Free · Runs in your browser · Nothing is uploaded
Split a sheet of stickers into separate PNGs
Drop in a sheet of stickers, clipart or icons on a plain background. Every separate object is detected and cropped out as its own transparent PNG — open each one in a new page or download them all at once.
- JPG · PNG · WebP
- Transparent cut-outs
- Batch download
- No sign-up
Contents
Everything on this site
Jump straight to the tool, or read the guides and policies behind it.
The tool
Cut out your objects
Upload or paste an image, tune the sliders, and download the results.
Cropped objects
Upload or paste an image to begin. Your cropped objects will appear here.
Features
Built for real sticker sheets
Not a magic wand — a small set of well-behaved rules you can actually tune.
Automatic detection
Connected-component labelling walks the pixels and finds every separate shape, however irregular.
Clean, soft edges
Border pixels fade to transparent instead of leaving a white halo around the cut-out.
Inner whites survive
Only the background touching the outside is cleared. Eyes, highlights and paper inside the object stay put.
Join detached parts
A stray leaf, a dot or a caption within a few pixels of its object gets pulled into the same crop.
Dust filter
Anything smaller than a share of the biggest object is dropped — shadows and JPEG specks disappear.
Batch output
Open every crop on one contact sheet, or download them one after another as numbered PNGs.
Use cases
What people cut out
If it sits on a plain background, it can be split.
Sticker sheets
Turn a printable sheet into individual PNGs for chat packs, journals or a shop listing.
Icon & sprite sheets
Chop a grid of icons into named files, ready to drop into a design system.
Scanned artwork
Scan a page of doodles on white paper and get one file per drawing.
Product photos
Several items shot on a white sweep become separate cut-outs for a catalogue.
Worksheets
Pull every cut-out shape off a printable page for classroom resources.
Emoji & reaction packs
Split a grid of expressions into individual transparent images.
Are my images uploaded anywhere?
No. Everything happens in your browser using the HTML canvas. Your image never leaves your device, and there is no server to send it to.
What kind of sheet works best?
A flat, evenly lit image with a plain background — white, off-white or a single flat colour — and a little breathing room between objects.
Why did two objects end up in one crop?
They touch or overlap in the image, so the detector reads them as a single shape. Re-export the sheet with more spacing, or lower the background tolerance.
Is there a file size limit?
No hard limit, but very large images are slower because the whole picture is processed on your machine. Anything up to roughly 4000 × 4000 pixels feels instant on a modern laptop.
Ready to cut something out?
Scroll back up, drop in a sheet, and you'll have a folder of PNGs in under a minute.
Home / How it works
How Object Cutter works
Six steps, no machine learning, no server. Here's exactly what happens between dropping an image in and getting PNGs out.
The pipeline
-
Read the pixels
The image is drawn onto a canvas and read back as raw RGBA data. This all happens locally in your browser.
-
Guess the background colour
The border pixels are sampled and the median red, green and blue values are taken. On a sheet with a plain background, that median is the background.
-
Mark the foreground
Each pixel is compared to that background colour. If the total difference across R, G and B exceeds the tolerance, the pixel is foreground. Fully transparent pixels are always background.
-
Group the pixels
An 8-way connected-component pass walks the mask and labels every blob. Two shapes that touch share a label — which is why touching objects end up together.
-
Merge, join and filter
Blobs that heavily overlap merge (a sticker split by its own white border). Tiny blobs are dropped. Small blobs near a real object are absorbed into it, if they're within the "join nearby parts" distance.
-
Crop and cut out
Each surviving box becomes its own canvas. Pixels belonging to a neighbouring object are repainted as background, then a flood fill from the edges makes the outer background transparent and softens the boundary.
Why a flood fill? A naive "make every white pixel transparent" rule would punch holes through eyes, highlights and paper. Flooding inwards from the crop's edges only clears background that is actually connected to the outside.
