Batch Image Editor

Most picture chores are two or three jobs in a row. A shop photo needs cropping square, shrinking and saving small. A set of proofs needs shrinking and a name across the middle. A scan needs turning the right way up and squeezing under an upload limit. This page does all of those in one pass, on one photo or on hundreds. Below: the order the steps run in and why, every control in every step, how the file is written out, worked jobs that combine several steps, and the mistakes that catch people most often.

One page, one pass, one save

This page replaced nine separate tools: the Image Resizer, Image Compressor, Image Converter, Image Cropper, Rotate and Flip, Watermark Adder, Circle Crop, Photo Effects and Add Border. They were always the same machine underneath. You dropped pictures in, set a few options, and the same change was made to every picture. The only real difference was the job in the middle. So now the jobs sit side by side in the settings rail as steps, and you switch on the ones you want.

The practical gain is bigger than saving a few clicks. Before, cropping, shrinking and watermarking a photo meant three trips through three pages, and a JPG was compressed again at the end of each trip. Every one of those saves threw a little more detail away. Here each picture is opened once at full size, every step you switched on is applied in memory, and the file is written once at the end. One save, one round of compression, however many steps you used.

The old addresses still work. Someone who arrives from a link to the compressor lands here with the Save as settings open, and someone looking for the cropper finds Crop open and switched on.

The order the steps run in, and why it is fixed

The steps always run in the same order, top to bottom as they appear in the rail: rotate and flip, crop, resize, effects, watermark, circle crop, border. Then the file is saved in the format you chose. You cannot shuffle them, and that is on purpose, because each position was picked so the steps do not fight each other.

Straightening comes first because it trims the edges anyway. Cropping after it means you frame what is actually left, rather than guessing. Cropping comes before resizing because a crop chooses what stays and a resize chooses how many pixels it is made of, and that is the sensible order for any job: decide the frame, then the size.

Effects come after resizing so their radii are measured on the finished size. A blur, a grain or a sharpening radius means the same on the saved file as it does on the preview. The watermark comes after effects so your name is never glitched, sketched or pixelated along with the photo. The circle crop and the border come last because they shape the outside of the picture, and the border is the very last thing so the frame surrounds everything else.

One rule of thumb covers most jobs. Crop for what is in the frame, resize for how many pixels, compress for how many bytes. Most jobs that feel complicated are really two of those in a row, and here they happen in a single run.

Two meanings of the word big

A picture has two sizes, and they are easy to muddle. One is its measurement in pixels, say 4032 across and 3024 down. The other is how much room the file takes on disk, say 4.8 megabytes. Cutting the first usually cuts the second, but they are separate things and separate controls look after them.

So read the requirement before touching anything. A note saying maximum 1600 pixels wide is about measurement, and the Resize step answers it exactly. A note saying under 2 MB is about weight, and that is the Target size setting under Save as. Many forms ask for both, and you can switch on both. Shrinking the measurement first makes a weight target far easier to reach without the picture turning to mush, because halving the width of a photo quarters the number of pixels in it.

Phone cameras save far more picture than most jobs need. A single frame can be four thousand pixels across and several megabytes heavy. That is wonderful for printing and wasteful for a web page, an email or a council form with a 200 KB cap. The weight comes out, the picture stays.

Getting pictures in

The drop area takes files in four ways. Drag them over from a folder window, click to open the file picker, paste a picture straight from the clipboard, or press Choose a folder to take a whole directory, sub-folders included. Anything in a folder that is not a picture is simply left out, and a short message says how many usable files were taken. There is no cap on how many you add.

It reads JPG, PNG, WebP, AVIF, GIF, BMP, ICO and SVG. The HEIC photos iPhones save open wherever the browser itself can read them; Safari can, most others cannot yet. A file that cannot be read is taken off the list with a message naming it, so one bad file never stalls a batch.

Every file appears in the list the moment you drop it, with a spinner while its thumbnail is made, and a bar at the top of the list counts through them. The Save button waits until reading has finished. Add images tops the list up without disturbing what is already there, the bin icon on a row removes that one picture, and Clear all empties the workspace. None of it touches your disk. The page works on copies held in memory, so every original survives every experiment.

Switches, the preview and the list

Each step has a switch beside its heading, and every switch starts off. Touching any control in a step switches that step on, so you rarely need the switch itself; it is there to turn a step off again without losing its settings. A line in the bar above the rail lists the steps that are on. With none on, the pictures are only saved again in the format chosen under Save as, which is exactly what the old compressor and converter did.

The big preview shows one picture at a time, the first one you added until you click another row in the list. Result shows it with every step applied, and every change you make redraws it: effects, watermark, circle, border, and the Save as settings too, so a low quality setting shows its blocks and a JPG shows the colour that will fill its see-through parts before anything is saved. Original shows the picture untouched, and Adjust crop, which appears while Crop is on, shows the whole picture with the crop box on top. Changing any setting outside Crop takes the preview back to Result by itself.

