Shrinking Images to a File Size: What We Measured
What a 12 MP photo keeps from 1 MB down to 50 KB, and how photo type, JPG vs WebP, quality and resizing first change it, measured with Hanji tools.
An upload form says “200 KB or smaller”, and two questions follow: how much blurrier will the photo get, and will mine even shrink that far? Both depend on the photo. A detail-heavy 12 MP landscape loses more than half its width at 200 KB, while an ID photo on a plain background keeps its full size at the same 200 KB.
This guide is built on measurements, not general advice. We fed test photos to Hanji’s Image Compressor and Image Resizer and measured the file size, resolution and difference from the original of every result. Find the row in the tables below that looks most like your photo.
The short version
- A detail-heavy 12 MP photo keeps its full size only down to 1 MB. A 4000×3000 photo became 2717×2038 at 500 KB, 1718×1289 at 200 KB, 1215×911 at 100 KB and 838×628 at 50 KB. That is 43% of the width at 200 KB and 21% at 50 KB.
- An ID photo on a plain background kept its full 3000×4000 down to 200 KB. It became 2334×3112 at 100 KB and 1486×1981 at 50 KB. At the same 200 KB, the landscape kept only 43% of its width.
- A screenshot (PNG) is left alone when the target is above its size, and loses resolution quickly below that. A 281 KB screenshot was untouched at 500 KB and 1 MB. At 200 KB it stayed full size (PSNR against the original: 42.9 dB), at 100 KB it became 1020×2207 (32.0 dB) and at 50 KB 628×1358 (27.5 dB).
- The quality slider alone cannot shrink a 12 MP landscape. Even at quality 10 it was still 301 KB, so getting below 200 KB means reducing the pixel dimensions. For the ID photo, quality 60 gave 205 KB and quality 10 gave 95 KB, but near 100 KB lowering only the quality (31.9 dB) was clearly worse than letting size and quality shrink together (38.2 dB).
- WebP keeps more pixels at the same file size. At 50 KB the ID photo was 1486×1981 as JPG and 2494×3325 as WebP. At 100 KB the screenshot was 1020×2207 as JPG and its full 1170×2532 as WebP.
- Asking for 1 KB did not make the tool fail. The results were dots, though, between 23×23 and 90×90 pixels. Nothing useful.
How we measured
On October 27, 2026 we put files into the live Image Compressor and Image Resizer pages at hanji.page in Chrome 155.0.8059.85 on Windows, then downloaded and measured the results. 1 KB is 1,000 bytes. The compressor tries JPG quality from 92 down to 45 and reduces the resolution only if 45 is still too big. It has no quality slider, so the slider tests used the Image Resizer instead.
- Five test photos. Two are real photos and three are pictures we generated.
- Real photos: two portraits from our own test set (960×1200 JPG, 226 KB, and 512×512 JPG, 62 KB).
- Generated landscape: 4000×3000 JPG, 3.59 MB. Sky, mountains, a forest and 90,000 blades of grass, plus noise, so it is extremely detailed.
- Generated ID photo: 3000×4000 JPG, 1.15 MB. A flat blue background with a face, hair and jacket, and a little noise.
- Generated screenshot: 1170×2532 PNG, 281 KB. An order-history screen with text and white boxes.
- Quality numbers. We scaled each result back to the original size and compared it with the original: PSNR (in dB, higher is closer to the original) measures how far the colors moved, and SSIM (closer to 1 is more alike) measures how well the light and dark patterns match. The loss from shrinking the resolution is included. Use these to compare targets within one picture; do not compare numbers between different pictures directly.
- What we saw. We also opened the same files on screen and looked at them.
The limits are real. Generated pictures do not capture real faces, skin or blurred backgrounds. The generated landscape was made on purpose to be harder to shrink than a typical landscape photo, so a real one will probably keep more pixels at the same size. The two real photos followed the same direction (last two rows of the first table). Because Chrome saved the test pictures at JPG quality 90, quality 90 and above in the slider test produces files almost identical to the original and the comparison means nothing, so we left those numbers out.
