2025 In Review

Before we dive into 2026, let's review what we achieved in 2025:

Python to Nim transition

In the beginning of the year, auto-editor was written 100% in Python. By mid-August, it's now written 100% in Nim (excluding end-to-end tests). This resulted in audio detection being ~6x faster and rendering performance increasing by ~2x. This large increase in speed surprised me, however, I think it's because Clang/GCC is better at emitting wide instructions and/or better at leveraging ffmpeg's existing multi-processing. Either way, I expect future versions of Python to shrink h this gap, unless new low-level optimizations or algorithmic improvements are applied.

FFmpeg contributions

This December, I have made two PRs to the FFmpeg project. Both deal with the whisper filter and make more useful to work with. At least one should be merged into master by early 2026.

Audio transcription via Whisper-Cpp

For the last release of the year, auto-editor 29.5.0, was released, which enables experimental support for Whisper on MacOS. Windows and Linux support will be coming in 2026.

Conclusion

Have a happy new year!

---