Somewhere between you and something you need — a job application, a visa portal, a listing, a government form — sits a box that says "Maximum file size: 200 KB". Your photo is 4.2 MB. The form offers no help, and the obvious move (drag the quality slider left until the number looks small enough) either overshoots into a blurry mess or never quite gets under the line.
The reason that feels so unpredictable is worth understanding, because once you know it the whole problem becomes mechanical. You cannot dial in a file size directly. No image format has a "make this 200 KB" setting. What you can control are four separate inputs, and the resulting file size is an output of all four working together. Most people only ever touch one of them.
Why the quality slider alone keeps failing you
A JPEG quality setting is not a percentage of file size. It is a coarseness knob for how aggressively detail is rounded away (the quantization step described in how image compression actually works). How many bytes that saves depends entirely on the picture.
Two images exported at the exact same quality 80 can land wildly far apart:
- A portrait against a plain studio backdrop is mostly smooth, predictable colour. There is little fine detail to discard, so it was already small and compresses to a modest file.
- A photo of autumn leaves, gravel, or a crowd is dense with fine, random texture — precisely the high-frequency information compression struggles to throw away without visible damage. The same quality 80 can produce a file several times larger.
This is why "just use 60% quality" advice fails. It is a setting, not a destination. Quality 60 on a busy landscape may still exceed your limit, while quality 60 on a simple graphic destroys it for no reason — it was never the thing making the file big.
The four levers that actually determine file size
Ranked by how much leverage each one gives you:
1. Pixel dimensions — by far the strongest lever. File size scales roughly with the area of the image, meaning the pixel count, not the width. Halving both width and height quarters the pixel count. A 4000×3000 photo holds 12 million pixels; the same photo at 1000×750 holds 750,000 — about 6% of the data before compression even starts. Nothing else comes close to this. It is also usually free, because almost no upload form actually displays your image at 4000 px.
2. Format. The same picture stored as PNG, JPEG, WebP, and AVIF can differ by an order of magnitude. A photo saved as PNG is the classic trap — PNG is lossless and was designed for graphics, so it faithfully stores every scrap of sensor noise. Converting that one photo to JPEG or WebP can cut the file by 80–90% before you touch anything else. (Which format for which job: JPG vs PNG vs WebP vs AVIF.)
3. Quality / compression level. The lever everyone reaches for first. Real, but it should be the last adjustment, not the first — and its useful range is narrower than people assume. Between quality 85 and 75 you typically save a lot and see nothing. Below roughly 45–50 you start seeing blocky artefacts and colour banding, and you are trading visible damage for diminishing returns.
4. Everything else — metadata, colour depth, extra channels. Camera EXIF, embedded colour profiles, thumbnails, and GPS tags add anywhere from a few kilobytes to a couple of hundred. Irrelevant when you are aiming at 2 MB; occasionally decisive when you are fighting for the last few KB under a 20 KB limit. Stripping metadata is also a privacy win — see what EXIF data reveals.
The method: resize first, convert second, compress last
The order matters, because each step changes what the next one has to do. Work through it in this sequence and you will hit almost any realistic target on the first or second attempt.
Step 1 — Decide the largest dimensions you actually need. Not the dimensions you have. Ask what the image is for:
| Use | Sensible max dimension |
|---|---|
| Form / ID upload, document attachment | 1000–1500 px on the long edge |
| Marketplace or listing photo | 1600–2000 px |
| Full-width website hero | 1920–2400 px |
| Email attachment | 1200–1600 px |
| Thumbnail or avatar | 400–800 px |
Resize to that first. Very often the file drops under the limit here and you are done, with zero visible quality loss — because you removed pixels nobody was going to look at rather than degrading the ones that remain. The mechanics, including why enlarging is a different and much harder problem: resize images without losing quality.
Step 2 — Make sure you are in the right format. If it is a photograph, it should be JPEG or WebP, never PNG. If it has transparency or hard-edged text and graphics, keep it lossless. If the destination accepts WebP, it will typically get you 25–35% below an equivalent-looking JPEG. If it is a government or legacy portal, assume it wants JPEG or PNG only — many still reject WebP outright, and a rejected upload is worse than a slightly bigger file.
