What’s New in Subtitle Edit 5
Subtitle Edit 5 is the Avalonia-based, cross-platform version of Subtitle Edit. It keeps the familiar subtitle editing workflow from the Windows Forms version, but many features were expanded.
- Cross-platform Avalonia UI for Windows, Linux, and macOS — macOS is a new supported platform in SE 5.
- Cleaner, High-DPI-aware UI that scales correctly on modern displays.
- Follow system theme (light, dark, etc.) automatically, or pick a theme manually.
- New Flatpak packaging work for Linux.
- Many new settings, and the settings dialog has a built-in search to quickly find any option.
- Native pick-folder dialog wherever a folder is needed (was missing in the WinForms 4.x line).
- Update check settings — pick the stable or beta channel, and get a passive notification at startup when a newer build is out.
- Proxy settings — domain, system credentials and a bypass list, used by every download and online engine.
- Type-to-search in every combo box, a default save location setting, and the subtitle file name in dialog title bars.
Editing and Grid
- Show formatting in grid — formatting tags (italic, bold, color, etc.) can now be rendered visually in the subtitle grid.
- Tag coloring in the subtitle text box — HTML and ASSA tags are syntax-highlighted while you type, using a native text box so centering, IME/CJK input, right-to-left text, live spell check (red underlines, suggestions, add to dictionary, ignore all) and the full context menu all keep working with color tags enabled.
- Edit controls Show/Hide/Duration are now optional and can be toggled on/off.
- Deleting many lines at once in the subtitle grid / list view is dramatically faster.
- New Tools → Change formatting dialog for adding or removing italic, bold, underline, and other formatting across selected lines.
- New Tools → Merge two subtitles tool that combines two subtitles (or the loaded subtitle’s text + translation) into one bilingual subtitle. Output as SubRip (overlapping pairs stacked as line 1 / line 2) or ASSA with two configurable styles (font, color, outline, shadow, top/bottom alignment) and a live preview.
- AI assistant for the current line — ask a local AI model about the selected line or ask it for a change (fix errors, fit reading speed, more formal/casual, or a free-form request), from the text box context menu or the edit-box toolbar. Runs via llama.cpp, Ollama, or any OpenAI-compatible endpoint.
- Edit original — the original/reference column can be edited in place, and its non-matching lines are shown as reference rows in the grid.
- Hide tags — a third grid formatting mode that hides override tags so tag-heavy ASSA lines read as plain text; show formatting now shows the text of tag-heavy lines too.
- Layout 10 puts the edit box under the waveform, like SE 4 and Aegisub.
- New Tools → Remove/replace Unicode characters (the SE 4 plugin, built in), Sentence case in the casing options, and the SE 4 Minimum gap frame rate calculator.
- Multiple replace grew up — import/export of selected rule categories, move rules up/down/to top/bottom, select all/none/invert in the preview, the whole rule shown while editing, remembered expanded categories, and a toolbar button.
- Merge lines: keep end time (allow overlap with the next subtitle) as an option, and Show selected lines earlier/later in the grid context menu.
- Auto-break settings under Settings → Tools, with a do-not-break-after list editor and a bottom-heavy percentage.
- Keyboard parity with SE 4: Go to first/last line with video sync, Go to next empty line, toggle custom tags, recalculate duration, Ctrl+Shift+Home/End selection, Shift+Backspace as forward delete, and Copy/Paste (alternative) text box shortcuts for clipboard managers.
Spell Check
- Live spell check in the subtitle grid — misspelled words are underlined in the grid, and the grid context menu offers suggestions, Add to user dictionary and Ignore all (when a single line is selected).
See Spell Check for details.
Video
- New Video → Re-encode tool that re-encodes a video into a format better suited for subtitling work.
- New Video → Cut video tool for trimming video segments directly from Subtitle Edit.
- Video → Burn-in with logo — a logo/watermark image can now be included when burning subtitles into video.
- Many new speech-to-text engines (see Speech to Text below).
- New Video → OCR burned-in subtitle tool that extracts hardcoded subtitles from video via OCR (see Video OCR).
- Improved reading of subtitles embedded in MP4 files.
- PGS position monitor — the binary/image-based edit window has a BDSup2Sub-style position map for image-based subtitles (
.sup/.sub/BDN): every subtitle’s real rectangle on the video canvas, color coded by zone (active picture / letterbox bars), with configurable content aspect ratio, custom bar height, and title-safe margin.