The picture is drawn at its real size when it fits and scaled down when it does not, and the line under it says which, as in shown at 47% or shown at actual size. So a resize shows as a smaller picture, and Original beside Result shows how much smaller. The preview is worked from a copy of at most 2400 pixels so it keeps up as you drag, by the same code that writes the file. Two things are estimates rather than exact: Target size, whose real search runs on the full picture when you save, and JPG compression, which the saved file also tries with MozJPEG. AVIF compression is not shown, because it takes seconds per picture. A step that cannot be drawn yet, such as a logo watermark with no logo chosen, is left out of the preview and the line under it says why.

Each row in the list shows the name, the pixel size and the weight on disk. When a step will change the size, the row shows the planned size with an arrow before you save, so you can see the cost of a decision before committing to it. Small tags flag a picture with its own turn or its own crop, a vector drawing, or a GIF.

Your settings are remembered on your own machine for next time. The switches are not, so arriving next week to shrink a photo will not surprise you with last week's circle crop.

Every image the same, or each one its own

Most steps apply one setup to the whole batch. That is usually why you batch at all: one look, one size and one frame are what make a set of photos look deliberate rather than accumulated. If some photos need a different watermark or a different border colour, run the page twice.

Two steps are different, because pictures really do differ there: rotation and cropping. Each has a tick box called The same for every image. Left ticked, whatever you do to the picture in the preview is done to all of them, which suits a stack of scans that all came out sideways. Untick it and each picture keeps its own: pick one in the list, set it, then pick the next. Rows that have their own setting carry a small own turn or own crop tag, so you can see at a glance which ones you have done.

Ticking the box again copies the picture in the preview to the rest. For a crop, the box is copied in proportion, keeping its centre and its share of the picture, so it behaves sensibly when the photos are different sizes. Click through the list to check each one and nudge any stragglers by hand.

Rotate and flip: quarter turns, mirrors and straightening

Three jobs share this step, and only one of them costs anything. The Left 90°, Right 90° and 180° tiles turn the picture by quarter turns. Press them as often as you like; two rights make a 180, and four bring you back. A quarter turn simply moves each pixel to a new address, unchanged in colour, so the result is exactly as good as the original. A grey note under the tiles says so while the fine angle is at zero, and steps aside once you tilt, because the promise no longer holds.

The Horizontal and Vertical tiles under Mirror swap left with right, or top with bottom. People often reach for a mirror when they meant a 180, and writing in the shot shows the difference. Turned 180 degrees, text is upside down but still correct; tip your head and you can read it. Mirrored, it is the right way up but reversed, and no head tipping fixes that. The Mirrored readout shows H, V or none, so you always know the state the picture is in, and Reset puts it back as it arrived.

Fine angle is the straightening control: a slider from minus 45 to plus 45 degrees in tenths, with a box beside it for typing. Almost all real straightening lives between minus 3 and plus 3, so treat it as a nudge. The readout beside the label shows the total turn, quarter turns and fine angle together, so a right turn plus a 1.5 degree tilt reads 91.5°. Tilting is the one operation that recalculates pixels from a blend of their neighbours. The loss is tiny but real, and it repeats every time you save and reopen, so straighten once, from the original.

Tilt a rectangle inside its own frame and four empty wedges appear at the corners, because the camera never recorded anything there. Empty corners decides what happens to them, and only when the angle is not a plain quarter turn. Crop them away, the recommended choice, cuts inwards to the largest rectangle that holds only real picture, so the result is a little smaller and has no gaps. Expand canvas, leave transparent grows the frame to hold the whole tilted picture and leaves the wedges see-through. Expand canvas, fill with colour paints them with a Corner colour you pick, white to begin with. Keep original size holds the frame as it was, gaps and all.

Saving here also fixes a common puzzle: a photo that looks upright in one program and sideways in another. The orientation is written into the pixels themselves, so every program agrees from then on.

Crop: the box, the numbers and the ratio lock

A crop throws pixels away, and that is the whole operation. The ones you keep are not stretched or recalculated, which is why a crop never makes a photo worse on its own. People crop for two reasons. One is composition: the lamp post at the edge goes, the subject moves off centre, the empty sky shrinks. The other is shape: a slot on a page wants a wide strip or a perfect square, and the photo is neither.

Switch Crop on and the preview changes to Adjust crop, with a box over the whole picture. Press Result, fold the Crop panel away or change any other setting to see the cropped result. Drag inside the box to slide it. Drag any of the eight handles to resize it. A tag on the box reports its size the whole time, and that figure is in the pixels of your picture, not of your screen. For exact work, type into the X, Y, W and H boxes instead: X and Y are the distance from the left and top edges, W and H the size you keep. The Crop gives readout confirms what comes out, and Reset crop puts the box back where it started.