Step 3 — Now bring quality down, and only as far as needed. Start around 80 and step down. Check the number after each step. Stop the moment you are under the limit — there is no prize for going further, and every step down is permanent.
Step 4 — If you are still over, go back to step 1, not further down on quality. This is the part people get wrong. When quality 45 still is not enough, the answer is almost never quality 35. It is that the image is still too many pixels. Drop the long edge by another 25% and return to quality 80. You will get a visibly better result at the same file size than you would by crushing quality alone — a smaller, cleaner image beats a large, artefact-ridden one every time.
Hitting the common targets
Rough starting points for a typical photograph. Treat these as a first guess to verify, not a guarantee — as explained above, image content genuinely changes the outcome.
| Target | Typical approach for a photo |
|---|---|
| Under 20 KB | 400–600 px long edge, JPEG quality 60–70, metadata stripped. Expect visible softness — this is thumbnail territory. |
| Under 50 KB | 700–900 px, JPEG quality 65–75. Fine for a small form photo. |
| Under 100 KB | 1000–1200 px, JPEG quality 70–80. The most common form limit; usually comfortable. |
| Under 200 KB | 1200–1600 px, quality 75–85. Genuinely no visible loss for most photos. |
| Under 500 KB | 1600–2000 px, quality 80–85. Plenty for a detailed listing image. |
| Under 1 MB | 2000–2500 px, quality 82–88. |
| Under 2 MB | Most photos are already under this after a resize to ~2500 px. |
Two patterns are worth noticing in that table. First, the tight targets are dominated by dimension choices, not quality choices. Second, once you are above about 200 KB for a normal photo, you have enough headroom that quality stops being a real constraint — you should be at 80+ and thinking about dimensions instead.
Special cases that break the normal rules
Screenshots and images of text. These behave the opposite way to photos. JPEG is actively bad at them — its 8×8 blocks smear crisp black-on-white edges into grey halos, and the artefacts are far more noticeable than on a photo. Keep screenshots as PNG (or WebP lossless) and get your savings from resizing and reducing the colour palette instead. A screenshot with a handful of flat UI colours compresses extremely well losslessly.
Signatures, stamps, and scanned line art. Often needed at brutally small sizes like 20 KB. These are near-black-on-near-white with almost no real colour information. Convert to a limited palette or greyscale and stay lossless — a signature as a small PNG will beat a JPEG at the same size and will not have muddy grey fringes around the ink.
Logos and anything with transparency. You cannot use JPEG at all; it has no alpha channel and will fill your transparency with solid white or black. PNG or WebP only. See make an image transparent.
PDFs are a different problem entirely. A PDF is a container, and its size is usually dominated by the images embedded inside it rather than the text. Compressing one means recompressing those embedded images — which is why targets like 2 MB or 500 KB behave differently from image targets, and why a text-only PDF barely shrinks at all.
When you genuinely cannot hit the number
Occasionally a limit is simply unreasonable for the content — a detailed multi-page scan under 100 KB, for instance. Before you accept a badly degraded file, check these in order:
- Is the requirement per-file or total? Many portals cap the sum of all attachments. Splitting a document into two uploads can solve it outright.
- Does it accept a different format? A form that wants a PDF may accept a JPEG, or vice versa, at a fraction of the size.
- Can you crop instead of shrink? If half the frame is empty desk or background, cropping removes those pixels entirely and keeps full detail on the part that matters. This is very often the best available answer for document photos.
- Is greyscale acceptable? Dropping colour from a scanned text document cuts the data substantially and usually costs nothing that matters.
- Is the source itself the problem? A photo of a screen, or a screenshot that was already saved as a low-quality JPEG once, carries baked-in noise and artefacts that compress terribly. Re-exporting from the original file, if you still have it, beats any amount of processing on the damaged copy.
Verify before you upload, and keep your original
Two habits that save real frustration:
Check the actual file size on disk, not an estimate. A preview that says "approximately 190 KB" is not what the form will measure. Confirm the saved file. And remember that some limits are stated in KB where the portal means kibibytes — if you are within a hair of the line, leave 5–10% of margin rather than landing at 199.6 KB against a 200 KB cap.
