Features

• Use FFmpeg 8.1.
• Add setting Constant Rate Factor: -crf.
• Add --no-cache to main program.
• Add -encoders to info command.
• Build whisper-cpp for Windows AArch64.

Fixes

• Implement Neon SIMD for readChunk proc (audio). Makes audio analysis faster by 10% for AArch64.
• Use SvtAv1 4.1.0.

Full Changelog: 30.0.0...30.1.0

This is a big release with big changes. The most important being going forward, certain releases may require a license key to work. While these are the same license keys used in the proprietary GUI, the auto-editor cli (and all of its components) remains open source. I call this the FOSSIL model, FOSS + (I)ntegrated (L)icense keys, and I'll explain more of the details in an upcoming blog post.

Major

• If you have purchased an Auto-Editor license, you may now use multiple inputs auto-editor a.mp4 b.mp4 and they will be concatenated into a single output, with silence detection applied to each input independently.
• Will no longer open media files by default. Use the --open flag to restore the old behavior.
• Will no longer handle options with underscores. Use dashes instead. e.g. --sample_rate -> sample-rate
• Add --smooth for setting mincut and minclip respectively. It will apply them with equal precedence. --edit audio:mincut=,minclip= is no supported.

Features

• Add --set-action. Allows adding any action anywhere, and is preferred over video/silent speed.
• whisper: Add --language, --translate, --threads.

Fixes

• whisper: enable BLAS on MacOS for speedup when not using GPU.
• whisper: Downscale samplerate to 16k for better accuracy.
• Fix wrong colors when encoding HDR content with hevc_videotoolbox.
• Improve audio rendering performance for single channel streams.
• Improve audio rendering for exotic layouts (do: 7.1 -> 7.1 instead of: 7.1 -> stereo -> 7.1).

Full Changelog: 29.8.1...30.0.0

Fixes

• Remove "Cuda" builds because they were too big and bloated. You can compile whisper-cpp's cli with
  CUDA support if you really need it.
• Switch to config.nims so that the correct flags are always passed no matter what nimble command is used.
• Fix --scale not rounding to nearest 2 when most video formats require it.
• --preview: Fix crash when there are no audio clips in the timeline.
• Dramatically reduce peak memory usage for media rendering by optimizing audio code

Full Changelog: 29.8.0...29.8.1

Features

• Add fcp7 importer
• Use a constant sort algorithm to improve --preview performance
• Increase number of cache files limit
• Add completion command
• Implement 'zoom' and 'invert' actions.

Fixes

• Add -w:0 and --when-inactive as aliases for --when-silent.
• Add -w:1 and --when-active as aliases for --when-normal.
• Disable some obscure codecs/muxers

Full Changelog: 29.7.0...29.8.0

Features

• Build Windows AArch64 with all packages besides whisper-cpp.
• Use FFmpeg's AVIndexEntry feature for more efficient video seeking.
• Improve clip-sequence export, avoiding resource exhaustion for many clips outputs.

Fixes

• Add -e, a short for --edit.
• Add option hints for zsh completions.
• Wrap help text based on current terminal window size.

Fixes

• Allow disabling the libvpl package when building
• Cuda vs Non-cuda binaries for Linux
• Better zsh completions

Full Changelog: 29.6.1...29.6.2

Fixes

• Support hardware-accelerated QSV encoders and decoders available for discrete (Arc) and integrated (Xe) Intel GPUs.
• Fix importing and exporting v2 timeline formats.
• If auto-editor encounters a bad video frame, use the last frame instead of erroring out.
• svtav1 1.6.0 -> 1.6.2
• dav1d 1.5.2 -> 1.5.3

Full Changelog: 29.6.0...29.6.1

Features

• Linux Aarch64 is now available as an official binary.
• --edit will now look for external subtitle files if the input file doesn't have enough.
• Whisper is now built on all platforms.
• Cuda for whisper can now be built as an opt-in feature for Linux.
• NOTE: Cuda Whisper for Windows cannot be built because NVIDIA does not provide MingW binaries. An alternative is to use the Linux binary in WSL.

Fixes

• Use the last video frame instead of inserting a black frame.
• Fixed 6 month regression where and function performed 'logical or' instead of 'logical and'.
• Fixed varispeed action causing misalignments in FCP7 and FCP11.
• Whisper command: properly escape files with colons and backslashes.

Full Changelog: 29.5.0...29.6.0

Features

• Kdenlive: Support the varispeed effect
• Shotcut: Support the varispeed and volume effects
• Whisper is now built by default on MacOS
• Whisper: support splitting words via --split-words
• --cut is now an alias for --cut-out
• --keep is now an alias for --add-in

Fixes

• Upgrade opus 1.5.2 -> 1.6
• Video rendering: Improve worst case performance with a better seeking algorithm

Full Changelog: 29.4.0...29.5.0

Features

• Can now render subtitles
• Will now copy attachments

Fixes

• Upgrade ffmpeg 8.0 -> 8.0.1
• Upgrade dav1d 1.5.1 -> 1.5.2

Full Changelog: 29.3.1...29.4.0