pylibheif

Python bindings for libheif using pybind11.

Features

• HEIC/HEIF Support: Read and write HEIC images (HEVC/H.265 encoded)
• AVIF Support: Read and write AVIF images (AV1 encoded)
• JPEG2000 Support: Read and write JPEG2000 images in HEIF container
• NumPy Integration: Zero-copy access to image data via Python Buffer Protocol
• Metadata Support: Read and write EXIF, XMP, and custom metadata
• RAII Resource Management: Automatic resource cleanup with context managers

Supported Formats

Requirements

• Python >= 3.12
• NumPy >= 1.26.0
• CMake >= 3.15
• C++17 compatible compiler

System Dependencies

# macOS
brew install openjpeg

# Ubuntu/Debian
sudo apt install libopenjp2-7-dev

Installation

pip install pylibheif

Or with uv:

uv pip install pylibheif

Building from Source

# Clone with submodules
git clone --recursive https://github.com/twn39/pylibheif.git
cd pylibheif

# Install
uv pip install -e .

Usage

Reading HEIC/AVIF Images

Using context manager (recommended):

import pylibheif
import numpy as np

# Open HEIC file with context manager
with pylibheif.HeifContext() as ctx:
    ctx.read_from_file('image.heic')
    
    # Get primary image handle
    handle = ctx.get_primary_image_handle()
    print(f'Image size: {handle.width}x{handle.height}')
    print(f'Has alpha: {handle.has_alpha}')
    
    # Decode to RGB
    img = handle.decode(pylibheif.HeifColorspace.RGB, 
                        pylibheif.HeifChroma.InterleavedRGB)
    
    # Get as NumPy array (zero-copy)
    plane = img.get_plane(pylibheif.HeifChannel.Interleaved, False)
    arr = np.asarray(plane)  # shape: (height, width, 3)

Explicit creation (for more control):

import pylibheif
import numpy as np

# Create context explicitly
ctx = pylibheif.HeifContext()
ctx.read_from_file('image.heic')

handle = ctx.get_primary_image_handle()
img = handle.decode(pylibheif.HeifColorspace.RGB, 
                    pylibheif.HeifChroma.InterleavedRGB)
plane = img.get_plane(pylibheif.HeifChannel.Interleaved, False)
arr = np.asarray(plane)

# Resources are automatically freed when objects go out of scope

Writing HEIC Images (H.265)

import pylibheif
import numpy as np

# Create image from NumPy array
width, height = 1920, 1080
img = pylibheif.HeifImage(width, height, 
                          pylibheif.HeifColorspace.RGB,
                          pylibheif.HeifChroma.InterleavedRGB)
img.add_plane(pylibheif.HeifChannel.Interleaved, width, height, 8)

# Fill with data
plane = img.get_plane(pylibheif.HeifChannel.Interleaved, True)
arr = np.asarray(plane)
arr[:] = your_image_data  # your RGB data

# Encode and save as HEIC
ctx = pylibheif.HeifContext()
encoder = pylibheif.HeifEncoder(pylibheif.HeifCompressionFormat.HEVC)
encoder.set_lossy_quality(85)
encoder.encode_image(ctx, img)

ctx.write_to_file('output.heic')

Writing AVIF Images (AV1)

import pylibheif
import numpy as np

# Prepare image (same as above)
width, height = 1920, 1080
img = pylibheif.HeifImage(width, height, 
                          pylibheif.HeifColorspace.RGB,
                          pylibheif.HeifChroma.InterleavedRGB)
img.add_plane(pylibheif.HeifChannel.Interleaved, width, height, 8)

# Encode and save as AVIF
ctx = pylibheif.HeifContext()

# Use AV1 format for AVIF
encoder = pylibheif.HeifEncoder(pylibheif.HeifCompressionFormat.AV1)
encoder.set_lossy_quality(85)
encoder.set_parameter("speed", "6") # Optional: Tune speed (0-9)

encoder.encode_image(ctx, img)

