pngdeer

Smaller PNGs. Same picture. pngdeer is a lossy PNG compressor with no quality slider: for each image it searches palette sizes from 2 to 256 colors, with and without dithering, scores each result with a perceptual metric, and keeps the smallest file that still looks right. At the same quality, the median image comes out 8–12% smaller than with TinyPNG, the best result of any open-source compressor we've measured.

Try it in your browser → Nothing is uploaded: images are compressed on your device.

brew install heydeer/tap/pngdeer
pngdeer input.png -o output.png

• Picks the quality for you: about what TinyPNG would give the image, never above 90 on SSIMULACRA2's scale, which is indistinguishable at 1:1.
• Same bytes everywhere: the command line (macOS and Linux) and the browser (WebAssembly, on every core) run the same Rust code and write identical files.
• Big images too: up to 40 megapixels in the browser, more on the command line.
• JPEG photos too: re-encoded with mozjpeg at the lowest quality that keeps the target, 13% smaller in total than TinyPNG's at its quality (details).
• AVIF too: from PNGs and JPEG photos, 19% smaller in total than TinyPNG's AVIF at its quality; on the command line for now (details).
• Animated PNGs stay animated: every frame on one palette and at the target quality, timing kept; 28% of the original size on 12 animated emoji, about even to 16% smaller than TinyPNG (notes).
• Free and open source: MIT.

Leaderboard

We ran pngdeer, TinyPNG and eight open-source quantizers on 180 PNGs: icons, UI screenshots, illustrations, photos, cut-outs and 3D renders. Every output was scored against its original with SSIMULACRA2, a perceptual metric where 90 means indistinguishable at 1:1. TinyPNG is the yardstick:

• Points: median SSIMULACRA2 points above (+) or below (−) TinyPNG when a tool's file is as big as TinyPNG's.
• Reaches: images where a tool can match TinyPNG's quality at all.

Each cell gives the held-out set (71 images) / the tuning set (109 images):

Compressor	Points	Reaches
pngdeer	+1.7 / +2.0	67 / 106
TinyPNG (the yardstick)	0	71 / 109
pngquant	0.0 / +0.3	53 / 87
quantizr	−0.1 / −1.0	39 / 53
exoquant	−3.7 / −5.1	19 / 24
NeuQuant	−6.5 / −9.8	4 / 5
FFmpeg	−7.0 / −8.8	16 / 15
quantette	−7.6 / −9.6	18 / 21
ImageMagick	−11.3 / −14.5	3 / 8
Pillow	−16.8 / −19.5	4 / 4

At TinyPNG's quality, pngdeer's files are 12% smaller on the held-out set and 8% smaller on the tuning set (median image).

• Held out: pngdeer's parameters were tuned on the 109 images; the other 71 were collected afterwards and never used for tuning.
• Generous to the others: each open-source tool gets its best of ten settings per image, picked after the fact. pngdeer's results are what you get, searched at compression time.
• A second opinion: scored with butteraugli instead, which the search never looks at, pngdeer's files are still 6–9% smaller than TinyPNG's at the same quality.
• TinyPNG ran through its API in October 2026. bench/ reruns everything; the website screenshots in the sets aren't ours to share, so a rerun covers 155 of the 180 images.

WebP

pngdeer also writes WebP (an output name ending in .webp, or the switch in the browser): the smaller of its palette image stored losslessly and the lowest lossy quality that keeps the target (a JPEG photo: below). Against TinyPNG's WebP on 120 of the images, the size each tool needs to match TinyPNG's quality (median image / in total), and how many images it can bring up to that quality:

Compressor	Size at TinyPNG's quality	Reaches
pngdeer	95.0% / 93.9%	118
TinyPNG (the yardstick)	100%	120
cwebp, best of 27 settings	97.9% / 110.2%	120
pngquant + cwebp, best of 10	135.0% / 212.1%	115

• cwebp is libwebp's own encoder, the library Squoosh, sharp and Pillow encode with. It gets its best of 27 settings per image, picked after the fact: 13 qualities with and without the auto filter, and lossless. pngquant + cwebp stores pngquant's palette image as lossless WebP, TinyPNG's palette route from open-source parts.
• At TinyPNG's file size, on the 40 images where all three can be compared, pngdeer scores 2.6 SSIMULACRA2 points above TinyPNG, pngquant + cwebp 1.9, cwebp −0.3.
• Out of the box, cwebp's default quality makes files 6% bigger than TinyPNG's on the median image, and slightly worse. pngdeer's automatic quality makes them 5% smaller, as good.
• WebP's settings were chosen on these 120 images. On the other 60, against a stand-in that matches TinyPNG's WebP on these (same size and score, median), pngdeer is 4% smaller. The WebP notes have the details.