Results
Resolution and quality kept at each target
Each cell is “resulting size · PSNR (dB) / SSIM”. “Unchanged” means the original was already under the target and the tool returned it as is (JPG, PNG and WebP files are kept in that case). Output is JPG.
| Your photo | 1 MB | 500 KB | 200 KB | 100 KB | 50 KB |
|---|---|---|---|---|---|
| Detail-heavy 12 MP landscape (4000×3000, 3.59 MB) | 4000×3000 · 30.3 / 0.871 | 2717×2038 · 27.6 / 0.822 | 1718×1289 · 25.0 / 0.751 | 1215×911 · 23.5 / 0.690 | 838×628 · 22.5 / 0.632 |
| ID photo, plain background (3000×4000, 1.15 MB) | 3000×4000 · 46.0 / 0.985 | 3000×4000 · 43.6 / 0.971 | 3000×4000 · 40.5 / 0.966 | 2334×3112 · 38.2 / 0.962 | 1486×1981 · 35.6 / 0.957 |
| Screenshot with text (1170×2532 PNG, 281 KB) | Unchanged | Unchanged | 1170×2532 · 42.9 / 0.999 | 1020×2207 · 32.0 / 0.977 | 628×1358 · 27.5 / 0.936 |
| Real portrait (960×1200, 226 KB) | Unchanged | Unchanged | 960×1200 · 36.8 / 0.985 | 864×1080 · 28.8 / 0.898 | 575×719 · 26.3 / 0.840 |
| Real photo (512×512, 62 KB) | Unchanged | Unchanged | Unchanged | Unchanged | 512×512 · 37.2 / 0.990 |
The files filled the budget without going over it. The landscape came out at 197,471 bytes for the 200 KB target and 97,951 bytes for 100 KB, for example.
How to read it:
- The more fine texture, the more resolution is given up. A photo with lots of fine texture is big at any given quality, so resolution is cut first. The ID photo, with its large flat areas, kept its resolution at the same file size.
- A high SSIM can still hide a blurrier picture. The ID photo at 50 KB has an SSIM of 0.957 even though the width was halved. Pictures with large flat areas tend not to lose SSIM easily. Look at resolution and PSNR together.
- What we saw. The 100 KB landscape keeps the clouds and the mountain outline, but individual grass blades smear into a soft texture. It is fine as a whole picture on screen and blurry when you zoom in. The 50 KB screenshot shrank to 628 pixels wide, yet both the large and the small text were still readable, with only the smallest line a little soft. The 50 KB ID photo has clean face edges and background; only the fine hair strands get rough.
How different are different kinds of photo?
The same results, written as the share of the width that is left:
| Your photo | Width left at 200 KB | 100 KB | 50 KB |
|---|---|---|---|
| Detail-heavy landscape (4000) | 43% (1718) | 30% (1215) | 21% (838) |
| ID photo, plain background (3000) | 100% (3000) | 78% (2334) | 50% (1486) |
| Screenshot (1170) | 100% (1170) | 87% (1020) | 54% (628) |
| Real portrait (960) | 100% (960) | 90% (864) | 60% (575) |
As an extreme case, a 1000×1000 picture filled with random colored noise (1.02 MB) became 726×726 at 200 KB, 514×514 at 100 KB, 363×363 at 50 KB and 270×270 at 10 KB. The more fine texture, the more resolution you pay.
File size by quality-slider value
In the Image Resizer at 100% (same size) with JPG output, we changed only the quality. Sizes are in KB, with PSNR (dB) against the original in brackets. Quality 90 and above has no PSNR because the original was saved at 90.
| Quality | 12 MP landscape | ID photo, plain background | Screenshot |
|---|---|---|---|
| 100 | 9,425 KB | 3,127 KB | 592 KB |
| 90 | 3,555 KB | 1,147 KB | 231 KB |
| 80 | 2,440 KB (33.2) | 727 KB (44.0) | 179 KB (41.6) |
| 70 | 1,578 KB (32.2) | 290 KB (43.0) | 152 KB (39.9) |
| 60 | 1,294 KB (31.1) | 205 KB (40.5) | 135 KB (38.3) |
| 50 | 1,038 KB (30.5) | 173 KB (40.3) | 124 KB (37.6) |
| 40 | 852 KB (29.9) | 142 KB (40.4) | 114 KB (36.7) |
| 30 | 681 KB (29.1) | 139 KB (38.5) | 102 KB (35.6) |
| 20 | 504 KB (27.9) | 114 KB (37.0) | 85 KB (33.8) |
| 10 | 301 KB (25.8) | 95 KB (31.9) | 61 KB (31.2) |
- Quality 100 is far bigger than the original. The landscape went from 3.59 MB to 9.4 MB and the 281 KB screenshot to 592 KB. If the goal is a smaller file, avoid 100.