# Save with .avif extension
ctx.write_to_file('output.avif')

Writing JPEG Images

import pylibheif
import numpy as np

# Prepare image (same as above)
width, height = 1920, 1080
img = pylibheif.HeifImage(width, height, 
                          pylibheif.HeifColorspace.RGB,
                          pylibheif.HeifChroma.InterleavedRGB)
img.add_plane(pylibheif.HeifChannel.Interleaved, width, height, 8)

# Encode and save as JPEG
ctx = pylibheif.HeifContext()

# Use JPEG format
encoder = pylibheif.HeifEncoder(pylibheif.HeifCompressionFormat.JPEG)
encoder.set_lossy_quality(90) # Quality 0-100

encoder.encode_image(ctx, img)

# Save with .jpg extension
ctx.write_to_file('output.jpg')

Writing JPEG2000 Images

import pylibheif
import numpy as np

# Prepare image (same as above)
width, height = 1920, 1080
img = pylibheif.HeifImage(width, height, 
                          pylibheif.HeifColorspace.RGB,
                          pylibheif.HeifChroma.InterleavedRGB)
img.add_plane(pylibheif.HeifChannel.Interleaved, width, height, 8)

# Encode and save as JPEG2000 in HEIF container
ctx = pylibheif.HeifContext()

# Use JPEG2000 format
encoder = pylibheif.HeifEncoder(pylibheif.HeifCompressionFormat.JPEG2000)
encoder.set_lossy_quality(85) # Quality 0-100

encoder.encode_image(ctx, img)

# Save with .jp2 or .heif extension
# Note: libheif typically saves JPEG2000 in a HEIF container
ctx.write_to_file('output.heif')

Encoder Selection

By default, pylibheif selects the best available encoder for the requested format (e.g. x265 for HEVC). You can also explicitly select a specific encoder (e.g. Kvazaar) if available.

import pylibheif

# 1. Get all available HEVC encoders
descriptors = pylibheif.get_encoder_descriptors(pylibheif.HeifCompressionFormat.HEVC)

# Print available encoders
for d in descriptors:
    print(f"ID: {d.id_name}, Name: {d.name}")

# 2. Find specific encoder (e.g. Kvazaar)
kvazaar_desc = next((d for d in descriptors if "kvazaar" in d.id_name), None)

if kvazaar_desc:
    # 3. Create encoder explicitly using the descriptor
    encoder = pylibheif.HeifEncoder(kvazaar_desc)
    
    # Verify which encoder is used
    print(f"Using encoder: {encoder.name}")
    
    encoder.set_lossy_quality(85)
    # encoder.encode_image(...)
    # encoder.encode_image(...)

Reading Metadata

import pylibheif

ctx = pylibheif.HeifContext()
ctx.read_from_file('image.heic')
handle = ctx.get_primary_image_handle()

# Get metadata block IDs
exif_ids = handle.get_metadata_block_ids('Exif')
for id in exif_ids:
    metadata_type = handle.get_metadata_block_type(id)
    metadata_bytes = handle.get_metadata_block(id)
    print(f'Metadata type: {metadata_type}, size: {len(metadata_bytes)}')

Writing Metadata

import pylibheif
import numpy as np

# Create and encode an image
width, height = 64, 64
img = pylibheif.HeifImage(width, height,
                          pylibheif.HeifColorspace.RGB,
                          pylibheif.HeifChroma.InterleavedRGB)
img.add_plane(pylibheif.HeifChannel.Interleaved, width, height, 8)
plane = img.get_plane(pylibheif.HeifChannel.Interleaved, True)
arr = np.asarray(plane)
arr[:] = 128  # fill with gray

ctx = pylibheif.HeifContext()
encoder = pylibheif.HeifEncoder(pylibheif.HeifCompressionFormat.HEVC)
encoder.set_lossy_quality(85)
handle = encoder.encode_image(ctx, img)

# Add EXIF metadata (with 4-byte offset prefix for TIFF header)
exif_data = b'\x00\x00\x00\x00' + b'Exif\x00\x00' + b'II*\x00...'  # your EXIF data
ctx.add_exif_metadata(handle, exif_data)