- New Video → More → Chapters editor with Matroska, MP4 and OGM chapter formats (see Chapters).
- Margin is part of the subtitle area — the preview subtitle may use the letterbox bars, keeping a translation clear of burned-in forced narrative.
- Burn-in gains Apple VideoToolbox hardware encoders on macOS,
.webm/.ts output, letter spacing, and lists only the encoders and containers the OS and codec actually support.
- Matroska track chooser shows an image-subtitle preview, the number of forced cues, and can export VobSub directly.
- Go to sub position and pause and Go to video position… shortcuts, and a toggle-subtitles shortcut for the video player.
Sync
- Visual Sync now includes a waveform, making it easier to pick precise sync points.
- Visual sync and Set sync point show the subtitle on the video, and Visual sync finds or opens a video when only a subtitle is loaded.
- Waveform toolbar buttons can be customized, sorted, imported, and exported.
- Waveform themes can be imported and exported.
- Spectrogram style can be changed at runtime — no re-generation needed.
- More customization options for the waveform and spectrogram, including colors, shot-change colors, and visual style.
- Snap to shot changes is consistent across every way of doing it — dragging an edge, dragging a whole subtitle, and the snap shortcuts all land the cue the Beautify time codes profile’s in/out cues gap away from the cut. Dragging snaps when the cue looks close (a pixel distance, so it feels the same at any zoom), and the snap distances are now in Options → Settings → Waveform. Shift while dragging temporarily disables the snap.
Beautify Time Codes
- New Tools → Beautify time codes… brings the SE 4 beautifier across, but as a live tool: two stacked waveform visualizers (original / beautified) show the result before you accept it, with prev/next navigation, frame and millisecond deltas, and a per-cue reason line (snapped to shot change · min. gap enforced · min. duration enforced, etc.).
- The full profile editor (zones, chaining, connected-subtitle handling, per-cue gap, presets for Netflix and SDI) is available from the tool window and from Options → Settings → Waveform. Profile edits persist into
Settings.json.
- Beautify time codes is also a batch convert step, so a whole folder can be snapped to shot changes and frames in one run.
OCR
Beyond the classic engines, image-to-text now includes AI-based local and online options:
- CrispEmbed — a new local ggml-based OCR engine (same family as CrispASR) with GLM-OCR, GOT-OCR2 and Qwen3-VL backends, downloaded and managed by Subtitle Edit.
- llama.cpp vision models — OCR through a server-managed llama.cpp instance, also available in Batch Convert and
seconv.
- More engines overall: Tesseract, nOCR, Binary OCR, Google Lens, Google Vision, Ollama, Mistral OCR, and Paddle OCR (standalone or Python).
- Show only forced subtitles — a filter plus a Forced column, so only the forced lines are OCR’ed and returned (SE 4 parity).
- Histogram-based color isolation for VobSub/DVD subtitles, making it easier to separate text from borders and background.
- The spell-check dictionary is auto-selected from the OCR language, and the OCR fix replace lists are applied even when no Hunspell dictionary is installed.
- Train nOCR from SE 4 is back, and nOCR matching is faster thanks to a result cache.
- HunyuanOCR 1.5 joins the llama.cpp OCR model list, with install-status dots for engines and models and an engine download/update button in the settings.
- Save all images with HTML index from the OCR window, and CrispEmbed’s hardware build can be re-picked after the first install.
See OCR and Video OCR for details.
Speech to Text
Speech recognition is no longer limited to classic Whisper workflows. Subtitle Edit 5 includes a broader set of local and downloadable engines:
- Purfview Faster-Whisper XXL, CTranslate2, Whisper.cpp (with cuBLAS and Vulkan backends on Windows), OpenAI Whisper, OpenAI-compatible STT, and Const-me’s Whisper.
- Qwen3 ASR with multiple GGUF model sizes.
- Crisp ASR variants including GLM, Qwen3, Granite, Omni, Parakeet, Canary, Cohere, Fire Red, Mega, and Kyutai.
- Forced-aligner picker (built-in / Canary CTC / Qwen3 / 12 language-specific wav2vec2 aligners) for word-level timestamps.
- Per-engine advanced parameters and batch transcription improvements.
- Automatic language selection for several newer engines.
- Voxtral backend for Crisp ASR, a way to switch VAD off, and an automatic retry without VAD when a clip comes back empty.