The Aspect ratio list is the shape control. Freeform lets the box go anywhere. Pick a ratio and the box keeps that proportion whichever handle you pull: square, the usual photo and screen shapes, upright shapes for phones, A4 in both directions, two passport shapes, and ready-made sizes for common social posts, covers, banners and video thumbnails. Choosing a ratio re-centres the box on every picture at the new shape.

The social presets carry a pixel size as well as a shape, and they show one more tick box, Scale output to exactly, followed by that size. Ticked, the cropped area is scaled to that exact size. Unticked, you keep the raw cropped pixels at whatever size they happen to be. One caution on the passport shapes: they get the proportions right and nothing more. Official photos also have rules about head size and background that no ratio lock can check, which is what Passport Photo Maker is for.

The crop box works on the picture after rotation, so a photo you turned sideways is cropped the way you now see it. Since the source file never changes, it costs nothing to experiment: cut a bold square from a wide scene, save it, then try a letterbox strip from the same shot.

Resize: Fit, Exact and percent

Resize by is a three-way switch and it decides what the numbers mean. Fit treats them as a box. Every picture is shrunk until it sits inside, with its shape untouched and nothing trimmed. Feed it a mix of upright and sideways photos with a 1600 box and each one comes out with its longest side at 1600. This is what most people want most of the time. Exact aims for the precise numbers you type. The % button scales every picture by a share of its own size, so a 4000 pixel photo and an 800 pixel one both shrink by the same proportion and end up at very different sizes. A grey line under the switch spells out the current choice.

Width and Height take any value up to 30000 pixels. The first picture you add fills them with its own size, so switching Resize on starts from something real. Clear one and it shows auto: leave one blank and it is worked out from the other, keeping the shape. Typing 1200 into Width alone is a perfectly good instruction. In % mode the two boxes give way to Scale to, from 1 to 500 per cent, starting at 50.

Keep aspect ratio appears only in Exact mode, and it is the setting that catches nearly everybody. Ticked, typing a width fills in the matching height for the first picture's shape. When both numbers are already set, as a preset sets them, each picture is scaled to fit within them without distortion, which means it may not land on both. Choose the square post preset for a 3:2 photo and it comes out 1080 by 720, not 1080 square. Untick the box and the picture really is forced to 1080 by 1080, stretched to get there, and everybody in it gets slightly fatter or thinner. If you want a square without stretching, crop to Square first and then resize.

The Preset list fills the fields for you: web widths from a 300 pixel thumbnail through 800, 1280 and full high definition at 1920 to 4K at 3840, all in Fit mode, plus exact sizes for a square post, an upright story, a video thumbnail, a cover, a banner and a 512 pixel avatar. Typing your own numbers returns the list to Custom.

Never enlarge smaller images is ticked when the page opens. Leave it that way. A picture already smaller than your target passes through at its own size instead of being blown up. Enlarging cannot add detail that was never recorded, so all it makes is a soft picture in a bigger file, and a mixed folder almost always hides one small image you had forgotten about. Shrinking, by contrast, looks clean: it is done in steps rather than one jump, which keeps edges and fine lines from going jagged.

Effects: adjustments, sharpening, glitch and duotone

The Effects step holds nine looks, chosen from Which effect, and the panel shows only the controls that look uses. One effect runs per pass. Every effect is built so one or two controls do the real work, and pushing every slider at once is what makes a photo look processed rather than styled.

Adjustments is the plain one: Brightness, Contrast, Saturation, Sepia, Greyscale, Hue rotate, Blur and Invert colours. It is where to start when a set of photos do not match, such as shots from two cameras. Brightness and contrast are different levers. Brightness moves every value up or down together, lifting shadows and highlights alike, which can flatten a picture. Contrast pushes values away from the middle, so darks darken and lights lighten. A washed-out photo usually wants contrast. Sepia and Greyscale work in part: at 40 per cent you get a hint rather than a conversion. Hue rotate turns the whole colour wheel, blunt on a photo and useful on a graphic.

Preset look sets all the sliders to a recipe: Vivid, Soft, Black and white, Vintage, Cool, Warm, Dramatic or Faded, or None to go back to the original. It is a starting point you can argue with. Move any slider and the menu honestly reports Custom, and set the sliders back and it recognises the preset again.

Sharpen and blur cannot add detail a camera never recorded. Sharpening exaggerates the edges already there, which the eye reads as crispness. The method is unsharp masking: the picture is blurred, the blur is subtracted from the original to find the edges, and that difference is added back. Strength is how much goes back in. Radius is how far the blur reached, which decides what counts as an edge. A small radius with high strength sharpens fine texture and makes noise worse; a large radius with modest strength adds punch to big shapes without touching grain. Set What to do to Blur and the same radius softens instead. A pale outline along a hard edge, a halo, means the strength is too high for the radius, and sharpening a JPG saved at low quality sharpens its compression blocks too.