# Add XMP metadata
xmp_data = b'''<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/">
  <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
    <rdf:Description rdf:about="" xmlns:dc="http://purl.org/dc/elements/1.1/">
      <dc:creator>My App</dc:creator>
    </rdf:Description>
  </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>'''
ctx.add_xmp_metadata(handle, xmp_data)

# Add custom/generic metadata
custom_data = b'{"app": "myapp", "version": "1.0"}'
ctx.add_generic_metadata(handle, custom_data, "json", "application/json")

# Save
ctx.write_to_file('output_with_metadata.heic')

API Reference

class pylibheif.HeifContext

Manages the valid lifetime of libheif context. It is the main entry point (root object) for high-level API.

Methods

__init__() Creates a new empty context.

read_from_file(filename: str) -> None Reads a HEIF file from the given filename.

• filename: Path to the HEIF file.

read_from_memory(data: bytes) -> None Reads a HEIF file from a bytes object.

• data: Bytes containing the file content.

write_to_file(filename: str) -> None Writes the current context to a file.

• filename: Destination path.

write_to_bytes() -> bytes Writes the current context to a bytes object.

• Returns: bytes object containing the encoded file data.

get_primary_image_handle() -> HeifImageHandle Gets the handle for the primary image in the file.

• Returns: HeifImageHandle for the primary image.

get_image_handle(id: int) -> HeifImageHandle Gets the handle for a specific image ID.

• id: The ID of the image (see get_list_of_top_level_image_IDs).
• Returns: HeifImageHandle.

get_list_of_top_level_image_IDs() -> List[int] Gets a list of IDs of all top-level images in the file.

• Returns: List of integer IDs.

add_exif_metadata(handle: HeifImageHandle, data: bytes) -> None Adds EXIF metadata to the specified image.

• handle: Image handle from encoding.
• data: Raw EXIF bytes (with 4-byte offset prefix for TIFF header).

add_xmp_metadata(handle: HeifImageHandle, data: bytes) -> None Adds XMP metadata to the specified image.

• handle: Image handle from encoding.
• data: XMP XML as bytes.

add_generic_metadata(handle: HeifImageHandle, data: bytes, item_type: str, content_type: str = "") -> None Adds generic/custom metadata to the specified image.

• handle: Image handle from encoding.
• data: Raw metadata bytes.
• item_type: Metadata item type (e.g. "json", "iptc").
• content_type: Optional MIME content type (e.g. "application/json").

class pylibheif.HeifImageHandle

Represents a compressed image within the HEIF file.

Properties

• width (int): The width of the image.
• height (int): The height of the image.
• has_alpha (bool): True if the image has an alpha channel.

Methods

decode(colorspace: HeifColorspace = HeifColorspace.RGB, chroma: HeifChroma = HeifChroma.InterleavedRGB) -> HeifImage Decodes the image handle into an uncompressed HeifImage.

• colorspace: Target colorspace (default: RGB).
• chroma: Target chroma format (default: InterleavedRGB).
• Returns: Decoded HeifImage.

get_metadata_block_ids(type_filter: str = "") -> List[str] Gets a list of metadata block IDs attached to this image.

• type_filter: Optional filter string (e.g. "Exif", "XMP").
• Returns: List of metadata ID strings.

get_metadata_block_type(id: str) -> str Gets the type string of a specific metadata block.

• id: Metadata ID.
• Returns: Type string (e.g. "Exif").

get_metadata_block(id: str) -> bytes Gets the raw data of a metadata block.

• id: Metadata ID.
• Returns: bytes object containing the metadata.

class pylibheif.HeifImage

Represents an uncompressed image containing pixel data. Supports the Python Buffer Protocol for zero-copy access with NumPy.

Properties

• width (int): The width of the image.
• height (int): The height of the image.

Methods

__init__(width: int, height: int, colorspace: HeifColorspace, chroma: HeifChroma) Creates a new empty image.

• width: Image width.
• height: Image height.
• colorspace: Image colorspace.
• chroma: Image chroma format.

add_plane(channel: HeifChannel, width: int, height: int, bit_depth: int) -> None Adds a new plane to the image.