- The ID photo curve is not smooth. Going from quality 80 to 70 cuts 727 KB to 290 KB, below 60 it shrinks only a little at a time, and at 10 the PSNR drops sharply to 31.9 dB. Where the big drop happens differs from photo to photo.
- The landscape is still 301 KB at quality 10. The slider alone does not get it to 200 KB.
- Saving a screenshot as JPG can make it bigger than the PNG. Quality 100 gave 592 KB, more than twice the PNG’s 281 KB. From quality 90 it is 231 KB.
JPG vs WebP
In the Image Compressor we changed only the output format and used the same targets. Cells are “size · PSNR / SSIM”.
| Your photo | Target | JPG | WebP |
|---|---|---|---|
| 12 MP landscape | 500 KB | 2717×2038 · 27.6 / 0.822 | 3139×2354 · 29.2 / 0.834 |
| 12 MP landscape | 200 KB | 1718×1289 · 25.0 / 0.751 | 1985×1489 · 26.2 / 0.770 |
| 12 MP landscape | 100 KB | 1215×911 · 23.5 / 0.690 | 1264×948 · 24.4 / 0.702 |
| 12 MP landscape | 50 KB | 838×628 · 22.5 / 0.632 | 891×668 · 23.1 / 0.643 |
| ID photo | 200 KB | 3000×4000 · 40.5 / 0.966 | 3000×4000 · 43.1 / 0.968 |
| ID photo | 100 KB | 2334×3112 · 38.2 / 0.962 | 3000×4000 · 42.6 / 0.966 |
| ID photo | 50 KB | 1486×1981 · 35.6 / 0.957 | 2494×3325 · 39.3 / 0.963 |
| Screenshot | 200 KB | 1170×2532 · 42.9 / 0.999 (199 KB) | 1170×2532 · 46.4 / 1.000 (111 KB) |
| Screenshot | 100 KB | 1020×2207 · 32.0 / 0.977 | 1170×2532 · 45.9 / 1.000 |
| Screenshot | 50 KB | 628×1358 · 27.5 / 0.936 | 1053×2279 · 32.9 / 0.981 |
In every row WebP had the same or higher resolution and PSNR. The gap was biggest for pictures with large flat areas (the ID photo and the screenshot); for the detail-heavy landscape WebP kept 4–16% more width. At the 200 KB target the WebP screenshot was only 111 KB even at the highest quality, so it ended far below the target.
First check whether the place you are submitting to accepts WebP; if the form only takes JPG, a WebP file does not help. The image formats guide compares the formats. Also, a PNG with a transparent background turns the transparent areas white when it is shrunk to JPG.
Resizing first vs lowering only the quality
We reached the same target in three ways: letting the tool do it, lowering only the quality, and choosing a size first and then the quality. The tool’s own method is the Image Compressor; the other two use the Image Resizer. Cells are “size, file size, PSNR / SSIM”.
12 MP landscape, target around 200 KB
| Method | Result |
|---|---|
| Let the tool do it | 1718×1289, 197 KB, 25.0 / 0.751 |
| Lower only the quality (down to 10) | 4000×3000, 301 KB. Does not reach 200 KB |
| Resize to 1600 wide first, quality 60 | 1600×1200, 194 KB, 24.8 / 0.740 |
| Resize to 2000 wide, quality 60 | 2000×1500, 314 KB. Does not reach 200 KB |
ID photo on a plain background, target around 200 KB
| Method | Result |
|---|---|
| Let the tool do it | 3000×4000, 194 KB, 40.5 / 0.966 |
| Lower only the quality (60) | 3000×4000, 205 KB, 40.5 / 0.967 |
| Resize to 2250 wide, quality 75 | 2250×3000, 155 KB, 39.3 / 0.964 |
| Resize to 2250 wide, quality 90 | 2250×3000, 526 KB. Does not reach 200 KB |
ID photo on a plain background, target around 100 KB
| Method | Result |
|---|---|
| Let the tool do it | 2334×3112, 92 KB, 38.2 / 0.962 |
| Lower only the quality (10) | 3000×4000, 95 KB, 31.9 / 0.961 |
| Resize to 2250 wide, quality 60 | 2250×3000, 104 KB, 38.0 / 0.962 |
What to take from it:
- For detail-heavy photos, reducing the pixel size is practically the only way. Quality alone did not bring the 12 MP landscape to 200 KB. When we picked a size by hand, the result was close to what the tool chose.
