PNG Compressor

No, and it is worth being precise about this. A true PNG-8 stores a palette table and one index per pixel. What this tool does is round each colour channel onto a coarser ladder — with Floyd-Steinberg error diffusion so gradients dither rather than band — and then re-encode as an ordinary full-colour PNG. The file has far fewer distinct colours in it, which is exactly what the compressor exploits, but it is not an indexed image. Each finished file reports the colour count it actually ended up with, rather than the number you asked for.

Because PNG has no quality dial to turn. It is a lossless format, and the browser canvas encoder ignores the quality argument outright when the output is PNG — any tool presenting you a PNG quality slider is presenting a control that does nothing. Real PNG compression therefore has exactly two levers: reduce how many distinct colours the file has to describe, which is what palette quantisation does, and reduce how many pixels there are, which is resizing. This tool uses the first; Resize PNG uses the second.

It depends entirely on what the image contains. Flat graphics — logos, icons, diagrams, UI screenshots — usually contain well under 256 distinct colours already, so quantisation is close to free and the savings are large. Photographs and smooth gradients contain tens of thousands, and forcing them into 256 shows up as banding or dither speckle in skies and skin tones. Use the before-and-after comparison on your own image rather than a rule of thumb.

Usually not. PNG was designed for images with sharp edges and flat colour and it stores photographs very inefficiently — the same photo as a good-quality JPEG or WebP is routinely five to ten times smaller with no visible difference. If your PNG is a photo or a screenshot of one, converting the format will beat any amount of PNG compression. Keep PNG for anything needing transparency or genuinely lossless pixels.