Settings explained
Five controls, in the order you'll usually reach for them.
| Setting | What it does | Raise it when… | Lower it when… |
|---|---|---|---|
| Background tolerance | How different from the background colour a pixel must be to count as an object. | The background is off-white, grey or slightly noisy and edges are being missed. | Pale objects are being eaten, or the background is bleeding into the cut-outs. |
| Join nearby parts | Pulls small detached pieces into the nearest object if they're within this distance. | Objects have detached details — a dot on an "i", a leaf, a small caption. | Separate small items keep getting swallowed by their neighbours. Set to 0 to disable. |
| Ignore pieces smaller than | Drops any blob smaller than this percentage of the largest object's size. | Dust, shadows, watermarks or scan noise show up as tiny extra crops. | Genuinely small objects are going missing. |
| Padding | Transparent margin added around each crop. | You need room for a drop shadow or consistent spacing in a UI. | You want the tightest possible bounding box. |
| Remove background | Turns the outer background transparent instead of filling it with the detected colour. | Almost always — unless you specifically want an opaque rectangular crop. | You're pasting the result onto a coloured background and prefer a flat fill. |
Reading the preview
After each run the source image is redrawn with a dashed box and a number over every object that was found. The numbers match the file names — object-01.png, object-02.png and so on — and they're assigned in reading order: top to bottom, then left to right.
Performance
Detection runs on a downscaled grid capped at 1600 pixels on the longest side, so the labelling pass stays fast even on big scans. Cropping itself always happens at full resolution, so no detail is lost in the output.
Nothing is uploaded. There is no backend. You can disconnect from the network after the page has loaded and everything still works. See the privacy policy.
Try it on your own sheet
The fastest way to understand the sliders is to move them once.
Home / Guide
Preparing a sheet that cuts cleanly
Object Cutter is a pixel tool, not a mind reader. Ten minutes spent on the source image saves an hour of cleanup afterwards.
The short version
- Flat, plain background. Pure white, or one solid colour. No gradients, no texture, no vignette.
- Even lighting. If the background brightness varies across the sheet, one tolerance value can't cover all of it.
- Space between objects. At least a few pixels of clear background — ideally 2–3% of the image width.
- Objects clearly darker or lighter than the background. A pale yellow star on white is hard; the same star on mid-grey is easy.
- No drop shadows or glows unless you want them included in the crop.
- Reasonable resolution. 1000–4000 px on the long edge is the sweet spot.
Backgrounds, ranked
- Pure white (#FFFFFF) — the classic. Works with the default tolerance of 40.
- Flat mid-grey — the safest choice for pale or pastel artwork.
- Flat colour that nothing else uses — magenta and chroma-key green are traditional for a reason.
- Off-white paper — fine, but raise the tolerance to 50–70 to absorb the variation.
- Gradients, textures, photos — avoid. No single tolerance value will work.
Exporting from your design tool
From Illustrator, Figma or Inkscape
- Hide guides, grids and artboard labels before exporting.
- Export as PNG at 2× if the artwork is vector — the extra pixels give the edge softening more to work with.
- Flatten any transparency onto a solid background colour first, otherwise the "background" is genuinely transparent and detection gets confused.
From a scanner
- Scan at 300 dpi or higher.
- Turn off automatic colour correction and dust removal if they introduce halos.
- Expect some paper texture — a tolerance of 55–80 usually handles it.
From a phone photo
- Shoot straight down, not at an angle.
- Diffuse the light — a window on an overcast day beats a desk lamp.
- Keep the sheet flat. Curled paper creates soft shadows that read as objects.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Only one giant crop | Objects are touching, or the background is being read as foreground | Re-export with more spacing; lower the tolerance |
| Objects are chipped or eaten away | Tolerance too high | Lower it in steps of 10 |
| White halo around every object | Tolerance too low — background pixels survived | Raise the tolerance; keep "Remove background" on |
| Dozens of tiny junk crops | Dust, noise or watermark specks | Raise "Ignore pieces smaller than" |
| Small objects are missing | Speck filter is too aggressive | Lower "Ignore pieces smaller than" to 10–20% |
| Detached details became their own crops | Join distance too small | Raise "Join nearby parts" to 20–30 px |
| Everything is slow | Very large image | Downscale the sheet to ~3000 px on the long edge |
| Inner white areas turned transparent | Those pixels matched the background and touched the edge | Not possible to fix in one pass — use a different background colour for the source sheet |
Naming and output
Files are named after the uploaded image, with a two-digit counter: uploading summer-pack.png gives you summer-pack-01.png, summer-pack-02.png and so on. Numbering follows reading order, so it usually matches how you'd name them by hand.
Tip: if your browser asks to allow multiple downloads, say yes — otherwise "Download all" will only save the first file. Chrome, Edge and Firefox each ask once per site.
Put the guide to work
Export a clean sheet, then split it.