- For photos with large flat areas, keeping the size is better around 200 KB. None of the resize-first combinations we measured beat 40.5 dB (the highest was 39.9 dB at 526 KB).
- Lowering only the quality is a bad deal around 100 KB. Going down to quality 10 dropped the PSNR from 38.2 dB to 31.9 dB.
- The 500 KB landscape was an exception. Lowering only the quality to 20 gave a 504 KB file at 27.9 dB / 0.837, slightly higher than the file the tool made at the 500 KB target by shrinking the size (27.6 dB / 0.822). It missed the target by 4 KB, though, and we did not compare the two by eye. The tool does not go below quality 45 and reduces the size first.
When the target is too small
The smallest target in the Image Compressor is 1 KB. These are the resolutions we got on the three generated photos as we lowered the target down to 1 KB.
| Target | 12 MP landscape | ID photo | Screenshot |
|---|---|---|---|
| 30 KB | 649×487 | 1140×1521 | 413×893 |
| 20 KB | 530×397 | 887×1182 | 352×761 |
| 10 KB | 375×281 | 543×723 | 241×523 |
| 5 KB | 256×192 | 305×407 | 151×327 |
| 3 KB | 178×134 | 183×243 | 109×237 |
| 2 KB | 138×103 | 114×153 | 78×169 |
| 1 KB | 52×39 | 29×38 | 32×68 |
In every cell the tool did not fail and returned a file within the target. The same held for the two real photos (23×29 and 23×23 at 1 KB) and for the random-noise picture (90×90 at 1 KB). The code does have a path that fails when a side would drop below 16 pixels, but none of our photos reached it. Not failing does not mean the result is useful.
- Once the width falls below 500 pixels (landscape at 20 KB, ID photo at 10 KB), do not expect small text. The screenshot in that range is 352×761 (20 KB) and 241×523 (10 KB). We did not measure whether text is readable at that size.
- Near 1 KB the result is a dot, smaller than an icon. The real photos shrank to about 23 pixels.
Choosing with these results
- Check the target first. If the photo is already smaller than the target, the tool returns the original (for JPG, PNG and WebP). There is nothing to do.
- Pick the row of the table that looks like your photo. Lots of fine texture (grass, leaves, noise) is the landscape row, a plain background is the ID photo row, and a screen full of text is the screenshot row.
- Check that enough resolution is left. If the place you submit to asks for a minimum width in pixels, pick a target that does not go below it. If the file size limit comes first, crop the photo beforehand so the pixels you keep are the ones that matter.
- If WebP is accepted, use WebP. It keeps more pixels at the same size. If you do not know, use JPG.
- Use the resizer only when you want to decide the size yourself. The Image Resizer’s results were not very different from what the tool chose on its own.
What each size is good for (signatures, ID photos, printing, email) is in the upload size limits guide. For PDFs, see making a PDF smaller.
Privacy
The Image Compressor and Image Resizer process the photo inside your browser, so it is never uploaded. We ran the test photos for this guide on this PC only.
Related tools
- Image Compressor: pick a target size (20 KB to 1 MB, or your own number) and the image is shrunk to fit inside it.
- Image Resizer: change the size by percentage or by width and height in pixels, and choose the quality yourself.
Guides
- Photo too large to upload? How file size limits work
- HEIC photos from iPhone: why forms reject them and fixes
- JPG, PNG, WebP, AVIF and HEIC compared
- Pixels, cm and DPI: how to convert, with tables
- How to Make a PDF Smaller (and What You Give Up)
- Scan Documents to PDF with Your Phone
- ID and passport photo sizes in mm and pixels
- How to Take a Passport Photo at Home That Gets Accepted
- Photo file rules for online applications: pixels, KB and age
- Do Online File Converters Keep Your Files?
- Print Passport Photos on 4×6: CVS, Walgreens, Walmart
- Character and Word Counts: Why Counters Disagree
- QR Codes That Keep Working: Static, Dynamic and Print
- Email Attachment Size Limits: Gmail, Outlook, Yahoo, iCloud
- Why Text Extraction From Photos Goes Wrong, and How to Fix It