• channel: The channel type (e.g. HeifChannel.Interleaved).
• width: Width of the plane.
• height: Height of the plane.
• bit_depth: Bit depth (e.g. 8).

get_plane(channel: HeifChannel, writeable: bool = False) -> HeifPlane Gets a plane object that supports the buffer protocol.

• channel: The channel to retrieve.
• writeable: Whether the buffer should be writable.
• Returns: HeifPlane object (wrappable with np.asarray()).

class pylibheif.HeifEncoder

Controls the encoding process.

Methods

__init__(format: HeifCompressionFormat) Creates a new encoder for the specified format.

• format: Compression format (e.g. HeifCompressionFormat.HEVC).

set_lossy_quality(quality: int) -> None Sets the quality for lossy compression.

• quality: Integer between 0 (lowest) and 100 (highest).

set_parameter(name: str, value: str) -> None Sets a low-level encoder parameter.

• name: Parameter name (e.g. "speed" for AV1).
• value: Parameter value.

encode_image(context: HeifContext, image: HeifImage, preset: str = "") -> HeifImageHandle Encodes the given image and appends it to the context.

• context: The destination HeifContext.
• image: The source HeifImage to encode.
• preset: Optional encoder preset (e.g. "ultrafast", "slow"). Default is empty (balanced/default). Note: This maps to the 'preset' parameter in libheif. It works for x265 (check version), but AOM and others may use different parameters (e.g. 'speed') which should be set via set_parameter instead.
• Returns: HeifImageHandle for the encoded image. Can be used to add metadata.

Enums

pylibheif.HeifColorspace

• RGB, YCbCr, Monochrome, Undefined

pylibheif.HeifChroma

• InterleavedRGB: Interleaved R, G, B bytes.
• InterleavedRGBA: Interleaved R, G, B, A bytes.
• C420: YUV 4:2:0 planar.
• C422: YUV 4:2:2 planar.
• C444: YUV 4:4:4 planar.
• Monochrome.

pylibheif.HeifChannel

• Interleaved: For interleaved RGB/RGBA.
• Y, Cb, Cr: For YUV planar.
• R, G, B: For RGB planar.
• Alpha: For Alpha channel.

pylibheif.HeifCompressionFormat

• HEVC: H.265 (libx265).
• AV1: AV1 (AOM/RAV1E/SVT).
• JPEG: JPEG.
• JPEG2000: JPEG 2000 (OpenJPEG).

Building from Source

# Clone with submodules
git clone --recursive https://github.com/your-username/pylibheif.git
cd pylibheif

# Build
uv pip install -e .

Performance

Performance

Benchmarks on 1280x720 RGB image (Apple Silicon), comparing encoders at Iso-Quality (approx. 29.2 dB PSNR) to ensure fair speed comparison.

Baseline: x265 (HEVC) at Quality 80 (Medium Preset).

Key Findings:

1. AOM AV1 Efficiency: Surprisingly, libaom (at speed 6) is the fastest encoder in this test, outperforming even the highly optimized Kvazaar HEVC encoder while maintaining excellent compression efficiency (smallest file size).
2. HEVC Choice: If you need HEVC, Kvazaar is significantly faster than x265 for similar quality.

----------------------------------------------------------------------------------
Test: 1280x720 RGB Image, 10 Rounds
----------------------------------------------------------------------------------
Name (time in ms)                     Mean            OPS
----------------------------------------------------------------------------------
test_benchmark_encode_aom_av1       115.64           8.65  (Q86)
test_benchmark_encode_kvazaar       125.19           7.96  (Q80)
test_benchmark_encode_x265          206.03           4.65  (Q80 / Limit)
----------------------------------------------------------------------------------

Run benchmarks yourself:

uv pip install pillow-heif pytest-benchmark
uv run pytest tests/test_benchmark.py --benchmark-only --benchmark-min-rounds=20

License

This project is licensed under the LGPL-3.0 License - see the LICENSE file for details.

Acknowledgments

• libheif - HEIF/AVIF codec library
• pybind11 - C++/Python bindings