Using the tool
Are my images uploaded to a server?
No. The entire tool runs in your browser using the HTML canvas. There is no backend, no account and no upload step. You can even disconnect from the internet after the page loads and it will keep working.
What file types can I use?
Anything your browser can decode: JPG, PNG, WebP, AVIF and BMP. Animated GIFs are read as their first frame. You can also paste an image straight from the clipboard with Ctrl+V (or ⌘V).
Is there a size or file limit?
No hard limit. Very large images just take longer, because all the work happens on your device. Anything up to about 4000 × 4000 pixels feels near-instant on a modern laptop.
Does it work on a phone or tablet?
Yes. The layout stacks into a single column and you can pick an image from your photo library. Precision slider work is easier on a larger screen, though.
Can I use it offline?
Once the page and its fonts have loaded, yes. Save the page to your device and it will keep working without a connection.
Detection problems
Why did several objects get merged into one crop?
They touch or overlap in the image, so the connected-component pass sees them as one continuous shape. Re-export the sheet with more space between objects. Lowering the background tolerance can also help when a soft shadow is bridging the gap.
Why are small objects missing from the results?
The speck filter drops anything smaller than the percentage you set. If your smallest objects are genuinely tiny compared to the biggest one, lower "Ignore pieces smaller than" to 10–20%.
There's a white fringe around my cut-outs.
The tolerance is slightly too low, so near-background pixels survived as part of the object. Raise the tolerance in steps of 10 until the fringe disappears. Keeping "Remove background" checked also helps, since it softens the boundary.
Why did a white area inside my object turn transparent?
Because that area is the same colour as the background and is connected to the outside edge of the crop. The flood fill can't tell it apart from real background. The fix is at the source: export the sheet on a background colour that doesn't appear inside your artwork.
Can I cut out objects from a photograph?
Only if the objects sit on a flat, evenly lit background. Against a real-world scene there's no single background colour to detect, so results will be unpredictable. This tool is built for sheets, not for portrait-style cut-outs.
Output & downloads
What format are the results?
Always PNG. PNG is the only common web format with a real alpha channel, which is what makes clean transparent cut-outs possible.
How do the files get named?
After the uploaded file, with a two-digit counter in reading order — my-sheet-01.png, my-sheet-02.png, and so on.
"Download all" only saved one file.
Your browser is blocking automatic multiple downloads. Look for a permission prompt in the address bar and allow downloads from this page, then try again. Every crop also has its own individual Download button.
What does "Open all in new page" do?
It opens a temporary page showing every crop at full size on a checkerboard, with file names underneath. From there you can right-click any image and choose "Save image as…". It's handy for a quick visual check before downloading.
Privacy & licensing
Do you store or see my images?
No. Nothing is transmitted anywhere. See the privacy policy for the full detail — it's short.
Can I use the output commercially?
The tool makes no claim on your images. What you can do with the output depends entirely on the rights you hold in the source artwork. See the terms of use.
Something's broken. How do I report it?
Head to the contact page and describe what happened, which browser you're using, and roughly what the source image looked like.
Still stuck?
Send a description of the sheet and what came out — that's usually enough to diagnose it.
Why it exists
Splitting a sticker sheet by hand is miserable work. You open the image, drag a rectangle around object number one, export, undo, drag around object number two, export, and repeat until your wrist gives out. For a sheet of forty icons, that's forty rounds of the same three clicks.
Object Cutter replaces the clicking with detection. It reads the pixels, works out where one object ends and the next begins, and hands you a folder of numbered PNGs instead of a single flat sheet.
What makes it different
- It runs locally. There is no server component at all. That means no upload wait, no file size limit imposed by a host, and no question about where your artwork ends up.
- It's tunable, not magic. Five sliders cover the vast majority of real-world sheets. Every one of them is explained in plain language on the settings page.
- It respects inner detail. The flood-fill cut-out keeps white areas inside an object intact, which is where most naive background removers fall over.
- It's small. One page, no framework, no tracking, no build step. The whole thing is a single HTML file.
How it's built
Plain HTML, CSS and JavaScript. Detection uses an 8-way connected-component pass over a downscaled mask, followed by a merge-and-filter stage. Cropping uses getImageData and putImageData on an offscreen canvas, with a flood fill for transparency and a distance-based alpha ramp for soft edges.
