How to Compress a PNG Without Losing Quality (Lossless PNG, 2026)
Last updated:
Quick Answer
True lossless PNG compression (via pngquant palette reduction and zlib optimization) reduces PNG file sizes 10–60% without changing any pixel. Lossless means the decompressed image is byte-for-byte identical to the original. For a PNG with transparency (logo, icon), this is the only way to compress without destroying the alpha channel. FileCurve runs pngquant-equivalent compression in your browser — upload, select "Lossless PNG", download.
Step-by-step
- 1
Upload your PNG to FileCurve Image Compressor.
- 2
Select "PNG (Lossless)" as the output format.
- 3
FileCurve applies zlib optimization and metadata stripping — typically 10–30% savings for already-reasonable PNGs.
- 4
For PNG graphics with flat colors (logos, icons, screenshots): enable "Indexed PNG (palette)" for 40–60% savings. This reduces to 256 colors — check if your PNG uses more.
- 5
Verify: download and open both original and compressed in a hex viewer or image diff tool. For true lossless, every pixel is identical.
- 6
If you need even smaller and can accept palette reduction, use "indexed PNG" mode — imperceptible for logos, may cause banding on gradients.
Expected output
Format
PNG (lossless)
Quality setting
Lossless — pixel-perfect identical to original
Estimated size
Metadata strip: 5–15% savings / zlib optimize: 10–25% / indexed: 40–60%
Why you might need this
- →Logo files that must preserve exact brand colors and transparency
- →Icon sets for web/app use where every pixel must be exact
- →Screenshot documentation where text must remain crisp
- →Medical or scientific images where pixel accuracy is mandatory
- →Game sprite assets where color accuracy and alpha channel are critical
Troubleshooting
My PNG is not getting smaller after lossless compression
It was already optimized. PNGs exported by Photoshop, GIMP, or pngquant are usually already near-optimal. Nothing further is losslessly possible.
Indexed PNG shows color banding on gradients
Indexed PNG uses 256 colors maximum — gradients need more. Use full-color lossless PNG for gradients, or convert to WebP lossless for best ratio with gradients preserved.
Alpha channel (transparency) is gone after compression
Make sure you selected "PNG" output, not "JPG". JPG does not support transparency. In FileCurve, always verify the output format shows PNG.
Screenshot PNG is 4MB and lossless only got it to 3MB
Screenshots with many colors (desktop screenshots) compress poorly losslessly. For screenshots, convert to JPG for best compression. Only keep PNG if transparency matters.
I need to send to a client — will they notice lossless vs lossy?
Lossless means they cannot — it is pixel-perfect. Lossy JPG at quality 90 is also imperceptible at display size. The difference only matters at pixel level.
Frequently asked questions
What is the difference between lossless and lossy PNG compression?
Lossless: the decompressed PNG is byte-identical to the original — no data lost. Lossy (palette/indexed PNG): color depth is reduced from millions to 256 colors — minor visible change possible. Both produce a PNG file, but indexed PNG is technically "near-lossless" not true lossless.
Is PNG always better than JPG for lossless?
For text, logos, and flat-color graphics — yes, PNG lossless is ideal. For photos, WebP lossless is 25–35% smaller than PNG. JPG has no lossless mode at all.
When does transparency preservation matter?
Logos on colored backgrounds, app icons, web buttons with rounded corners, cutout product images for e-commerce, and overlaying images in design tools.
What is pngquant and does FileCurve use it?
pngquant is a command-line tool that reduces PNG color palette (lossy). FileCurve uses equivalent browser-based techniques. True zlib lossless optimization is available separately.
Can I compress a 24-bit PNG to 8-bit (indexed) without visible loss?
For logos and icons with fewer than 256 distinct colors — yes, completely invisible. For photos or complex gradients — color banding becomes visible. Always visually check the output.