Smart Crop Image Online — the Subject Stays in Frame
Reframe any photo for Instagram, YouTube, TikTok and more — without beheading the subject. This cropper measures where the eye actually goes (edges, contrast, colour, skin tones, and real faces where your browser detects them), frames the shot around it, and still lets you drag the result anywhere. Batch mode, an all-sizes pack and full-resolution exports included; your photos never leave your device.
How the smart part works — honestly: this is attention-based cropping, not a neural network. Your browser measures edge density, local contrast, colour saturation and skin-tone weighting on a small copy of the photo, then frames the window that captures the most of it, with the subject placed by the rule of thirds. Where your browser ships a built-in face detector (Chromium), faces are weighted in too — the panel above tells you which mode ran. Toggle Attention map to see exactly what the engine saw.
How to Use
- 1Add photos — drag & drop, browse, paste, or try the built-in sample scene whose subject sits off-centre.
- 2Pick a target shape (1:1, 4:5, 9:16, 16:9 and more, a custom ratio, or freeform). The smart frame lands on the subject automatically.
- 3Fine-tune if you like: drag the frame, pull a corner, zoom, set subject placement, or inspect the attention map and coverage scores.
- 4Download the crop — or the whole batch as a ZIP, or one photo in every social size at once.
Key Features
- ✓
Attention-based framing, explained honestly
The smart in this cropper is a measured attention map, not a buzzword. Your browser shrinks the photo to a small analysis copy and scores every region for edge density, local contrast, colour saturation and skin-tone weighting — the visual signals human eyes are drawn to. It then slides a window of your chosen aspect ratio across the photo and keeps the framing that captures the most attention, with a rule-of-thirds bonus for where the subject lands inside the frame. No neural network runs, and the page says so: the method is classical, instant, deterministic and fully inspectable.
- ✓
Face-aware when your browser allows it
Where the platform provides a built-in face detector — the FaceDetector API in Chromium browsers — detected faces are weighted into the attention map, so portraits and group shots anchor on the people instead of the wallpaper behind them. On browsers without the API, the skin-tone weighting in the attention map still pulls faces forward, just less precisely. The tool never guesses silently: a status line in the panel tells you, for every photo, whether a face was found, whether the detector ran and found none, or whether it is unavailable in your browser.
- ✓
Drag, resize and overrule it — free
Automatic framing is a starting point, not a lock-in. Drag the frame anywhere, pull its corners to zoom into the subject, use the slider for fine zoom steps, or tell the engine where the subject should sit — left third, centre or right third — and let it re-frame around your composition choice. Some competing tools put manual adjustment behind a premium tier; here the override is the point. A badge always shows whether the current frame came from the analysis or from your hand, and one click restores the smart suggestion.
- ✓
See what the engine saw
Two transparency features no rival bothers with. The attention-map overlay paints the actual scoring the engine computed — hot red where it thinks eyes will go, cool blue where they won't — so a surprising crop stops being a mystery. And the coverage readout measures it: the percentage of the photo's total attention inside your current frame, next to the percentage a plain centre crop would have captured. When the bars tie, your subject genuinely is in the middle; when they don't, you can see exactly what the centre crop would have thrown away.
- ✓
One photo, every platform — plus batches
Repurposing is the real job, so the tool is built around it. Ten target shapes cover the platforms people actually publish to: 1:1 posts and avatars, 4:5 Instagram portraits, 9:16 stories, Reels and TikTok, 16:9 YouTube, 3:2 photo frames, 2:3 pins, 4:3, 3:1 headers, 21:9 banners — plus any custom ratio and a freeform mode. The all-sizes pack re-frames one photo into six social shapes and zips them in a single download. And batch mode takes up to twenty photos at once, smart-crops each one individually (its subject, not a shared template), and delivers one ZIP.
- ✓
Full-resolution exports, exact sizes
Every crop is cut from your original pixels at full resolution — never from the small analysis copy, never from an upscaled preview. Export as JPG, PNG or WebP with a quality slider, at the crop's natural size, an exact pixel size (1080×1080 for Instagram, 1280×720 for a thumbnail), a set width, or a set longest edge. If you ask for more pixels than the crop contains, the tool caps at the real resolution and tells you — upscaling only happens when you explicitly switch it on. EXIF orientation is honoured on the way in, and the canvas re-encode leaves metadata behind on the way out.
Frequently Asked Questions
How does smart cropping decide where to crop?
Is this an AI cropper?
Does it detect faces?
Can I adjust the crop manually?
What sizes and formats can I export?
Can I crop many photos at once?
Will cropping reduce my photo quality?
Are my photos uploaded anywhere?
