Changelog

We are operating with semantic versioning.

v14.0.0

Major:

  • Drop FFmpeg 6.

  • Remove AVError alias (use FFmpegError directly instead).

  • Remove the template arg from OutputContainer.add_stream().

Features:

v13.1.0

Features:

Fixes:

  • Remove the deprecation module in anticipation of PEP 702.

  • Add type stubs to previously unannotated API sections.

  • Improve type stubs for both mypy and mypy –strict.

  • Permit only setting time_base with a Fraction, as mypy is unable to respect different types in getters vs setters.

  • Declare write_packet function as const by @hmaarrfk in (#1517).

v13.0.0

Major:

  • Drop FFmpeg 5, Support FFmpeg 7.

  • Drop Python 3.8, Support Python 3.13.

  • Update FFmpeg to 7.0.2 for the binary wheels.

  • Disallow initializing an AudioLayout object with an int.

  • Disallow accessing gop_size, timebase as a decoder (Raised deprecation warning before).

  • Remove ticks_per_frame property because it was deprecated in FFmpeg.

Features:

  • Add AttachmentStream class.

  • Add best() method to StreamContainer.

  • Add set_audio_frame_size() method to Graph object.

  • Add nb_channels property to AudioLayout object.

  • Add from_bytes() method to VideoFrame object.

Fixes:

  • Fix VideoCC’s repl breaking when self._format is None.

  • Fix getting pix_fmt property when VideoCC’s self._format is None.

v12.3.0

Features:

Fixes:

  • Include libav headers in final distribution by @materight in (#1455).

  • Fix segfault when calling subtitle_stream.decode() by @WyattBlue in (#1460).

  • flushing subtitle decoder requires a new uninitialized packet by @moonsikpark in (#1461).

  • Set default color range for VideoReformatter.format() by @elxy in (#1458).

  • Resampler: format, layout accepts str int too by @WyattBlue in (#1446).

v12.2.0

Features:

  • Add a make_writable method to Frame instances (#1414).

  • Use av_guess_sample_aspect_ratio to report sample and display aspect ratios.

Fixes:

  • Fix a crash when assigning an AudioLayout to AudioCodecContext.layout (#1434).

  • Remove a circular reference which caused AudioSampler to occupy memory until garbage collected (#1429).

  • Fix more type stubs, remove incorrect __init__.pyi.

v12.1.0

Features:

  • Build binary wheels with webp support.

  • Allow disabling logs, disable logs by default.

  • Add bitstream filters by @skeskinen in (#1379 #1375).

  • Expose CodecContext flush_buffers by @skeskinen in (#1382).

Fixes:

v12.0.0

Major:

  • Add type hints.

  • Update FFmpeg to 6.1.1 for the binary wheels.

  • Update libraries for the binary wheels (notably dav1d to 1.4.1).

  • Deprecate VideoCodecContext.gop_size for decoders by @JoeSchiff in (#1256).

  • Deprecate frame.index by @JoeSchiff in (#1218).

Features:

Fixes:

  • Convert deprecated Cython extension class properties to decorator syntax by @JoeSchiff.

  • Check None packet when setting time_base after decode by @philipnbbc in (#1281).

  • Remove deprecated Buffer.to_bytes by @WyattBlue in (#1286).

  • Remove deprecated Packet.decode_one by @WyattBlue in (#1301).

v11.0.0

Major:

  • Add support for FFmpeg 6.0, drop support for FFmpeg < 5.0.

  • Add support for Python 3.12, drop support for Python < 3.8.

  • Build binary wheels against libvpx 1.13.1 to fix CVE-2023-5217.

  • Build binary wheels against FFmpeg 6.0.

Features:

  • Add support for the ENCODER_FLUSH encoder flag (#1067).

  • Add VideoFrame ndarray operations for yuv444p/yuvj444p formats (#788).

  • Add setters for AVFrame.dts, AVPacket.is_keyframe and AVPacket.is_corrupt (#1179).

Fixes:

  • Fix build using Cython 3 (#1140).

  • Populate new streams with codec parameters (#1044).

  • Explicitly set python_requires to avoid installing on incompatible Python (#1057).

  • Make AudioFifo.__repr__ safe before the first frame (#1130).

  • Guard input container members against use after closes (#1137).

v10.0.0

Major:

  • Add support for FFmpeg 5.0 and 5.1 (#817).

  • Drop support for FFmpeg < 4.3.

  • Deprecate CodecContext.time_base for decoders (#966).

  • Deprecate VideoStream.framerate and VideoStream.rate (#1005).

  • Stop proxying Codec from Stream instances (#1037).

Features:

  • Update FFmpeg to 5.1.2 for the binary wheels.

  • Provide binary wheels for Python 3.11 (#1019).

  • Add VideoFrame ndarray operations for gbrp formats (#986).

