One image, a dozen shapes
Publishing the same picture across social platforms is not one task repeated; it is a series of different crops that happen to share a source. A square feed post, a full-screen vertical Story, a wide video thumbnail and a letterbox profile banner are four genuinely different compositions. Producing them by opening an editor, cropping by eye, exporting, renaming and repeating is where an afternoon goes.
This tool inverts that. You choose the destinations once, and it crops your image separately for each target ratio, resizes each crop to that platform's exact pixel dimensions, and returns the whole set as a single ZIP with filenames that say which is which. The image is decoded once and the work happens in parallel on your own machine.
The reference table above lists every size it exports, grouped by platform. It is generated from the tool's own preset definitions rather than transcribed into prose, which means it cannot drift out of step with what the export actually produces. When a preset changes, the table changes with it.
Aspect ratio matters more than pixel count
The numbers platforms publish look like a demand for a specific resolution, but the part that is genuinely load-bearing is the ratio. Upload an image with the right shape at slightly the wrong size and the platform scales it, which costs you a negligible amount of sharpness. Upload one with the wrong shape and the platform crops it, using an algorithm that does not know which part of your image mattered.
That is where the visible failures come from: the headline sliced in half, the logo cropped out of the corner, the subject's forehead removed by a square thumbnail. The platform is not being careless; it is being given an image it cannot display and taking the only action available. Supplying the correct ratio is what takes that decision back.
Resolution still matters, in one direction. Every platform re-encodes what you upload, and the quality of that re-encode depends on the size of the input. Uploading at the recommended dimensions rather than half of them gives the platform's compressor room to work, which is why a full-size upload usually looks better after processing than a smaller one that looked identical before.
Subject-aware cropping, and why centre-cropping fails
Almost every automatic resizer crops from the centre, because it is the only assumption you can make without looking at the image. It is also wrong most of the time. Photographs are usually composed with the subject off-centre — a face on a third, a product to one side, a horizon low in the frame — and a centre crop to 9:16 will happily deliver a picture of somebody's shoulder.
This tool looks first. It attempts face detection and, when there is no face, falls back to a saliency estimate: a fast pass that finds the region of highest local detail, which for most photographs is the subject rather than the sky or the wall. The crop window for each target ratio is then placed around that point and clamped so it stays inside the image.
It is an estimate, not a judgement, and there are cases where it will choose differently from you — an image with several equally salient subjects, or one where the meaning is in a small detail rather than the busiest area. The framing control exists for exactly that, and the extreme ratios are where it earns its place. A 3:1 header keeps roughly a third of a portrait photograph's height, so the difference between a good and a bad vertical position is the difference between a face and a chin.
Designing for the surfaces, not just the sizes
Correct dimensions are necessary and not sufficient, because several of these placements have things drawn on top of them. Full-screen vertical formats — Stories, Reels, Shorts, TikTok posts — carry interface elements at both ends: profile information across the top, and reply fields, captions and action buttons across the bottom. Roughly the outer 250 pixels at each end of a 1920-pixel-tall frame are contested. Text and faces belong in the middle band.
Profile and page banners are harder still, because they are cropped differently on different devices. A wide desktop banner is usually reduced to a much narrower strip on mobile, taken from the centre. Anything important placed near the left or right edge of a banner is effectively desktop-only. Keep the essential content central and treat the outer thirds as decoration.
Then there is the size the image will actually be seen at. A YouTube thumbnail is designed at 1280 pixels wide and viewed at around 210. A feed post is designed at 1080 and viewed on a phone at perhaps 400. Anything that depends on fine detail or small type disappears at those sizes. The reliable test is to shrink your design to its real display size and look at it there — if it still reads, it works.
Everything runs on your device
The whole pipeline is local. Your image is decoded in the browser, face detection and saliency run on your machine, each crop is resampled with Lanczos in a WebAssembly codec worker, and the ZIP is assembled in memory before it is handed to your downloads folder. No byte of your photograph is transmitted, and there is nothing on our side to log, cache or delete.
For anyone handling client work, unreleased campaign assets, or photographs of people who did not consent to a third-party upload, that is the difference between a usable tool and one that has to be avoided. It also means the tool works offline once the page has loaded, and that there is no upload limit dictating what you can open.
Lanczos is worth a note, because it is the reason downscaled exports hold their detail. It is a higher-quality resampling filter than the bilinear scaling browsers apply by default, and the difference is most visible exactly where social images live: fine text, hard edges and detailed textures reduced to a fraction of their original size. It costs a little more computation, which is affordable when the computation is happening on your own machine rather than on a server we would have to pay for.