Image Compressor

Files are routed by extension and come back in the same format: JPG through mozjpeg, PNG through oxipng (lossless), WebP through libwebp and AVIF through libaom — each format's reference encoder compiled to WebAssembly. A file the encoder cannot make smaller is returned as your original bytes and the card reads "already optimized", so a batch is never made worse. GIF, BMP and TIFF are not accepted here; convert them first.

Balanced is the right default — for photographs the difference from the original is effectively invisible at normal viewing sizes. Use High Quality when images will be inspected closely (portfolio, product zoom) and Max Compression when speed beats fidelity (thumbnails, chat, email).

PNG is lossless — quality presets don't apply to it, so savings come only from re-encoding efficiency and are typically modest. For big PNG wins, convert photographic PNGs to WebP or JPG instead (one click away in Related tools); keep PNG for graphics that need sharp edges or transparency.

NV Toolkit's compressor never returns a larger file: when the re-encoded output is not smaller than the input, the original bytes are kept and the card is marked "optimal". You can only win or break even — which is why it is safe to run an already-optimised folder through it.

Close to it. Switch to the Target Size tab, pick a byte budget (20/50/100/200/500 KB or a custom value), and the tool binary-searches the JPEG/WebP/AVIF quality setting for the highest quality whose output still fits under your target — it always aims slightly under, never over, so a 100 KB target typically lands at 95–100 KB rather than exactly 100.000 KB. PNG has no quality lever (it's lossless here), so target size on a PNG runs a single optimization pass and reports whether that already met the target.

Only as a last resort. The search tries quality alone first; if even the lowest usable quality still produces a file larger than your target — a very small KB budget on a large photo — the tool steps the dimensions down (to roughly 85%, 70%, 55%, then 40% of the original) and re-runs the quality search at each size until one fits. You'll see this happen on aggressive targets like 20 KB for a multi-megapixel photo; ordinary web-size images usually hit the target on quality alone.

The tool returns its smallest achievable result — the lowest quality at the smallest tried scale — and marks the card so you can see the target wasn't quite reached, rather than silently returning something misleadingly labeled. In practice this only happens with extreme targets (a few KB) on large or highly detailed source photos.