JPEG

pngdeer also takes JPEG photos (pngdeer photo.jpg -o smaller.jpg, or drop them in the browser) and writes a smaller JPEG: re-encoded with mozjpeg at the lowest quality that keeps the target, never bigger than the photo. On 60 photos from Wikimedia Commons, 30 web-size copies and 30 camera originals, the size each tool needs to match TinyPNG's quality (median photo / in total), and how many photos it can bring up to that quality:

Compressor	Size at TinyPNG's quality	Reaches
pngdeer	87.8% / 85.4%	60
TinyPNG (the yardstick)	100%	60
jpegli, best of 12 settings	89.2% / 87.2%	60
mozjpeg, best of 7 qualities	94.2% / 89.1%	59

• jpegli is Google's JPEG encoder from the JPEG XL project, mozjpeg Mozilla's, which Squoosh uses. Each gets its best setting per photo, picked after the fact: qualities 40–90, jpegli also at 4:2:0. pngdeer's files are pooled the same way (its encoder, mozjpeg with its MS-SSIM-tuned table at 4:2:0, at qualities 50–85, and its own runs), as in the PNG and WebP tables. The file it writes with TinyPNG's score as the target is 89.5% / 86.7% of TinyPNG's size.
• At TinyPNG's file size, pngdeer scores 3.3 SSIMULACRA2 points above TinyPNG on the median photo, jpegli 3.1, mozjpeg 1.8.
• At its automatic quality, pngdeer's files are 82% of TinyPNG's size on the median photo (84% in total), scoring about the same (median +0.2 points). mozjpeg's default quality makes them 7% bigger than TinyPNG's, 5 points better.
• Like TinyPNG, pngdeer gives big photos a lower quality than small ones: a large photo is seen scaled down. Unlike TinyPNG, it keeps the color profile (unless sRGB) and the orientation, and drops the rest of the camera's metadata.
• WebP from a JPEG (pngdeer photo.jpg -o photo.webp, or the switch in the browser): turned upright, at the same automatic quality. At TinyPNG's WebP quality it needs 95.3% / 91.3% of TinyPNG's WebP size; cwebp, given its best of 13 settings per photo, about the same. At equal quality it is no smaller than pngdeer's own JPEG (99% on the median photo), so it is there for sites that want WebP.
• The JPEG notes have the details. TinyPNG ran through its API in October 2026.

AVIF

pngdeer also writes AVIF (an output name ending in .avif; on the command line for now), from PNGs and JPEG photos: libaom's lowest quality that keeps the target as Chrome shows the file. On 54 images, 30 of the photos above and 24 PNGs of every kind, against TinyPNG's AVIF (median image / in total):

Compressor	Size at TinyPNG's quality	Smaller and no worse
pngdeer	79.7% / 80.5%	52 of 54
TinyPNG (the yardstick)	100%

• TinyPNG writes every AVIF at one fixed quantizer, so how good its files look follows the image (47 to 89 on SSIMULACRA2 here). pngdeer's automatic quality aims at TinyPNG's on the median: its files are 79% of TinyPNG's size on the median image (76% in total) and 0.7 points better, though better or worse than TinyPNG's image by image.
• Scored as browsers show the files: pngdeer's scores match Chrome's own decoding pixel for pixel, and TinyPNG's files were decoded by Chrome too.
• Small images are tried at four settings and the smallest file kept; large ones are searched at a fast speed, then encoded once or twice more slowly. libaom is built so that macOS and Linux, ARM and x86, write identical files.
• The AVIF notes have the details. TinyPNG ran through its API in October 2026.

How it works

• Searches the palette: palette sizes from 2 to 256 colors, each tried a few ways (with and without Floyd–Steinberg dithering, with neighbour smoothing, with a second palette builder), and the smallest file that reaches the target wins.
• Scores what you'd see: SSIMULACRA2, measured in-process. Colors are matched in Oklab, smooth gradients are always dithered, and transparent pixels are judged over light and dark backgrounds.
• Stays fast on big images: large images are searched on a downscaled preview and checked at full size.

The research notes follow every step, from the first comparison with TinyPNG to today's defaults.

Command line

pngdeer input.png -o output.png              # automatic quality
pngdeer input.png -o output.png --target 85  # a fixed SSIMULACRA2 quality
pngdeer input.png -o output.webp             # WebP, automatic quality
pngdeer photo.jpg -o photo.min.jpg           # JPEG, automatic quality
pngdeer photo.jpg -o photo.webp              # a JPEG as WebP
pngdeer input.png -o output.avif             # AVIF, from a PNG or a JPEG
pngdeer --help

From source: cargo build --release, and the binary is target/release/pngdeer.

Browser

The web app runs the same code compiled to WebAssembly. To run it locally (web/build.sh lists the toolchain it needs):

sh web/build.sh
python3 web/serve.py   # http://localhost:8770

License

MIT. Dependencies keep their own licenses (web/THIRD_PARTY_LICENSES.txt), as do the test and sample images (the SOURCES.md next to them). JPEG output is encoded with mozjpeg: this software is based in part on the work of the Independent JPEG Group. AVIF output is encoded with libaom.