Glitch is five separate faults. Colour split, pushing the red channel one way and the blue the other, is the one to set first and often the only one needed. Sliced bands shove strips sideways and read as a transmission fault. Corrupted blocks swap, invert or crush small rectangles and read as data corruption. Scanlines and Noise sit on top as surface texture, which is what makes the result look photographed off a screen. For the classic tape look, use colour split, scanlines and a little noise, and keep bands and blocks low. The randomness is fixed, so the preview and the saved file match and the same settings give the same damage. Press Re-roll the randomness for a different arrangement; it lasts until you leave the page.

Duotone throws the original colours away and keeps only how bright each pixel was, then maps that onto colours you choose: Shadows for the dark end, Highlights for the bright end, and everything between blended. Set Tones to three and a Midtones colour bends the ramp, which separates a considered result from an obvious one, or pick one of the preset pairs. A flat, evenly lit snapshot only uses the middle of the ramp and comes out muddy, which is what the Contrast slider rescues. Blend with the original brings some true colour back; at 20 or 30 per cent the photo looks graded rather than recoloured.

Effects: film, sketch and halftone

Film gives away its age through several separate details, and the step offers them one by one. Lift the blacks does the most work quietly: film is never truly black, so raising the darkest tones a little is more convincing than any amount of grain. Halation is the one most filters forget, light bleeding around a bright edge, which is why a window glows in a film photo and simply clips in a digital one. Grain has an amount and a size, and size matters more: fine grain reads as digital noise, coarser grain at a lower amount reads as film. Grain is scattered afresh every time, so the saved file carries the same amount and size of grain as the preview but not the same specks. Vignette darkens the corners as a lens does, Light leak adds a streak of colour from the edge, Warmth shifts towards amber or blue, and Black and white drops colour for mono stock. Starting point moves the sliders to imitate a type of film, warm portrait, saturated slide, high contrast mono, faded or cross processed, and leaves them all in view.

Sketch finds edges and draws them, then decides what to do with everything that is not an edge. The five styles differ in that decision: Pencil sketch keeps soft shading, Ink drawing drops it for hard black and white, and Clean outlines, Charcoal and Cross hatched sit between. Line strength sets how much ink goes down: in the edge styles it decides how faint an edge may be and still be drawn. Detail appears only for Pencil and Charcoal, the two styles that shade, and sets how wide the shading reaches: low values keep fine lines, high values build broader, heavier strokes. Contrast adjusts the photo before anything is drawn. Paper sets the background colour, Paper texture adds grain to it, and Keep some of the original colour lets colour survive. The photo matters more than any slider: a subject clearly apart from its background works, and a busy scene at one distance turns into a tangle.

Halftone is how presses print photos. They cannot print grey, only ink or no ink, so the picture becomes dots that grow where it is darker, and the eye blends them at normal distance. Dot or cell size decides whether the result reads as a technique or a mistake. Screen angle matters more than it sounds: dots on a square grid look mechanical, which is why printers tilt them, and 45 degrees is the classic choice. Contrast shapes the picture before it is screened, and a flat photo screens into a flat grey field of identical dots. Ink colour and Paper colour are real colour pickers, so dark blue on warm cream gives a two-colour poster, and Invert swaps them.

The same menu also offers Newsprint, four colour, the dithering modes and Characters instead of dots. Dithering takes a different route to the same problem, scattering same-sized pixels rather than sizing dots, which suits pixel art and small sizes. Levels sets how many tones it keeps, and the dot size control steps aside for it, since there are no dots to size. The panel only ever shows the controls the chosen mode actually uses.

Effects: pixelate and tilt-shift

Pixelate replaces each block of pixels with their average colour. Used on a face or a number plate, it removes the detail that identified them, and how much it removes depends entirely on Block size, which is the safety dial rather than a style choice. Apply to offers The whole image, A region I set below, placed with the four region sliders, and Everything except the middle, which keeps that same rectangle sharp and pixelates the rest, the way to keep one person clear in a crowded room. Blocks can be square, round dots, hexagons or triangles, and Soften the block edges takes the hard steps off.

Take care when it is for privacy. Large blocks over a small face leave nothing; small blocks over a large one can leave enough structure to recover. Pixelated text is weakest of all, because a short, predictable string can sometimes be rebuilt by pixelating guesses and comparing. Round shapes leave gaps between blocks, so the area gets a solid base coat first; a gap would be a hole you could see through. And no visual effect touches the location, camera and time stored inside a file. Saving here drops those, as described under Save as, and the EXIF Viewer and Remover shows exactly what a file carries.