  • Add VideoFrame ndarray operations for gbrpf32 formats (#1028).

  • Add VideoFrame ndarray operations for nv12 format (#996).

Fixes:

  • Fix conversion to numpy array for multi-byte formats (#981).

  • Safely iterate over filter pads (#1000).

v9.2.0

Features:

  • Update binary wheels to enable libvpx support.

  • Add an io_open argument to av.open for multi-file custom I/O.

  • Add support for AV_FRAME_DATA_SEI_UNREGISTERED (#723).

  • Ship .pxd files to allow other libraries to cimport av (#716).

Fixes:

  • Fix an ImportError when using Python 3.8/3.9 via Conda (#952).

  • Fix a muxing memory leak which was introduced in v9.1.0 (#959).

v9.1.1

Fixes:

  • Update binary wheels to update dependencies on Windows, disable ALSA on Linux.

v9.1.0

Features:

  • Add VideoFrame ndarray operations for rgb48be, rgb48le, rgb64be, rgb64le pixel formats.

  • Add VideoFrame ndarray operations for gray16be, gray16le pixel formats (#674).

  • Make it possible to use av.open() on a pipe (#738).

  • Use the “ASS without timings” format when decoding subtitles.

Fixes:

  • Update binary wheels to fix security vulnerabilities (#921) and enable ALSA on Linux (#941).

  • Fix crash when closing an output container an encountering an I/O error (#613).

  • Fix crash when probing corrupt raw format files (#590).

  • Fix crash when manipulating streams with an unknown codec (#689).

  • Remove obsolete KEEP_SIDE_DATA and MP4A_LATM flags which are gone in FFmpeg 5.0.

  • Deprecate to_bytes() method of Packet, Plane and SideData, use bytes(packet) instead.

v9.0.2

Minor:

  • Update FFmpeg to 4.4.1 for the binary wheels.

  • Fix framerate when writing video with FFmpeg 4.4 (#876).

v9.0.1

Minor:

  • Update binary wheels to fix security vulnerabilities (#901).

v9.0.0

Major:

  • Re-implement AudioResampler with aformat and buffersink (#761). AudioResampler.resample() now returns a list of frames.

  • Remove deprecated methods: AudioFrame.to_nd_array, VideoFrame.to_nd_array and Stream.seek.

Minor:

  • Provide binary wheels for macOS/arm64 and Linux/aarch64.

  • Simplify setup.py, require Cython.

  • Update the installation instructions in favor of PyPI.

  • Fix VideoFrame.to_image with height & width (#878).

  • Fix setting Stream time_base (#784).

  • Replace deprecated av_init_packet with av_packet_alloc (#872).

  • Validate pixel format in VideoCodecContext.pix_fmt setter (#815).

  • Fix AudioFrame ndarray conversion endianness (#833).

  • Improve time_base support with filters (#765).

  • Allow flushing filters by sending None (#886).

  • Avoid unnecessary vsnprintf() calls in log_callback() (#877).

  • Make Frame.from_ndarray raise ValueError instead of AssertionError.

v8.1.0

Minor:

  • Update FFmpeg to 4.3.2 for the binary wheels.

  • Provide binary wheels for Python 3.10 (#820).

  • Stop providing binary wheels for end-of-life Python 3.6.

  • Fix args order in Frame.__repr__ (#749).

  • Fix documentation to remove unavailable QUIET log level (#719).

  • Expose codec_context.codec_tag (#741).

  • Add example for encoding with a custom PTS (#725).

  • Use av_packet_rescale_ts in Packet._rebase_time() (#737).

  • Do not hardcode errno values in test suite (#729).

  • Use av_guess_format for output container format (#691).

  • Fix setting CodecContext.extradata (#658, #740).

  • Fix documentation code block indentation (#783).

  • Fix link to Conda installation instructions (#782).

  • Export AudioStream from av.audio (#775).

  • Fix setting CodecContext.extradata (#801).

v8.0.3

Minor:

  • Update FFmpeg to 4.3.1 for the binary wheels.

v8.0.2

Minor:

  • Enable GnuTLS support in the FFmpeg build used for binary wheels (#675).

  • Make binary wheels compatible with Mac OS X 10.9+ (#662).

  • Drop Python 2.x buffer protocol code.

  • Remove references to previous repository location.

v8.0.1

Minor:

  • Enable additional FFmpeg features in the binary wheels.

  • Use os.fsencode for both input and output file names (#600).

v8.0.0

Major:

  • Drop support for Python 2 and Python 3.4.

  • Provide binary wheels for Linux, Mac and Windows.

Minor:

  • Remove shims for obsolete FFmpeg versions (#588).

  • Add yuvj420p format for VideoFrame.from_ndarray() and VideoFrame.to_ndarray() (#583).

