Image compression is a trade: fewer bytes in exchange for some change in representation. The useful question is not βwhich quality number is best?β but βwhat is the smallest result that remains acceptable for this image, at its real display size?β That answer changes for a photograph, a logo, a screenshot, and a transparent graphic.
What the ToolsFree compressor does
The compression tab accepts JPEG, PNG, and WebP inputs of up to 20 MB, with a maximum batch of 10 files. The browser decodes each image, draws its pixels to an HTML canvas, and exports a JPEG using the selected quality. The original dimensions are preserved.
This behavior has two important consequences. First, compression is a lossy JPEG re-encode even when the input was PNG or WebP. Second, JPEG cannot preserve transparency. Use the separate conversion tab when you need an explicit JPEG, PNG, or WebP output, and keep the original file until you have checked the result.
Try a copy in the image tools. The file is decoded and re-encoded in your browser; ToolsFree does not receive its contents.
Choose a format from the image's job
- JPEG
- Suits photographs and other continuous-tone images. It is lossy, does not support transparency, and can produce ringing or block artifacts around text and sharp edges.
- PNG
- Uses lossless compression and supports an alpha channel. It is a safer choice for interface captures, line art, and graphics where exact edges or transparency matter, though photographs can be large.
- WebP
- Supports lossy and lossless coding plus transparency. Browser output still depends on the canvas encoder, so compare the actual result instead of assuming a fixed saving from the file extension.
- AVIF
- AVIF is included here for format comparison, but ToolsFree does not currently accept or export it. Use a dedicated encoder when an AVIF workflow is required.
What the quality slider actually saves
Advice about quality settings is usually given as a single number β βuse 80β β as though the picture did not matter. It does, more than the setting does. We ran four kinds of image through this compressor at four quality levels and recorded what it printed:
| What is in the picture | Input | q90 | q70 | q50 | q30 |
|---|---|---|---|---|---|
| Smooth gradient β sky, blur, soft background | 539.9 KB | β92% | β96% | β97% | β98% |
| Fine detail β foliage, fabric, gravel | 1.6 MB | β69% | β81% | β86% | β92% |
| Flat colour β chart, logo, diagram | 23.8 KB | β17% | β31% | β37% | β46% |
| Text and line art β a screenshot | 13 KB | +2611% larger | +1735% larger | +1368% larger | +1033% larger |
Read the columns and the rule falls out on its own. At quality 70 the smooth gradient loses 96 per cent of its bytes and the flat graphic loses 31 β the same setting, a threefold difference in outcome. Detail is where the slider earns its keep: the busy image swings from 69 per cent saved at quality 90 to 92 per cent at quality 30, while the gradient only moves from 92 to 98. If an image is already mostly smooth, dropping the quality buys you very little and costs you visible artefacts.
The last row is the one worth remembering. Text and line art do not compress as JPEG at all β they inflate. A 13 KB synthetic pattern of thin high-contrast strokes β how rendered text behaves, not a real screenshot β came back at 351.1 KB at quality 90, and even at quality 30 it was 146.7 KB, 11.3 times the original. This is not a defect in the encoder. JPEG works by throwing away high-frequency detail, and a hard black-on-white edge is nothing but high-frequency detail; the encoder spends its bits trying to reproduce an edge it is built to blur, and produces ringing around the strokes for the trouble. Keep those as PNG.
The compressor now says so directly: when the result comes back larger than the input it reports the increase in amber and tells you to keep the original, instead of the β0%β it used to show. We found that while measuring for this article β the tool was clamping the growth away, so a visitor could download a file many times bigger than the one they started with and see nothing to suggest it.
A repeatable quality check
- Duplicate the source; never make the only copy your experiment.
- Resize to the largest dimensions the destination actually displays.
- Export one high-quality candidate and one smaller candidate.
- Compare faces, gradients, text, thin lines, and high-contrast edges at 100% zoom.
- Check a representative phone as well as a desktop display.
- Record dimensions, format, quality setting, and byte size so the decision is reproducible.
If the compressed file is larger than the original, keep the original. Re-encoding an already optimized image can add artifacts without saving bytes. A quality slider is an encoder input, not a universal percentage of visual quality.
Privacy and operational limits
Local processing avoids uploading the image to ToolsFree, but it does not remove metadata as a documented guarantee, sanitize hostile files, or eliminate device risk. Browser memory grows with decoded pixel dimensions: a compressed file can expand substantially in memory. Large images or batches may therefore fail on a low-memory phone even below the 20 MB input cap.
For a publishing workflow, validate the downloaded file, keep the source, and test the page or application where the image will actually appear.