โ Farbfeld ยท Demo
Build a farbfeld v2 file
Drop an image or an animated GIF, get a .ff in the v2 format: lossless 8-bit by default, or near-lossless with a hard
error bound. Everything runs locally in a worker (ffimg.js); nothing is uploaded. The result is
decoded again right here and compared with your pixels, so you see whether it is exact.
Encode
independent tiles can be decoded alone or in parallel, at a cost of 2โ12% in size
Limits: up to 6 megapixels here. Colour-profiled images and semi-transparent pixels go through a canvas and can shift by a level or two before encoding; the verify line compares against what the canvas gave the encoder. For exact archival work or bigger images use the command line below.
Build it yourself
Same encoder, from a shell. Needs node and ImageMagick (for reading the source image).
ff2-encode.cjs ffimg.js test.cjs
node ff2-encode.cjs photo.png photo.ff # lossless v2 node ff2-encode.cjs --near 4 photo.png photo.ff # every channel within 4 node ff2-encode.cjs photo.png photo.ff photo.v1.ff # also write classic v1 for png2ff-style tools node ff2-encode.cjs anim.gif anim.ff # animation: frames, delays and loop count in one stream node ff2-encode.cjs --16 photo.png photo.ff # keep 16 bits per channel (lossless) node ff2-encode.cjs --tile 256 big.png big.ff # independent 256x256 tiles node ff2-encode.cjs --dpi 300 --cs p3 --meta title=Sunset photo.png photo.ff # optional metadata prefix node ff2-encode.cjs old.ff new.ff # classic v1 farbfeld in, 16-bit v2 out node test.cjs # self-test + corrupt-file fuzz
The encoder tries plain MED and the blended predictor, plus palette mode when the image has 256 colours or fewer, and keeps the smallest.