image optimizer for web

pingo a fast image optimizer for web

pingo is an experimental lossless and lossy image optimizer (PNG, JPEG, WebP, APNG) designed to be used for web context — it does image reductions over iterations by performing its own heuristics to target the image content. since most of features are not fully implemented, it would be recommended to do backup or use only on test files

• a6 — download (64-bit, win)
• ! alpha state ! for v1.00 notes
• if you want to support this work, consider making a donation

pingo — optimizing 400 PNG files losslessly for web

usage (examples)

pingo [options] files/folders

pingo -s0 *.png

pingo -s0 myfolder

pingo -s0 c:\myfolder

PNG — lossless

pingo -sN file.png

the lossless PNG optimizer in pingo does heuristic ordered transforms to make the data more compressible losslessly for web. it performs an highly optimized deflation according to the user-level. see also the PNG lossless benchmark — -sa or -sb do more trials, -fast=N does faster encoding

JPEG — lossless

pingo -sN file.jpg

the JPEG lossless optimizer in pingo is just like mozjpeg with some automatic transform on image data — rotation according to the orientation flag. instead of using the automatic lossless profile, user could add jpgtype=N to force the encoding

• -jpgtype=0, baseline: fast encode and fast decode, legacy compatible but usually do less savings
• -jpgtype=1, progressive: average encode speed but slower decode, usually better savings than baseline
• -jpgtype=2, progressive with more scans: slow encode, slower decode, usually best savings

WebP — lossless

the WebP lossless optimizer uses a tweaked libwebp — it could try its own heuristics to determine transformations. this implementation is incomplete at this moment: it makes sense to try it on paletted samples where it uses a different way to sort colors

APNG — lossless

the APNG optimizer would try to re-create the animation by using some transformations. this implementation is incomplete at this moment: it could not be able to select an optimal colortype or bitdepth

PNG — lossy

pingo would let the user to try several lossy transformations for PNG. it makes sense to use those transformations from high quality source — not to apply those several times on the same file

-auto or -auto=N

N = 1 to 100 as quality factor: 1 = very low, 100 = best effort. this would attempt to optimize the file automatically, by picking the lossy transformation (eventually lossless) by itself. this implementation is incomplete at this moment: it uses only fast heuristics which could be wrong

-pngfilter or -pngfilter=N

N = 1 to 100 as quality factor: 1 = low, 100 = near-lossless. this would try to transform image data using predictors. would be efficient on image with lot of colors / noise, and less efficient on low colored images

-pngpalette or -pngpalette=N

N = 1 to 100 as quality factor: 1 = very low, 100 = best effort. this would try to transform image data by palettization. would be efficient on low colored images, and could have limitations for high colored ones

-pngcolor or -pngcolor=N

N = 1 to 100 as quality factor: 1 = low, 100 = near-lossless. this would try to reduce image data using color transformation. this usually get the closest near-lossless result if used with highest value, but could do less savings; could be also useful on moderately colored images

JPEG — lossy

the JPEG lossy uses sjpeg — it does 4:4:4 by default, with automatic settings according to the sample. eventually, user could set the following flag:

-jpgquality=N

N = 1 to 100 as quality factor: 1 = low, 100 = best effort. without any other flag, it should not try to do subsampling, unless source is already 4:2:0 JPEG

-jpgsub

this would enable the JPEG subsampling but by using a sharp filter on high quality source. this attempt could offer a better rendering than regular subsampling

WebP — lossy/auto

the WebP lossy optimizer in pingo uses sharp_yuv by default on high quality source. it could give different results from the reference encoder — e.g. image pre-processing on PNG with alpha, automatic compression, etc.

-webp-lossy or -webp-lossy=N

N = 1 to 100 as quality factor: 1 = low, 100 = best effort

-webp or -webp=N

N = 1 to 100 as quality factor: 1 = low, 100 = best effort. just like -auto, it would pick a transform according to the image data, including a lossless one if it seems more suitable. this implementation is incomplete at this moment: it uses only fast heuristics which could be wrong

-webp-palette or -webp-palette=N

N = 1 to 100 as quality factor: 1 = low, 100 = best effort. just like -pngpalette, this would try to transform image data by palettization

WebP — near-lossless

if -webp-lossy has some limitations (YUV420) regarding quality, one could try a near-lossless transform instead. pingo includes several way to do this according to the sample

-webp-near or -webp-near=N

N = 1 to 100 as quality factor: 1 = low, 100 = best quality. this is a slighlty tweaked version from the one in libwebp

-webp-color or -webp-color=N

N = 1 to 100 as quality factor: 1 = low, 100 = best quality. use color transformation: it should be faster than -webp-near but could be less efficient on various image type. provide different quality/size ratio — not always better

-webp-filter or -webp-filter=N

N = 1 to 100 as quality factor: 1 = low, 100 = best quality. use predictors to transform data: only efficient on image with lot of colors, probably less efficient than -webp-color

-webp-nigh or -webp-nigh=N

just like -auto, but would pick a near-lossless transform automatically according to the sample. this implementation is incomplete at this moment: it uses only fast heuristics which could be wrong

-fast=N

N = 1 to 4. -fast=1 works with -sN and could be combined with other lossy transforms (as grayscale, resize, pngfilter, etc.) and affects (A)PNG, JPEG or WebP. 2, 3 and 4 are only for PNG lossless encoding

image transformations

color profiles

pingo converts image data to sRGB by perceptual intent from the color profile if a lossy transformation is done, whatever the format — it could be disabled by using -nosrgb. once converted, the color profile would be removed from the optimized file. this would also convert CMYK JPG

PNG

JPEG

WebP

-strip

even without using -strip, pingo would remove most of non-critical chunks/segments from PNG/JPG except some, as colors profile, which could be useful or affect rendering. -strip would be more strict, and should leave only critical data

-resize=N

N = 24 to 1920. a downscaler that resizes PNG, JPEG, or WebP from the highest value of width/height and keep ratio (from height only for APNG). value must be inferior to the image — resizing is always done *with* an additional targeted lossy transform

PNG

JPEG

WebP

-enhance or -enhance=N

N = 1 to 8 as factor: 1 = low, 8 = high sharpness. increase image details in some area while transforming image data to make it more compressible. this implementation is incomplete at this moment: it uses only fast heuristics

-grayscale

convert image data to grey level. used with -sN in non-automatic mode, it should try to convert image with no loss over the transform. used in automatic mode, it would try to apply loss over the grey level conversion

PNG

JPEG

WebP

APNG

-notrans

transform image data to make it non-transparent by colors blending. in automatic mode, this could convert PNG to JPG (could be disabled by adding -sN). user could set -notrans=N N N, where N is R, G, and B color value from 0 to 255 (default is 255,255,255 (white))

-svgz or -svgz=N

N = 1 to 100 as quality factor: 1 = low, 100 = near-lossless. convert PNG images with lot of colors to compressed SVG container — this have to be served accordingly, but could be used in most of browsers (~2011). could be useful where recent codecs are not supported. -nocompression creates SVG files instead of SVGZ