WebP Python bindings

Installation

pip install webp

Requirements

• libwebp (tested with v1.0.3)
  • Install libwebpmux and libwebpdemux components as well.
  • Check out the Dockerfile for steps to build from source on Ubuntu.
• Python 3 (tested with v3.6)
• cffi
• Pillow
• numpy

Usage

import webp

Simple API

# Save an image
webp.save_image(img, 'image.webp', quality=80)

# Load an image
img = webp.load_image('image.webp', 'RGBA')

# Save an animation
webp.save_images(imgs, 'anim.webp', fps=10, lossless=True)

# Load an animation
imgs = webp.load_images('anim.webp', 'RGB', fps=10)

If you prefer working with numpy arrays, use the functions imwrite, imread, mimwrite, and mimread instead.

Advanced API

# Encode a PIL image to WebP in memory, with encoder hints
pic = webp.WebPPicture.from_pil(img)
config = WebPConfig.new(preset=webp.WebPPreset.PHOTO, quality=70)
buf = pic.encode(config).buffer()

# Read a WebP file and decode to a BGR numpy array
with open('image.webp', 'rb') as f:
  webp_data = webp.WebPData.from_buffer(f.read())
  arr = webp_data.decode(color_mode=WebPColorMode.BGR)

# Save an animation
enc = webp.WebPAnimEncoder.new(width, height)
timestamp_ms = 0
for img in imgs:
  pic = webp.WebPPicture.from_pil(img)
  enc.encode_frame(pic, timestamp_ms)
  timestamp_ms += 250
anim_data = enc.assemble(timestamp_ms)
with open('anim.webp', 'wb') as f:
  f.write(anim_data.buffer())

# Load an animation
with open('anim.webp', 'rb') as f:
  webp_data = webp.WebPData.from_buffer(f.read())
  dec = webp.WebPAnimDecoder.new(webp_data)
  for arr, timestamp_ms in dec.frames():
    # `arr` contains decoded pixels for the frame
    # `timestamp_ms` contains the _end_ time of the frame
    pass

Features

• Picture encoding/decoding
• Animation encoding/decoding
• Automatic memory management
• Simple API for working with PIL.Image objects

Not implemented

• Encoding/decoding still images in YUV color mode
• Advanced muxing/demuxing (color profiles, etc.)
• Expose all useful fields

Developer notes

Running tests

The CircleCI local CLI should be used to run tests in an isolated environment:

$ circleci local execute

Known issues

• An animation where all frames are identical will "collapse" in on itself, resulting in a single frame. Unfortunately, WebP seems to discard timestamp information in this case, which breaks webp.load_images when the FPS is specified.