Tilt-shift makes a real scene look like a model. Up close, only a narrow band of anything is sharp, so the brain reads a thin band of focus as a small subject seen from near by. Focus position sets where the sharp band sits and Sharp band height how thick it is. Blur ramp sets how quickly sharpness falls away, and it separates convincing from obviously edited: a sudden edge reads as a mistake, a gradual one as a lens. Blur strength sets how soft the far parts get. Colour boost is part of the trick, because model villages are painted brighter than the real world. Run the band vertically instead suits a street shot along its length. The photo matters most: it wants a scene seen from above at a distance, with small, separate subjects. A street at eye level never works, because the geometry contradicts the illusion.

Watermark: a name in the corner, or a lattice across the middle

A watermark does one of two jobs, and the settings depend on which. Attribution puts your name on work so viewers know whose it is. It should be small and quiet, because the picture is the point, and a corner mark at low opacity does it perfectly. Protection makes a copy hard to use without permission: proofs before payment, gallery samples, competition entries. There the mark must be in the way, across the middle, where it cannot be cropped off. A tiny corner logo protects nothing, and a heavy lattice over a portfolio piece just annoys people.

With Type on Text, the Text box holds up to 120 characters and starts with a copyright line to replace. Font offers Sans-serif, Serif, Monospace and Cursive; sans-serif stays readable when small and faint, and cursive is the hardest to read at low opacity. Bold starts on and earns its place, because thin letters at 30 per cent vanish over a busy photo. Colour starts at white, which suits most photos; dark grey often reads better on pale ones. Drop shadow, also on, paints a soft dark edge behind the letters. It is the setting people turn off before wondering where their white text went against a bright sky.

Switch Type to Logo image and press Choose logo image. A small preview appears with the file name and size, so you can confirm you picked the right one. Use a PNG with a truly transparent background; a JPG logo brings a white rectangle with it and looks like a sticker. The logo is scaled by its longest side, so a wide logo and a square badge at the same size look equally prominent, and a white logo reads over almost any photo.

Size runs from 2 to 60 and is a percentage of each picture's shorter side, not pixels, which is what lets one setting suit a mixed batch. On a photo with a 4000 pixel short side, 3 gives 120 pixel letters; on a 1000 pixel export it gives 30, and both look the same on screen. For text, Size sets the letter height, never the line width, so a long studio name at 20 runs off both edges. Opacity runs from 5 to 100 and starts at 45; between 30 and 50 is the usual balance. Angle tilts the mark from minus 90 to plus 90 degrees around its own centre, and a slant of 30 or 45 is the classic proof look.

Tile across the whole image switches between the two jobs. Unticked, Position offers nine places from Top left to Bottom right, where it starts, and Margin sets the gap from the edges, again as a percentage of the short side, from 0 to 15. Ticked, those give way to Gap between tiles, from 1 to 40, the space between one copy and the next. The whole grid is turned around the middle of the picture and drawn out past the diagonal, so no clean corner is left to crop to.

A watermark is a deterrent, not a lock. A corner mark can be cropped away in seconds. A tiled one is much harder, because no undamaged copy of the picture is left to work from, but harder is not impossible. And nothing of ours is ever added: no badge, no hidden signature, no paid tier that removes something we put there. The only mark on the saved picture is the one you set.

Circle crop: round pictures that stay round anywhere

Plenty of apps show a photo inside a round window, while the file itself stays a rectangle. Move it into a document, a slide or an email footer and the corners come back. This step really removes them: what falls outside the shape is stored as empty, so the picture sits on a coloured panel, a dark slide or a printed page with no box behind it. If the picture only ever goes to an app that draws its own circle, you can skip this step and give the app the plain photo.

Before the shape is cut, a square is taken from the picture, always as wide as its shorter side. A photo 3024 wide and 4032 tall gives a 3024 square, centred from side to side. Up and down is your choice: Crop from Centre takes the middle, Top pushes the square against the top edge, where a face usually is in a portrait, and Bottom suits a subject that sits low. Zoom, from 100 to 250 per cent, tightens the square: at 130 it is about three quarters as wide, so the subject fills more of the shape. Zoom only cuts closer; it cannot reach past the edges or invent detail.

Shape offers Circle, Rounded square and Hexagon, which is flat at the sides and pointed at the top and bottom. Rounded square shows Corner radius, from 2 to 50 per cent of the finished picture, so it looks the same at any size: around 8 gives an app-icon look, and 50 makes the corners meet in a circle.

Output size is exact, from 32 to 4096 pixels square, starting at 512, whatever the size of the original. 512 covers nearly every use online; 1024 suits print or a sharp screen. The square is cut at full resolution and only then shrunk, in steps, so hair and glasses frames stay clean. Ring thickness starts at 0, no ring. Move it and a Ring colour well appears; the number is a percentage of the finished picture, so 4 on a 512 avatar draws a band about 20 pixels wide, just inside the edge, never clipped. Past about 10 the ring starts to crowd the face.