No libraries, no dependencies, no analytics. The only external request is the web font stylesheet, and the page degrades gracefully without it.
Who it's for
Designers turning a sticker pack into shop assets. Teachers cutting up a worksheet. Developers extracting a sprite sheet. Anyone who has ever thought "I wish I could just… select all of these."
Limitations, honestly
- It assumes a flat, fairly uniform background. Photographs of objects in the real world won't work.
- Objects that physically touch in the image will be treated as one object.
- Pale artwork on a pale background needs a tolerance the eye can't easily predict — experiment.
- Very large images are limited by your device's memory, not by us.
Credits
Typefaces are Bricolage Grotesque and Figtree, served by Google Fonts. Everything else is hand-written.
Questions, bugs or ideas?
The contact page is the place for all three.
Bug reports
Include your browser, the slider values you used, and what you expected to happen.
Response time
Usually within two working days. This is a small project, so please be patient.
Send a message
This form opens your email client with the details filled in — nothing is sent to a server.
Before you write
- Check the FAQ — most detection oddities are covered there.
- Check the troubleshooting table for the symptom you're seeing.
- If a sheet won't cut correctly, describe it: background colour, roughly how many objects, and whether any of them touch.
Home / Privacy
Privacy policy
Last updated: 1 January 2026. The short version: your images never leave your device.
Your images
Object Cutter processes images entirely in your browser using the HTML canvas API. No image, or any part of an image, is uploaded, transmitted, stored or seen by anyone but you. There is no server-side component to send it to.
When you close or reload the page, everything you loaded is discarded from memory.
What we don't collect
- No accounts, sign-ups or passwords.
- No analytics, telemetry, heatmaps or session recording.
- No advertising or third-party trackers.
- No cookies.
- No server logs of your activity, because there is no server.
What is stored on your device
One thing: your colour-theme preference (light or dark) is saved in localStorage under the key oc-theme. That's it. It never leaves your browser and you can clear it at any time through your browser's site-data settings.
Third-party requests
Loading the page makes one external request: the web font stylesheet from Google Fonts (fonts.googleapis.com) and the font files themselves from fonts.gstatic.com. As with any request to a third party, Google receives your IP address and user-agent string. The fonts are not used to identify you by this site.
If you'd prefer to avoid that request entirely, block fonts.googleapis.com — the page falls back to your system fonts and works exactly the same.
Children's privacy
The tool collects no personal data from anyone, including children under 13. There is nothing to collect.
Your rights
Because no personal data is collected or processed by this site, there is no data to access, correct, export or delete. If you believe otherwise, please get in touch.
Changes to this policy
If this policy changes, the "last updated" date at the top will change with it. Material changes will be noted on the home page.
Contact
Questions about privacy can go to privacy@example.com or through the contact page.
1. What this is
Object Cutter is a free, browser-based tool for splitting an image containing multiple objects into separate cropped images. By using it you agree to these terms.
2. Your images stay yours
We claim no ownership, licence or rights over anything you load into the tool or anything it produces. The output belongs to you, subject to whatever rights you hold in the original artwork.
3. Your responsibilities
You are responsible for making sure you have the right to use and modify the images you process. Do not use the tool to infringe copyright, trademarks or any other rights, and do not use it for anything unlawful.
4. No warranty
The tool is provided "as is", without warranty of any kind, express or implied. Detection is heuristic and will not be perfect on every image. We do not guarantee that any particular object will be found, separated or cut out to your satisfaction.
5. Limitation of liability
To the fullest extent permitted by law, we are not liable for any loss or damage arising from your use of the tool — including lost time, lost work, or output that doesn't meet your requirements. Always keep the original files.
6. Availability
The tool is offered free of charge and may change, break or be withdrawn at any time without notice. Because it runs entirely in your browser, a copy you have saved locally will keep working even if the hosted version changes.
7. Acceptable use
- Don't attempt to use the tool to process material that is illegal in your jurisdiction.
- Don't redistribute the tool as your own original work.
- Don't misrepresent the tool's output as being produced by a different service.
8. Third-party components
The page loads web fonts from Google Fonts, which are governed by Google's own terms. Everything else — markup, styles and scripts — is original to this project.
9. Changes to these terms
These terms may be updated from time to time. The "last updated" date at the top reflects the most recent revision. Continued use after a change constitutes acceptance.
10. Contact
Questions about these terms: legal@example.com, or use the contact page.