Images

Compress, resize, convert, rotate, crop, strip metadata, make an icon set, read the text: what Raycast's TinyPNG and Image Modification do, with the tools on the machine first and TinyPNG only when you give it a key. One input palette: the rows are the images at hand, every operation is an action on a row, and every result is written next to its source with a suffix and its path copied.

What is at hand

Before you type, the palette lists:

With more than one image an All N images row leads, and every operation on it runs over all of them; tab (or x with nothing typed) marks rows for the same. A typed path lists a file (~/Desktop/a.png), a folder's images (~/Desktop/site/, with its own All row) or the completions of the last segment; anything else filters the rows by name. Each row shows a thumbnail, the size and the pixel size; the detail pane (open by default) shows the picture at 256 px over the info: dimensions, format, colour space and profile, camera, lens, exposure, date, location (the last five from exiftool when it is installed; sips knows make, model, copyright and artist).

The operations

Every operation ends the same way: the result's path on the clipboard (the paths, one per line, for a batch; the image itself when the input was the clipboard, so a screenshot compresses and pastes), the panel down, and the HUD saying what happened and with what: "Compressed photo.png: 1.4 MB → 312 KB (−78%), pngquant · path copied". A batch that takes longer than eight seconds answers with a progress toast and finishes behind the panel; the HUD then sums it up. With replace on the result takes the source's place and the source is copied to ~/Library/Caches/pal/images/originals/ first, so Restore original always has something to restore; nothing is ever deleted. A clipboard image's result goes to ~/Library/Caches/pal/images/clipboard/ (the source is pal's history file).

Which tool does it

The tools setting is the order the encoders are tried; one left out is never used. Every result says which tool made it. On macOS sips (in every install) does the geometry (resize, rotate, flip, crop, pad, grayscale) whenever it is listed, and converts to PNG, JPEG, TIFF, GIF, BMP, PDF, HEIC, AVIF and ICO; ImageMagick (magick, or convert on version 6) does all of that elsewhere and is the fallback for everything. The compressors:

A source an encoder cannot read is decoded to PNG first (sips, ImageMagick, dwebp, avifdec, heif-convert); SVG is rasterised to PNG. The empty listing names up to three encoders worth installing when none of the sources are there. TinyPNG (tinypng_api_key, from tinypng.com/developers, 500 compressions a month free) takes PNG, JPEG, WebP and AVIF and posts the file to api.tinify.com.

Settings

[extensions.images]:

How it is made

ops.ts is pure: the formats, the output naming, plan(job, input, opts) answering the steps ({ tool, argv }) for a job with the tools at hand, the built-in JPEG and PNG strip (APPn and COM segments; the text, time, EXIF and ICC chunks), and the parsers for sips -g all, identify -format and exiftool -j. exec.ts finds the binaries, runs the steps (a replace goes through a temp file, so a failed step leaves the source whole), keeps the thumbnails, reads the Finder selection, writes the clipboard, talks to TinyPNG and keeps the originals. index.ts is the rows, the levels, the view and the picks. The tests draw their own PNGs (host/test/extensions/images-png.ts) and turn one into a JPEG through sips or ImageMagick.

What it does not do

• Batch across folders recursively, or watch a folder.
• Edit pixels beyond the list above (no filters, no annotations, no background removal).
• Read RAW files: sips can, but the palette lists the common formats only.
• A per-file quality: quality is one setting; run again with another.

Platforms

macOS (sips in every install; the encoders from Homebrew) and Linux (ImageMagick and the encoders from the package manager; the Finder selection is macOS only, the clipboard and typed paths work everywhere).