  • Add support for palette formats in VideoFrame.from_ndarray() and VideoFrame.to_ndarray() (#601).

  • Fix Python 3.8 deprecation warning related to abstract base classes (#616).

  • Remove ICC profiles from logos (#622).

Fixes:

v7.0.1

Fixes:

  • Removed deprecated AV_FRAME_DATA_QP_TABLE_* enums. (#607)

v7.0.0

Major:

  • Drop support for FFmpeg < 4.0. (#559)

  • Introduce per-error exceptions, and mirror the builtin exception hierarchy. It is recommended to examine your error handling code, as common FFmpeg errors will result in ValueError baseclasses now. (#563)

  • Data stream’s encode and decode return empty lists instead of none allowing common API use patterns with data streams.

  • Remove whence parameter from InputContainer.seek() as non-time seeking doesn’t seem to actually be supported by any FFmpeg formats.

Minor:

Fixes:

  • Fix typing in CodecContext.parse() and make it more robust.

  • Fix wrong attribute in ByteSource. (#340)

  • Remove exception that would break audio remuxing. (#537)

  • Log messages include last FFmpeg error log in more helpful way.

  • Use AVCodecParameters so FFmpeg doesn’t complain. (#222)

v6.2.0

Major:

  • Allow av.open() to be used as a context manager.

  • Fix compatibility with PyPy, the full test suite now passes. (#130)

Minor:

Build:

  • Do not install the scratchpad package.

v6.1.2

Micro:

v6.1.1

Micro:

Build:

  • Fix more compiler warnings.

v6.1.0

Minor:

  • av.datasets for sample data that is pulled from either FFmpeg’s FATE suite, or our documentation server.

  • InputContainer.seek() gets a stream argument to specify the time_base the requested offset is in.

Micro:

  • Avoid infinite look in Stream.__getattr__. (#450)

  • Correctly handle Python I/O with no seek method.

  • Remove Datastream.seek override (#299)

Build:

  • Assert building against compatible FFmpeg. (#401)

  • Lock down Cython language level to avoid build warnings. (#443)

Other:

  • Incremental improvements to docs and tests.

  • Examples directory will now always be runnable as-is, and embeded in the docs (in a copy-pastable form).

v6.0.0

Major:

  • Drop support for FFmpeg < 3.2.

  • Remove VideoFrame.to_qimage method, as it is too tied to PyQt4. (#424)

Minor:

Micro:

Build:

  • Remove the “reflection” mechanism, and rely on FFmpeg version we build against to decide which methods to call. (#416)

  • Fix many more const warnings.

v0.x.y

Note

Below here we used v0.x.y.

We incremented x to signal a major change (i.e. backwards incompatibilities) and incremented y as a minor change (i.e. backwards compatible features).

Once we wanted more subtlety and felt we had matured enough, we jumped past the implications of v1.0.0 straight to v6.0.0 (as if we had not been stuck in v0.x.y all along).

v0.5.3

Minor:

Patch:

  • Fix Packet.time_base handling during flush. (#398)

  • VideoFrame.reformat() can throw exceptions when requested colorspace transforms aren’t possible.

  • Wrapping the stream object used to overwrite the pix_fmt attribute. (#390)

Runtime:

  • Deprecate VideoFrame.ptr in favour of VideoFrame.buffer_ptr.

  • Deprecate Plane.update_buffer() and Packet.update_buffer in favour of Plane.update(). (#407)

  • Deprecate Plane.update_from_string() in favour of Plane.update(). (#407)

  • Deprecate AudioFrame.to_nd_array() and VideoFrame.to_nd_array() in favour of AudioFrame.to_ndarray() and VideoFrame.to_ndarray(). (#404)

Build:

Docs:

  • Docstrings for many commonly used attributes. (#372 and #409)

v0.5.2

Build:

  • Fixed Windows build, which broke in v0.5.1.

  • Compiler checks are not cached by default. This behaviour is retained if you source scripts/activate.sh to develop PyAV. (#256)

  • Changed to PYAV_SETUP_REFLECT_DEBUG=1 from PYAV_DEBUG_BUILD=1.

v0.5.1

Build:

  • Set PYAV_DEBUG_BUILD=1 to force a verbose reflection (mainly for being installed via pip, which is why this is worth a release).

v0.5.0

Major:

  • Dropped support for Libav in general. (#110)

  • No longer uses libavresample.

Minor:

  • av.open has container_options and stream_options.

  • Frame includes pts in repr.

Patch:

  • EnumItem’s hash calculation no longer overflows. (#339, #341 and #342.)

  • Frame.time_base was not being set in most cases during decoding. (#364)

  • CodecContext.options no longer needs to be manually initialized.

