Runs 100% in your browser — your images are never uploaded
Image Compressor
Shrink JPEG, PNG, WebP and AVIF by up to 90%. Hit an exact size in KB, convert formats, resize in bulk — and let on-device AI keep your subject sharp while the background compresses harder.
Compress your images
Drop images here, or click to browse
You can also paste with Ctrl + V · up to 60 files, 60 MB each
JPEGPNGWebP AVIFGIFBMPOutput & quality
Auto measures each image and picks the format and quality to match — photos to WebP or AVIF, logos and screenshots kept crisp.
75–85% is the sweet spot for photos; below 50% expect blocking and banding.
Resize
Resizing is the single biggest win. A 4000 px phone photo shown in a 1200 px column is carrying about ten times the pixels it needs.
AI subject detection ON-DEVICE
How to compress an image
Three steps, no account, and nothing leaves your device.
Add your images
Drag files in, click to browse, or paste with Ctrl+V. Up to 60 at a time — and your first batch starts compressing straight away, no button needed.
Adjust if you need to
Leave the format on Auto and set a quality, or switch to Target file size and type the exact number of kilobytes you need. Press Compress to re-run.
Check and save
Compare each result with the before/after slider, then save individually or grab the whole batch as a ZIP.
What this compressor does
Built for people who care about the number in the corner of the page-speed report.
Exact target size
Need it under 100 KB for a form that rejects anything bigger? Type the number. The encoder runs up to eight times, narrowing in on the highest quality that still fits.
AI that earns its place
U²-Net finds your subject, then detail outside it is softened before encoding. Blurred regions carry almost no high-frequency data, so the encoder spends its bits where you are actually looking.
Genuinely private
No upload, no queue, no server that could log your file. Canvas and Web Workers do the work locally, and the page keeps running if you disconnect after it loads.
Smart Auto mode
Each image is measured — unique colours, edge energy, transparency — and routed accordingly. Photos take AVIF or WebP at 78%; logos and UI screenshots stay near-lossless so text keeps its edges.
Batch and ZIP
Sixty images at once, each with its own before/after, savings percentage and download. Take the lot as a single ZIP when you are done.
Metadata stripped
Re-encoding rebuilds the file from pixels, so EXIF, GPS coordinates and camera serial numbers are dropped. The orientation flag is read first, so portrait photos are not rotated sideways.
Which image format should you use?
The format matters more than the quality slider. Here is the short version.
| Format | Best for | Typical vs JPEG | Transparency | Watch out for |
|---|---|---|---|---|
| WebP | Almost everything on the web today | 25–35% smaller | Yes | Nothing significant — every current browser supports it |
| AVIF | Large photos and hero images | Around 50% smaller | Yes | Slower to encode, and not every browser can create it |
| JPEG | Maximum compatibility, old software | Baseline | No | No alpha channel, and visible blocking at low quality |
| PNG | Logos, icons, flat art, sharp UI screenshots | Usually much larger for photos | Yes | Lossless, so a quality setting does nothing |
Frequently asked questions
Are my images uploaded to a server?
No. Every image is decoded, resized and re-encoded by your own browser using Canvas and Web Workers. Nothing is transmitted, so private photos, client work and screenshots containing sensitive data never leave your device. The page keeps working if you go offline after it loads.
How do I compress an image to an exact size like 100KB or 200KB?
Switch the mode from Quality to Target size and type the number of kilobytes you need. The tool then runs a binary search over the encoder quality, re-encoding the image up to eight times, and keeps the highest quality result that still fits under your limit. If even the lowest quality overshoots the target, the result is flagged as above target rather than silently failing.
Which format should I choose: WebP, AVIF, JPEG or PNG?
WebP is the safest default and is typically 25 to 35 percent smaller than JPEG at matching quality, with support in every current browser. AVIF compresses smaller still, often 50 percent below JPEG, but encoding is slower and not every browser can create it. Keep JPEG only when a tool or client insists on it. Use PNG for flat graphics, logos and screenshots with sharp edges, or when you need lossless output.