Background is the setting people misread. Solid colour paints inside the shape, underneath the picture, not around the outside. On an ordinary photo you will see no difference, because the photo covers the whole shape. It matters when the picture already has see-through parts, like a PNG logo, and then the colour fills those gaps. The outside of the shape stays clear either way, so with Save as on Same as original a round photo is saved as PNG, the common format that remembers which parts are empty.

One honest note for team pages. The step applies the same rule to every photo, not the same face position. A colleague shot close up and another shot from across the room will still look different. If they must match, crop the originals to a similar framing first; with Crop and Circle crop both on, that happens in the same pass.

Border: frames, mounts and rounded corners

A frame buys breathing room. A photo set straight against text or other photos bleeds into its surroundings, and a margin of clean colour restores the edge. Prints gain a white edge to hold, product shots gain consistency across a shop page, and a wide photo gains enough padding to fit a square slot without anything trimmed. A framed image also reads as finished, the way a mounted print does.

Style offers three treatments. Solid frame is an even band on all four sides. Polaroid mount keeps three sides even and deepens the bottom, leaving room for a caption as on an instant print. Rounded corners softens the corners of both the frame and the photo, with a Corner radius slider for the curve. Even margins suit galleries and prints, the deep bottom suits nostalgia and captions, and rounded corners suit screenshots and anything meant to feel soft and modern.

Width runs from 1 to 25, and its unit is a percentage of each photo's short side. A folder holding an 800 pixel export and a 5000 pixel camera file shows why: a fixed 40 pixel band would swamp the small one and vanish on the big one, while 5 per cent gives 40 and 250 pixels, and both frames look the same share of the picture. Around 3 to 5 looks like a classic print margin; past 15 the frame becomes a statement.

Colour starts at white, which vanishes against most pages and flatters most photos; cream feels warmer and near black suits high contrast work. Thin accent line draws a slim keyline between photo and frame in an Accent colour you choose, and that one line often separates homemade from professional. Inner shadow under the photo adds a hint of depth, as if the print sat above the board. Both scale with the photo, so they hold up across a mixed batch.

A border is added around the picture, never cut out of it, so the output is always bigger. On a 4000 by 3000 photo, one per cent of the short side is 30 pixels, so a width of 5 adds 150 pixels on every side and the file comes out 4300 by 3300. A 1080 pixel square at the same setting gains 54 a side and becomes 1188 square. The accent line adds a hair more. The Polaroid bottom is 3.2 times the side band, so on that same photo the bottom gains 480 pixels, which is why a mount needs a lower width than a plain frame to feel balanced.

Rounded corners leave areas outside the curve, and what fills them depends on Save as. PNG and WebP keep them see-through, so the picture sits cleanly on any page. JPG cannot, so the corners are filled with the Background behind transparency colour. If you have ever seen a nicely rounded image wearing wedges in its corners on a website, that is the reason.

Save as: the format, the quality and a target size

Format offers Same as original, JPG, PNG, WebP and AVIF, and a line under it explains the one you chose. JPG is the safe choice for photos, read by everything ever made, but it cannot store see-through areas. PNG loses nothing and keeps transparency, and pays with files often several times larger on photos, so keep it for logos, screenshots and drawings. WebP is usually a quarter to a third smaller than JPG at matching quality and keeps transparency. AVIF squeezes hardest but is slower to write. If you have no strong reason, WebP for your own web pages and JPG for anything sent to a stranger will serve well.

Same as original keeps JPG, PNG, WebP and AVIF files in their own format. Types a browser cannot write, GIF, BMP, HEIC, SVG and icons, become PNG when the result has see-through parts and JPG when it has none. A JPG made round by Circle crop becomes PNG so its corners stay clear.

What this browser can create lists the four types with a tick or a cross. It is not guessed from a table of browser versions: the page asks your browser to write a tiny picture in each type and checks the result, and anything that fails is greyed out in the Format list, so a batch cannot fail half way. JPG, PNG and AVIF go through real encoders compiled to WebAssembly, because they beat what a browser does alone. JPG is encoded both by the browser and by MozJPEG, and the smaller file is kept, so it is never bigger than the browser would have made. PNG is recompressed by oxipng, which often saves a third or more without changing a single pixel. AVIF comes from libaom, since no browser writes it; it downloads once, about 3.4 MB, and takes a few seconds per picture. WebP is written by the browser itself.