  • CodexContext.thread_type accepts its enums.

v0.4.1

Minor:

  • Add Frame.interlaced_frame to indicate if the frame is interlaced. (#327 by @MPGek)

  • Add FLTP support to Frame.to_nd_array(). (#288 by @rawler)

  • Expose CodecContext.extradata for codecs that have extra data, e.g. Huffman tables. (#287 by @adavoudi)

Patch:

  • Packets retain their refcount after muxing. (#334)

  • Codec construction is more robust to find more codecs. (#332 by @adavoudi)

  • Refined frame corruption detection. (#291 by @Litterfeldt)

  • Unicode filenames are okay. (#82)

v0.4.0

Major:

  • CodecContext has taken over encoding/decoding, and can work in isolation of streams/containers.

  • Stream.encode returns a list of packets, instead of a single packet.

  • AudioFifo and AudioResampler will raise ValueError if input frames inconsistent pts.

  • time_base use has been revisited across the codebase, and may not be converted between Stream.time_base and CodecContext.time_base at the same times in the transcoding pipeline.

  • CodecContext.rate has been removed, but proxied to VideoCodecContext.framerate and AudioCodecContext.sample_rate. The definition is effectively inverted from the old one (i.e. for 24fps it used to be 1/24 and is now 24/1).

  • Fractions (e.g. time_base, rate) will be None if they are invalid.

  • InputContainer.seek and Stream.seek will raise TypeError if given a float, when previously they converted it from seconds.

Minor:

  • Added Packet.is_keyframe and Packet.is_corrupt. (#226)

  • Many more time_base, pts and other attributes are writable.

  • Option exposes much more of the API (but not get/set). (#243)

  • Expose metadata encoding controls. (#250)

  • Expose CodecContext.skip_frame. (#259)

Patch:

  • Build doesn’t fail if you don’t have git installed. (#184)

  • Developer environment works better with Python3. (#248)

  • Fix Container deallocation resulting in segfaults. (#253)

v0.3.3

Patch:

  • Fix segfault due to buffer overflow in handling of stream options. (#163 and #169)

  • Fix segfault due to seek not properly checking if codecs were open before using avcodec_flush_buffers. (#201)

v0.3.2

Minor:

  • Expose basics of avfilter via Filter.

  • Add Packet.time_base.

  • Add AudioFrame.to_nd_array to match same on VideoFrame.

  • Update Windows build process.

Patch:

  • Further improvements to the logging system. (#128)

v0.3.1

Minor:

  • av.logging.set_log_after_shutdown renamed to set_print_after_shutdown

  • Repeating log messages will be skipped, much like ffmpeg’s does by default

Patch:

  • Fix memory leak in logging system when under heavy logging loads while threading. (#128 with help from @mkassner and @ksze)

v0.3.0

Major:

  • Python IO can write

  • Improve build system to use Python’s C compiler for function detection; build system is much more robust

  • MSVC support. (#115 by @vidartf)

  • Continuous integration on Windows via AppVeyor. (by @vidartf)

Minor:

  • Add Packet.decode_one() to skip packet flushing for codecs that would otherwise error

  • StreamContainer for easier selection of streams

  • Add buffer protocol support to Packet

Patch:

  • Fix bug when using Python IO on files larger than 2GB. (#109 by @xxr3376)

  • Fix usage of changed Pillow API

Known Issues:

  • VideoFrame is suspected to leak memory in narrow cases on Linux. (#128)

v0.2.4

  • fix library search path for current Libav/Ubuntu 14.04. (#97)

  • explicitly include all sources to combat 0.2.3 release problem. (#100)

v0.2.3

Warning

There was an issue with the PyPI distribution in which it required Cython to be installed.

Major:

  • Python IO.

  • Aggressively releases GIL

  • Add experimental Windows build. (#84)

Minor:

  • Several new Stream/Packet/Frame attributes

Patch:

  • Fix segfault in audio handling. (#86 and #93)

  • Fix use of PIL/Pillow API. (#85)

  • Fix bad assumptions about plane counts. (#76)

v0.2.2

  • Cythonization in setup.py; mostly a development issue.

  • Fix for av.InputContainer.size over 2**31.

v0.2.1

  • Python 3 compatibility!

  • Build process fails if missing libraries.

  • Fix linking of libavdevices.

v0.2.0

Warning

This version has an issue linking in libavdevices, and very likely will not work for you.

It sure has been a long time since this was released, and there was a lot of arbitrary changes that come with us wrapping an API as we are discovering it. Changes include, but are not limited to:

  • Audio encoding.

  • Exposing planes and buffers.

  • Descriptors for channel layouts, video and audio formats, etc..

  • Seeking.

  • Many many more properties on all of the objects.

  • Device support (e.g. webcams).

v0.1.0

  • FIRST PUBLIC RELEASE!

  • Container/video/audio formats.

  • Audio layouts.

  • Decoding video/audio/subtitles.

  • Encoding video.

  • Audio FIFOs and resampling.