pnggif

Pure-Crystal PNG / APNG / GIF reader, and PNG / APNG writer.

It decodes an image file (or in-memory buffer) into RGBA bitmaps and, optionally, a downscaled terminal-cell map suitable for solid-block rendering in a TUI (one cell per sampled pixel, the cell background set to the pixel color).

Ported from Blessed's bundled tng.js (MIT).

Features

• PNG: all color types (grayscale, RGB, palette, gray+alpha, RGBA), bit depths 1–16, all scanline filters, Adam7 interlacing, palette transparency. Decompression uses Crystal's stdlib Compress::Zlib.
• APNG: animation frames with dispose/blend compositing.
• GIF (87a/89a): ported LZW decoder, interlacing, transparency, NETSCAPE loop counts, multi-frame animation.
• JPEG / other: converted to PNG via ImageMagick convert when available.
• Cellmaps: nearest-neighbour downscaling with non-square-cell aspect correction (cell_aspect, default 2.0) so images don't look stretched.
• Encoding: still PNG and animated APNG written in-process, no external tools (see below).

Usage

require "pnggif"

img = PNGGIF::PNG.new("picture.png", cell_width: 40)

img.width        # => image pixel width
img.height       # => image pixel height
img.bmp          # => Array(Array(PNGGIF::Pixel)), full-resolution RGBA
img.cellmap      # => downscaled bitmap, one PNGGIF::Pixel per terminal cell
img.frames       # => Array(PNGGIF::Frame)? for animations, else nil

# Pre-composited animation frames ({cellmap, delay_ms}):
if frames = img.animation_cellmaps(40)
  frames.each { |cellmap, delay_ms| ... }
end

Each PNGGIF::Pixel has r, g, b, a channels (0–255).

Encoding

The shard also writes PNG and APNG, pure-Crystal (stdlib Compress::Zlib and Digest::CRC32 only):

bytes = PNGGIF.encode_png(bmp)                  # bmp : Array(Array(PNGGIF::Pixel))
bytes = PNGGIF.encode_apng(frames, num_plays: 0) # frames : Array({Bitmap, delay_ms})

Both have (…, io : IO) overloads for writing straight to a file or stream. Output is always truecolor + alpha (color type 6, 8-bit), matching the in-memory Bitmap format, so no quantization happens. encode_apng emits a spec-correct animation (acTL/fcTL/fdAT); the first frame doubles as the still IDAT, so non-APNG viewers show it. num_plays: 0 loops forever. Round-trips through the shard's own decoder (PNGGIF::PNG#frames).

Current limitations, chosen for simplicity:

• Not streaming: encode_apng takes all frames up front (acTL carries the frame count and precedes the frames). Long recordings buffer every raw bitmap in memory; a streaming variant would need a seekable IO to patch acTL, or buffering of the (much smaller) compressed chunks.
• No inter-frame optimization: every frame is a full-canvas fdAT with scanline filter 0 (dispose NONE / blend SOURCE), and all frames must match the first frame's canvas size. Files are therefore larger than what optimizing encoders (e.g. ffmpeg) produce via sub-rectangle diffs and filter selection.
• Duplicate frames are the caller's job: identical consecutive frames are encoded again in full. When capturing mostly-static content, fold repeats by extending the previous frame's delay_ms instead of appending a frame.

There is no GIF or JPEG encoder — for those, pipe frames to an external encoder instead.

License

AGPLv3