Compress by has two buttons. Quality means you set the dial and accept whatever size comes out. The Quality slider runs from 20 to 100 and starts at 85. It tells the encoder how much fine detail it may drop. Between 75 and 85 you will struggle to tell the result from the original; below 60, skies and skin start to show blocks; above 90 the file grows fast for little visible gain. PNG has no such dial. What makes a PNG heavy is how many distinct colours it holds, so PNG colours, from 8 to 256, rebuilds the picture with fewer. At 256 every pixel is kept exactly. On a screenshot, logo or chart, 64 or 128 can halve the file with nothing visible lost; on a photo it causes banding. With Same as original, the panel shows only the sliders your pictures will use: Quality when some will be JPG or WebP, PNG colours when some will be PNG.

Target size means you state the size and the page finds the dial. Type the ceiling, 200 KB to begin with, and pick KB or MB. Each picture is written several times over, closing in on the best setting that still fits, which is why a target run takes a little longer. It works with JPG, WebP and PNG; AVIF uses the quality setting. Sometimes the number is out of reach: a 12 megapixel photo will not become 30 KB at any setting. That row is then marked missed target, you get the smallest file it managed, and a message names the two moves that really help: switch Resize on to make it smaller, or save as WebP.

Background behind transparency is the colour that fills see-through areas when the format cannot keep them, white to begin with. Set it to the page the picture is going on and the join will not show. If any result has see-through areas and you chose JPG, a warning says so before you save. Rasterise SVG at appears when a vector drawing is in the list. An SVG is a set of drawing instructions with no fixed pixel size, so this sets its width, from 16 to 10000 pixels, starting at 1024, with the height following the drawing's own shape. Drawing it fresh at 2048 and again at 512 gives two crisp files, where shrinking one into the other would soften it. A GIF notice explains that an animated GIF becomes one still frame, because none of these formats stores motion; to keep it moving, turn it into a video instead.

Writing a fresh file drops the hidden notes a camera attaches to a photo, including the camera model and the place it was taken. For pictures going public, that is a quiet privacy win.

What comes out, and what it is called

The main button counts the pictures waiting, as in Save 12 images. Press it and a bar tracks the run. Pictures are handled one after another rather than all at once, which is why a folder of a hundred large photos does not run a modest laptop out of memory. Whenever more than one picture is being saved, a Stop button sits beside it. It finishes the picture it is on, and everything already done stays available.

With one picture, the finished file goes straight to your downloads folder. With several, they arrive together as one ZIP. Every row also gains its own Save button, and Download all as ZIP stays in the bar for another copy. After a run the button reads Save again, for a fresh set once you change a setting. The heading over the list adds up the batch, before and after, with the change as a percentage, and each row shows its new size, format and weight beside the old.

Compare, on any row whose picture kept its shape, opens a panel with your original on the left and the saved file on the right, split by a divider you drag across. Look at the busiest part of the busiest picture, grass and leaves or a face, since that is where compression shows first. When a crop, a circle or a border changes the shape, that row has no Compare, because there is nothing to line up.

Names tell you what was done. One step keeps that job's own ending: a resize adds the new size, so harbour.jpg becomes harbour-1920x1280.jpg, and the others add -cropped, -rotated, -watermarked, -circle or -bordered, or the effect's own word, such as -glitch or -sketch. Several steps add -edited. With no steps on, the name stays and only the extension changes. Either way the new file sits beside the original instead of replacing it, which makes a second try completely safe.

Worked jobs that use several steps

Twenty-four product shots for a shop page. A designer sent them as PNGs, about 3 MB each, and the shop wants square, light pictures. Drop them all in. Choose Square 1:1 under Crop with The same for every image ticked, then click through and nudge any product the box cut short. Switch Resize to Fit with 1200 in both boxes. Add a Solid frame border at width 3 in white, which puts 36 pixels around each 1200 pixel square and makes them 1272 square. Set Save as to WebP with Quality 85, then press Save 24 images. One ZIP arrives with every file under a few hundred kilobytes and named -edited.

Forty-two proofs for a client. The sports day photos are 6000 by 4000 JPGs, and nothing should be usable as it stands. Switch Resize to Fit with 2000 in both boxes, so landscape shots come out 2000 by 1333 and upright ones 1333 by 2000. Set Watermark text to PROOF followed by the studio name, Size 3, Opacity 22, Angle minus 30, and tick Tile across the whole image with a gap of 10. Because the watermark runs after the resize and its size is a share of the picture, the lattice looks the same as it would at full size, just on a file that is far too small to print. Save as JPG at 80.

A stack of scanned letters. Some came out sideways, one upside down. Untick The same for every image under Rotate and flip, then pick each crooked scan and press the turn it needs; the list tags each one own turn. Switch on Crop with the same framing for all to trim the scanner lid. Set Save as to JPG and Target size to 500 KB for the upload form, then save the lot.