Never overwrite your original. Lossy compression is not reversible, and the damage compounds: compress a JPEG, then edit and save it again, and you are quantizing already-quantized data. This is generation loss, and it is why a photo that has been through several rounds of "make it smaller" looks progressively worse in a way no amount of later processing can undo. Keep the untouched original and always export target-sized copies from that, so a second attempt at a different limit starts from full quality.
Doing it in bulk
For a folder of listing photos or a batch of documents, the same order of operations applies — you just want it applied consistently rather than image by image. Set one dimension cap and one quality level for the whole set, run them together, then spot-check the two or three busiest images in the batch, since those are the ones most likely to land above the target. Walkthrough: bulk image compression.
Every step described here — resizing, converting, compressing, stripping metadata — runs inside your browser tab on this site. Nothing is uploaded to a server, which matters when the files in question are passports, bank documents, or a client's unreleased product shots. How that works: in-browser image tools and privacy.
How to tell when an image is already at its floor
Sometimes an image resists every technique, and it helps to recognise when you are fighting the file rather than your settings. Three signals that you have hit a genuine limit:
The size barely moves as quality drops. If going from quality 70 to quality 45 saves only a few percent, the bytes are not in discardable detail — they are in the sheer number of pixels. Reduce dimensions instead; quality has nothing left to give.
Artefacts appear before the target does. If you can see blocking and banding while still above the limit, the image is being asked to carry more scene complexity than the budget allows. Either the dimensions must come down or the target is unrealistic for that content.
The file was already compressed once. A photo saved from a messaging app, downloaded off a website, or screenshotted from a screen has been through lossy encoding already. Its apparent detail includes compression noise, which the next encoder faithfully preserves as if it were real texture — so it compresses worse than a clean original of the same subject. If you can find the untouched source, you will get a smaller and better file with less effort.
Frequently asked questions
Does resizing an image reduce its file size? Yes, and more effectively than anything else. File size scales with pixel count, so halving both dimensions cuts the data to roughly a quarter before compression is applied at all.
Why is my PNG so much bigger than a JPEG of the same photo? PNG is lossless — it stores every pixel exactly, including invisible sensor noise. JPEG discards detail your eyes do not register. For photographs the difference is routinely 5–10×. PNG is the right choice for graphics, screenshots, and transparency, not for photos.
Will compressing an image reduce its dimensions? No. Compression and resizing are independent. Compressing keeps the same pixel grid and stores it in fewer bytes; resizing changes the pixel grid itself. Most real size problems need both.
Can I get quality back after compressing too far? No. Lossy compression permanently discards information — raising the quality setting afterwards produces a larger file containing the same damage. This is why keeping the original matters.
Is there a difference between 1 MB and 1024 KB in an upload limit? Sometimes. Some systems treat 1 MB as 1,000,000 bytes and others as 1,048,576. When you are within a percent of the line, leave margin rather than testing which convention a portal uses.
Does compressing an image remove EXIF data? Not automatically — many tools preserve metadata through compression. If you want GPS coordinates and camera details gone, strip them deliberately.
Why does the same setting give different file sizes on different photos? Because the quality setting controls how aggressively detail is discarded, not how many bytes result. A detailed photo of foliage holds far more information than a portrait on a plain backdrop, so the same setting leaves it much larger.
Key takeaways
- File size is an output, not a setting. No format lets you request a byte count directly; you control dimensions, format, quality, and metadata, and the size follows.
- The same quality setting gives wildly different sizes on different images, because busy, detailed pictures hold far more information than smooth ones.
- Resize first — it is the strongest lever by far. Halving width and height cuts the pixel count to a quarter, usually with no visible loss at all.
- Fix the format second. A photo stored as PNG is the single most common cause of an absurdly large file.
- Lower quality last, and stop as soon as you are under the limit. If quality 45 is not enough, reduce dimensions again instead of going lower.
- Screenshots, signatures, and transparent graphics break the photo rules — those stay lossless and get their savings from dimensions and palette.
- Keep the untouched original. Compression is permanent and compounds each time you re-save.