Subtitle Edit 5.1 is the first feature release after 5.0. A summary of the changes since 5.0.0:

• AI review (Tools > AI review) - proofread subtitles with a local LLM via llama.cpp or Ollama, with a before/after grid and an editable prompt

• AI assistant in the subtitle text box - quick actions like fix spelling/grammar, reading speed and change tone

• Auto-translate: new generic OpenAI-compatible engine, remote llama.cpp server option, DeepL formality picker, and many new local models (TranslateGemma, Gemma 4, Hy-MT2, Qwen 3.5/3.6, Aya Expanse)

• Auto-translate: retry with backoff on rate-limited (429) LLM engines, show the API URL in error details, and fix engines that ignored a custom URL

• Speech to text: new engines and models - MOSS-Transcribe-Diarize with speaker labels, OpenRouter and Alibaba Qwen3-ASR online engines, and Parakeet, Kyutai, SenseVoice, ARK-ASR and Cohere models - plus live progress for all CrispASR engines, and the input language is remembered again

• Text to speech: new MOSS-TTS engine with zero-shot voice cloning and a target language dropdown, new Q6_K/Q8_0 quants, a big window overhaul, and a large bug-fix pass across the engines and the review window

• OCR: new CrispEmbed local OCR engine (GLM-OCR, GOT-OCR2, Qwen3-VL), llama.cpp OCR vision models in Batch convert and seconv, a "Show only forced subtitles" filter, and histogram-based color isolation for VobSub/DVD subtitles

• New formats: EBU-TT-D (read/write), IMSC 1.1 image profile export, and "DVD sup (MuxMan/Scenarist)" image-based export

• macOS: multi-window support - "File -> New window" opens an independent editor window

• New "file saved" prompt with file info, copy path and Play/Open - used across text to speech and the video tools

• Waveform: SE4-style center/lock scrubbing (paused and while playing), configurable toolbar with seek and play buttons, drag a multi-line selection to shift time codes together

• Subtitle text box: new native tag-coloring text box - IME/CJK input, right-to-left and live spell check now work with color tags enabled

• Live spell check in the subtitle grid context menu, spell check window fixes, MS Word engine fixes, and new Georgian, Armenian, Luxembourgish and Faroese dictionaries

• Fix common errors: reworked fix list with category chips, sortable columns and live counts - and unticked fixes are no longer applied

• Shortcuts window: categories with icons, sortable columns, better search - plus many new SE4-parity shortcuts (column shortcuts, set start and go to next, play previous/next, extend, and more)

• Subtitle grid: type-a-number line navigation, alternating row colors, text fit setting (clip/wrap/ellipsis), and "Insert subtitle after current line..." keeping the inserted file's own time codes

• Auto-save of the open subtitle file (Options > General)

• Right-to-left: mirrored grid columns, content direction in text boxes, and an ASSA karaoke RTL option

• Accessibility: menu bar activation via Alt/F10, and named controls for screen readers across the app

• Image-based subtitle editor: PGS position monitor, adjust color for selected lines, change speed, and HTML index export

• Point sync and visual sync: sorted sync points, re-pointing, delete via menu/keyboard, and a time code box for the sync point

• Split/re-balance long lines: respect the single line max length

• Auto-trim white space: use the detected language, so Dutch keeps " 's"

• Network: always bypass the proxy for local (loopback) URLs, and proxy password/crash fixes

• Performance: non-English start-up about 3.5x faster, 10-25x faster auto line-breaking, much faster subtitle grid and waveform rendering, faster .srt/ASSA loading and MP4 parsing, and a wide allocation-cutting pass (source-generated regexes, reused StringBuilders) - thx ivandrofly

• Stability: hundreds of fixes, including a 46-bug codebase sweep, an async/await bug hunt, and hardened binary subtitle parsers

• Command line (seconv): llama.cpp auto-translate and OCR, image output styling options, fix-common-errors profile settings in --settings JSON, VobSub/.idx fixes, and clear errors on malformed SCC input

• The libse core library is now published on NuGet

• New UI languages: Arabic, Turkish, Persian, Traditional Chinese, Indonesian, Vietnamese, Greek and Thai - and all language files synced with English (3,000+ strings)

• Updated engines: CrispASR v0.8.24, llama.cpp b10142 (with a CUDA 13 build on Windows), whisper.cpp v1.9.1, CrispEmbed v0.16.1, Qwen3 ASR CPP v0.1.7 and yt-dlp 2026.07.04

  A big thank you to everyone who tested the pre-releases, reported issues and contributed code and translations - and a special thanks to Anthropic for sponsoring a Claude subscription used in the development of this release :)