Round headshots for a team page. Switch on Circle crop, leave Shape on Circle, set Output size to 512, Crop from to Top and Zoom to about 130, which pulls the framing in from shoulders to face. Set Ring thickness to 4 and pick the site's colour. Leave Save as on Same as original, and every JPG comes out as a 512 by 512 PNG called name-circle.png with a clear surround. Resize is not needed here, because Output size already decides the final size.

A leaning seafront. One holiday photo, 4032 by 3024, where the horizon drops to the right. Drag Fine angle to minus 1.0, look, then type minus 1.4 and it sits level. Leave Empty corners on Crop them away, and the list shows the result at 3961 by 2928, a little smaller with no gaps. Save as JPG, and it arrives as a new file beside the original.

Where people trip up

Most problems come from a setting doing exactly what it says, just not what was expected. These are the ones that come up again and again.

• The file got bigger. A photo was saved as PNG. Choose JPG or WebP for photos and keep PNG for logos, screenshots and drawings.
• Pressing Save changed nothing but the format. No steps are on. The bar above the rail lists the steps that are.
• Resize complains it needs a width or a height. Both boxes are empty in Fit or Exact mode. Type at least one.
• The picture looks squashed. Exact mode with Keep aspect ratio unticked stretches it. Tick it, or crop to shape first and then resize.
• Still over the upload limit. That limit is about weight, so use Target size under Save as, and switch Resize on to help it.
• A second pass through a JPG did little and looked worse. There was little left to remove. Start again from the original and do it once, harder.
• The sky turned into flat blocks. Quality is too low for that photo. Raise it, and take the weight out with Resize instead.
• The PNG barely moved when you dragged Quality. PNG never listens to it. Lower PNG colours, or save photos as JPG or WebP.
• Writing in a screenshot looks smeared. Sharp edges are what JPG handles worst. Keep screenshots as PNG.
• White or dark wedges in the corners. The result had see-through parts and was saved as JPG. Choose PNG or WebP, or set the background colour to match the page.
• The corner watermark vanished on a round picture. The watermark runs before the circle crop, and a corner mark sits outside the circle. Place it at the Centre, or leave one of the two off.
• A border around a circle came out square. The border frames the whole picture, corners included. For an edge that follows the shape, use Ring thickness instead.
• White corners on a round picture after uploading it somewhere. That site flattened the see-through parts. Give it the plain photo and let it draw its own circle.
• The face in a circle sits off to one side. Only up and down can be chosen; sideways is always centred. Switch on Crop to frame it first.
• The crop box will not move further. It has hit the edge of the picture. Make the box smaller first.
• The corner setting under Rotate seems to do nothing. It only applies to tilted angles; a quarter turn leaves no gaps.
• Text runs off both edges of the picture. Watermark Size sets letter height, not line width. Shorten the text or lower the number.
• The preview says it needs text or a logo. The Text box is empty, or Type is on Logo image with no file chosen yet.
• A small photo came out soft. It was enlarged. Keep Never enlarge smaller images ticked and start again from the original.
• Wanting a different watermark or frame colour on different photos. One run shares one setup, so run the page twice.

When a different page is the better answer

This page reshapes, resizes and restyles whole pictures. A few jobs belong elsewhere. If your phone's HEIC files will not open here, HEIC to JPG is built for exactly that complaint. For official photos with rules about head size and background, use Passport Photo Maker. To put several framed photos on one sheet, use Collage Maker. To hide a face, an address or a number plate by hand, Blur and Redact lets you paint exactly where, and to point at something in a screenshot, Annotate a Screenshot adds arrows and labels.

To change one colour rather than the whole picture, Image Studio can select an area and shift its hue, and to take a background out, Photo AI Studio removes or replaces it on your own machine. A building that leans because you pointed the camera up is a perspective problem, not a rotation one, and Perspective Correction fixes it. For layered work, masks and different effects on different parts of one picture, Image Studio is the page. And to see or strip the hidden data in a photo without re-saving it, use the EXIF Viewer and Remover.

One last habit is worth keeping whichever page you use. Always work from the best copy you own. Every trip through a format that trades detail for size takes a little more away, and going back never returns it. Squeeze the copies that travel, and keep the originals whole.

Nothing leaves your device

Every step on this page runs inside your browser, on your own processor. Pictures are read from your disk, worked on in memory and saved back to your downloads folder. None is uploaded, stored or logged anywhere, and there is no account, no daily limit and no watermark of ours. You can prove it: load the page, switch off your connection, and carry on. The only things fetched are the page's own encoders, from this site, the first time a format needs one.

That matters more here than it might seem, because the pictures people bring are often private: family photos, identity pictures, unreleased client work, screenshots with details in the margins, and pictures of other people's children. Speed depends on your machine rather than a queue or your connection, so a folder that would take minutes to push to a server finishes here in the time it takes to process. Your settings are remembered on your own machine, and nothing else is kept once the tab closes.