cost. AV1 encoding takes roughly two seconds per megapixel on a mid-range laptop, so the tool says so before you start instead of showing a bar that looks stuck.

The same gap runs the other way round. HEIC to JPG carries no HEIC decoder of its own. It leans on the one your browser may or may not have. Safari reads HEIC, and recent Chrome and Edge do on some systems. There is no way to ask in advance, so the tool tries a small decode on your first file and tells you plainly if the answer is no. AVIF can be written here because libaom carries a patent grant from the Alliance for Open Media. HEIC has no such grant, which is why no HEIC codec is bundled.

A canvas holds pixels and nothing else

Nearly every tool here works the same way. The browser decodes your file into pixels on a canvas, the tool changes those pixels, and something writes a new file at the end. Thirty of the image scripts ask the same shared helper for that canvas. Once you know this, a lot of the behaviour follows from it.

A canvas carries pixels, and nothing rides along with them. The camera model, the lens, the date and the GPS position are all left behind at the decode. A photo that has been through Image Resizer, Image Cropper, Rotate & Flip or Watermark Adder comes out with no metadata on it at all. That is welcome if you were about to post it in public. It is a nuisance if you wanted the date kept.

Two tools skip the canvas for exactly that reason. EXIF Viewer & Remover reads the JPEG APP1 segment, and the tables inside it, straight from the bytes of the file. When it strips the data it rewrites the list of segments and leaves the compressed picture untouched, so the photo is pixel for pixel the same and only the tags are gone. Most tools that offer to remove EXIF redraw the photo and squash it again without saying so.

Shutter Count reads bytes too, because a shutter count is not an EXIF field. It sits in the maker's own private block, and every brand invents its own layout there. Nikon writes the true count at tag 0x00A7 and it can be relied on. Canon store no count at all on nearly every model, though they do store a file number that people mistake for one. Pentax scramble theirs against the capture date, and Sony hide theirs in a block that shifts between models. Neither is decoded here. This number is what people price a used camera by, and a figure that is right some of the time is worse than none.

The canvas sets a size ceiling as well. Nothing here turns a file away for being big, but the shared helper holds the working area to 16,777,216 pixels. That is a square of 4096 by 4096, or near enough 16.7 megapixels. Above that, iOS Safari fails without a word and any browser can run out of room. Anything larger is scaled to fit, and the scaling is done by halving over and over rather than in one jump, because a single big step down leaves visibly rough edges.

The look-alike pairs, and which one you want

Perspective Correction and Document Scanner run the same geometry. Both ask you to put four corners on something that ought to be rectangular, then pull that shape flat through the same shared warp code. What differs is what comes next. Perspective Correction stops there and takes the output size from the edge lengths of your own quad, which is what you want for a building or a painting. Document Scanner guesses the corners from the edges it finds, then cleans the result up: grey, or a real adaptive threshold that judges each pixel against its own neighbourhood. That last part is why the text stays readable when half the page is in shadow.

Image Compressor and Image Converter both make files smaller, and they are not swaps for each other. The converter changes format, and that is where the WebAssembly encoders live. The compressor keeps the format and hunts for a setting. Give it a target size and it searches the quality range in eight steps, halving the gap each time, which lands within a few per cent. For PNG it does something else again. A canvas can only write full-colour PNGs, so the compressor cuts the number of colours instead, and on screenshots, logos and flat art that alone takes 40 to 70 per cent off.

Three tools can blur. Photo Effects runs one effect over a whole picture, and over a whole batch of them. Blur & Redact hides boxes that you drag, held in the coordinates of the source image so the saved file matches the preview exactly. Annotate a Screenshot blurs as well, but it is really there for arrows, boxes and numbered steps. The LUT pair splits along a cleaner line. LUT Maker needs no picture at all, because a LUT is only colour in and colour out, so it builds the .cube file from the adjustments alone. LUT Applier is the one that wants your photo and a .cube file together.

Image Studio overlaps with half of this page, and the thing that settles it is memory. A 24 megapixel photo is 96 MB of raw pixels for every layer holding it, and a browser tab gives up somewhere past a gigabyte. So the studio caps what you bring in at 4096 pixels on the long side, and tells you when it has. The single-job tools carry no layer stack and keep your file at its own size. Crop, resize or watermark: use the small tool. Masks, blend modes, adjustment layers or a cut-out: that is what the studio is for. Its cut-out is the one thing on this page that downloads anything, about 90 MB of model weights, fetched the first time you ask for it and then reused. It runs on your machine like everything else.