Why the centre crop keeps ruining good photos
Almost no photo is published in the shape it was shot. A landscape frame becomes a square post, a vertical story, a wide banner, a small round avatar — and every one of those reframings has to cut something away. The default cut, in nearly every editor and upload widget, is the centre: shave both sides equally and hope the subject was standing in the middle. Photographers rarely put it there. Composition rules — the rule of thirds above all — deliberately place subjects off-centre, faces near the top third, products to one side with breathing room for copy. A centre crop of a well-composed photo is therefore wrong in a specific, predictable way: it amputates exactly the part the photographer cared about, keeping the empty sky and losing the face at the edge. Smart cropping exists to make one decision well — which part of the frame survives — by looking at what is actually in the picture before choosing where the window goes.
How this engine decides, step by step
The method is the classical one behind the smartcrop.js library and the original Croppola service, reimplemented here and extended. First, the photo is downscaled to a small analysis copy — attention does not need megapixels, and the small copy is why the analysis is effectively instant. Each pixel of that copy earns attention from four signals: edge density (detail and texture attract eyes; flat sky and blank walls don't), local contrast (regions that differ from their surroundings stand out), colour saturation (vivid objects pull focus from muted backgrounds), and skin-tone weighting (humans look for humans, so skin-coloured regions score extra). A blur then consolidates the signals into regions rather than scattered hot pixels. Finally, the engine evaluates every position of a window at your target aspect ratio — efficiently, via a summed-area table — scoring each by the attention it captures, plus a bonus when the captured subject sits near a rule-of-thirds line inside the frame. The highest score wins. It is arithmetic all the way down: same photo, same answer, every time — and if you open the attention-map overlay you can watch it think. Tools that call this "AI" are describing the same family of heuristics with better marketing; genuinely trained detectors exist in paid cloud services, and for batch catalogue work they can be worth it — but they need your photos uploaded to someone else's server, which this page never asks for. For a stronger clean-up job after cropping, the image object remover works the same private way.
Faces, in three honest tiers
Faces deserve their own note because "face detection" is the most overclaimed phrase in this category — the previous version of this very page claimed it without running any. Here is what actually happens now, in descending order of precision. Tier one: if your browser ships the native FaceDetector API (Chromium browsers do), it locates real face boxes on-device and the engine boosts them strongly into the attention map, so a 9:16 story crop keeps heads instead of torsos. Tier two: everywhere else, the skin-tone term in the attention map still biases framing toward people — it cannot tell a face from a wooden table with certainty, but across ordinary portraits it pulls the right way. Tier three: you. Profile faces, tiny faces in a crowd, faces in shadow and faces turned away can defeat both automated tiers, which is why the drag override, the zoom slider and the placement control are always one gesture away, and why the panel states plainly which tier handled your photo instead of letting you assume the best one did.
The workflow this replaces
One shoot, six platforms. Drop the hero photo in, take the all-sizes pack, and the same image leaves as a square post, a 4:5 portrait, a 9:16 story, a 16:9 thumbnail, a 3:2 frame and a 2:3 pin — each framed around its subject, not resized copies of one compromise. Catalogue standardisation. Batch mode takes a folder of product photos shot at slightly different distances and returns uniform marketplace squares, each product centred by its own analysis; follow with the image compressor if the marketplace also has a file-size limit. Avatars and headshots. A 1:1 smart crop with face boost beats the platform's own centre-crop upload widget, which is how foreheads get lost. Vertical repurposing. Landscape video stills and event photos become Reels and TikTok covers with the speaker still in frame — check the placement control if they were standing stage-left. In each case the manual tools stay close: when a photo is mostly text, or deliberately mostly empty space, your judgement beats any attention map, and the tool is designed to get out of your way in one drag.
The honest limits — and your privacy
Attention is a proxy for importance, not a synonym. A bright billboard behind your subject can out-shout the subject; a scene with three equally strong subjects forces a choice no score can make for you (use the placement control, or widen the frame); a very small subject in a vast landscape may score less than the landscape's own texture. Panoramas cropped to 9:16 will always be a compromise — there is no window that keeps both ends. The coverage bars help you spot these cases: a smart crop that barely beats the centre crop is the engine telling you the photo is genuinely ambiguous. What never happens here is a data trade: the analysis, the optional face detection and the export all run in your browser's memory, your photos are never uploaded, and there is no account, quota or watermark. The exported file is written fresh from your original pixels by the canvas, which also means camera metadata (including GPS) does not survive into the download — strip it deliberately beforehand with the EXIF metadata cleaner if you need to inspect it first.
Related Tools You Might Like
Handy utilities in Image Tools
Image Compressor
Compress JPG, PNG and WebP images online for free — by quality slider or to an exact size like 100 KB. Batch ZIP download. No uploads, no sign-up.
WebP to JPG Converter
Convert WebP images to high-quality JPG format online for free. Batch convert multiple files with custom quality controls. No uploads, 100% private.
Image Object Remover
Remove unwanted objects, people, text & blemishes from photos free — Smart Fill runs 100% in your browser. No upload, no sign-up, full-resolution download.