What does AI saliency-guided compression actually do?
The U²-Net model runs in your browser and predicts which pixels belong to the main subject. The tool then softens detail outside that subject before encoding. Because blurred regions contain far less high-frequency information, the encoder spends dramatically fewer bits on the background while the subject keeps its full sharpness, so the file gets smaller without the subject looking degraded.
How large are the AI models and how often do they download?
U²-Net is about 4.4 MB and RMBG-1.4 is about 42 MB. Each is downloaded once and stored in the browser Cache API, so later visits load it instantly with no network request. The AI features are entirely optional and off by default, so you never pay that cost unless you switch one on.
Does compressing an image remove EXIF and GPS data?
Yes. Re-encoding through Canvas rebuilds the file from raw pixels, so EXIF blocks, GPS coordinates, camera model, serial numbers and editing history are all discarded. The tool does read the EXIF orientation flag before discarding it, so portrait photos from a phone stay the right way up instead of coming out sideways.
Why did my PNG get larger instead of smaller?
PNG is lossless, so the quality slider has no effect on it and a re-encode can be slightly larger if the original was optimised by a dedicated tool. If a PNG does not need transparency or pixel-exact fidelity, convert it to WebP instead, which is usually far smaller. Auto format does this for you.
Is there a file size or usage limit?
You can process up to 60 images at a time and each file may be up to 60 MB. There is no daily cap, no account, no watermark and no queue, because the work happens on your computer rather than on a shared server.
Does image compression reduce image quality?
Lossy formats such as JPEG, WebP and AVIF discard detail the eye is least likely to notice, so there is always some loss, but between roughly 75 and 85 percent quality the difference is very hard to see while the file is often a fraction of the original. Use the before and after slider to judge each image, and raise the quality if you can spot artefacts.
Does the compressor work on a phone or tablet?
Yes. The layout is fully responsive and the drop zone opens the camera roll on iOS and Android. Very large batches and the 42 MB RMBG-1.4 model are heavier on a phone, so the smaller U²-Net model is the better choice on mobile.
A short guide to compressing images well
Most pages are heavy because of their images, not their code. A single unoptimised phone photo can outweigh every script and stylesheet on a page combined. The good news is that image weight is one of the easiest things on the web to fix, and you rarely have to trade away visible quality to do it.
Resize before you compress
This is the step people skip, and it is the one that matters most. A modern phone shoots at around 4000 px wide. If that photo is displayed in a 1200 px column, you are shipping roughly ten times the pixels anyone will ever see, and no amount of quality tuning recovers that waste. Set the longest edge to about twice your largest display size — 2400 px for that 1200 px column — and you will cover high-density screens with room to spare.
Pick the format before you touch the slider
Switching a photo from JPEG to WebP typically saves a quarter to a third of the file at the same visual quality, and AVIF can halve it. That is a far bigger win than nudging quality from 80% to 70%, and it costs you nothing perceptually. Flat artwork works the other way: logos, icons and diagrams with large areas of solid colour belong in PNG or lossless WebP, because lossy encoders produce visible ringing around hard edges.
Where the quality slider actually sits
For photographs, 75–85% is where almost everyone should live. Above 90% the file grows quickly for detail nobody can see. Below about 60% you start to get blocking in smooth gradients — skies and skin are the first places it shows. Screenshots containing text are the exception: they need 90% or more, or the letterforms pick up coloured fringes.
What saliency-guided compression buys you
Every lossy encoder spends bits in proportion to local detail, and it has no idea which detail you care about. A busy background of leaves or a crowd will consume a large share of your file even though the viewer is looking at the face in front of it. Predicting the subject first and softening everything else redirects that budget: the background gives up detail that was never being examined, and the subject keeps its full sharpness. On images with a clear subject and a busy background, this is where the largest savings hide.
Check the result, not just the number
A savings percentage tells you nothing about whether the image still looks right. Use the before/after slider on at least one image from each batch, at full size, and look at the places compression fails first: smooth gradients, hair against a background, small text, and areas of saturated red.