- Batch speech to text can include the language code in the output file names.
See Speech to Text for the current engine list and workflow.
Text to Speech
Text to speech now includes more local and cloud engines:
- Edge-TTS.
- Mistral TTS.
- Qwen3 TTS with downloadable local server builds and models.
- Kokoro TTS with downloadable local server builds and models.
- OmniVoice TTS - a local CPU engine (many languages, voice cloning) with downloadable models.
- Review audio clips, regenerate individual lines, keep regeneration history, and export generated clips with metadata.
- IndexTTS 2.5 — a local engine with emotion and speaking-rate control.
- Find voices in video and clone them all — auto-cast every speaker from the video, or clone the voice of a single line from the video.
- Chatterbox in 23 languages (plus F16 and Q4_K model variants), and the CosyVoice3 RL talker models.
See Text to Speech for details.
Auto-translate
Subtitle Edit 5 adds local, downloadable auto-translate engines that run entirely on your own machine:
- Server-managed llama.cpp — Subtitle Edit downloads llama.cpp, manages a local
llama-server process, and offers a curated TranslateGemma model picker, so no manual server setup is required. CPU, Vulkan, and CUDA builds are available, and the server can be started and stopped from the Auto-translate window.
- CrispASR MADLAD — a local MADLAD-based translation engine with downloadable models (shown with size and install status), available in both the Auto-translate window and Batch Convert.
- OpenAI Compatible API — a generic engine for any service exposing an OpenAI-compatible
chat/completions endpoint (vLLM, KoboldCpp, a llama.cpp server on another machine, cloud providers, …).
- MiLMMT-46 translation models in the llama.cpp engine, and a llama.cpp advanced engine with custom prompt, server parameters, a stall watchdog and a token cap.
- Completion-format prompts for LM Studio, KoboldCpp, Ollama and the other local engines, a reset-to-default button for the prompt, and the chosen languages are kept when the engine changes.
See Auto-translate for the full engine list and workflow.
AI Review
- New Tools → AI review — an AI proofreading pass that catches typos, spelling, grammar, punctuation, and casing errors without rephrasing or changing meaning, tone, or style.
- Runs locally by default — uses a server-managed llama.cpp engine with a downloadable model picker, so no cloud service or API key is required. Ollama and any OpenAI-compatible endpoint are also supported.
- Review before you apply — suggestions are listed as before/after pairs with a per-line reason, grouped by category (Spelling, Grammar, Punctuation, Casing, Other) with filter chips. Tick the ones you want and apply only those.
- Safe by design — formatting tags (
<i>, {\an8}, etc.) and line breaks are preserved; suggestions that touch tags are dropped, and large rewrites are flagged for a closer look and left unselected.
- Editable prompt — the instructions sent to the model can be customized (with the subtitle language auto-detected and substituted in).
- Play current, Select none, and a Start/Stop server button so the local model’s VRAM can be released without leaving the window; the server is also stopped when a review is cancelled. AI review is in the grid’s Selected lines context menu too.
See AI Review for details.
Batch conversion
- OCR while converting — Batch Convert can turn image-based subtitles into text-based formats in bulk, using nOCR, Binary OCR, Tesseract, Ollama, llama.cpp vision models, or PaddleOCR. Language and pixels-are-space settings can be auto-detected for nOCR/Binary OCR, so converting many files with similar fonts needs far less manual setup.
- Local auto-translate in the queue — the new local engines (server-managed llama.cpp / TranslateGemma and CrispASR MADLAD) can be applied directly as a batch conversion step, fully offline.
- More chainable functions — including the new Change formatting (add/remove italic, bold, underline, etc.) alongside the existing fixes, replacements, casing, time-code, gap, merge, and split operations.
- Speech-to-text batch mode — transcribe many media files at once and save the results next to the source files.
- Optimized MKV parsing — reading subtitle tracks from Matroska (
.mkv) files is significantly faster, speeding up batch jobs that extract subtitles from many video containers.
- Beautify time codes, Convert colors to dialog, Snap time codes to frames, and the CrispEmbed OCR and llama.cpp advanced translate engines as batch steps; the llama.cpp launch flags are yours to set.
- Add folder…, a filtered file count, translation progress, and an option to keep the source file’s date/time on converted files (also in
seconv).
See Batch Convert, OCR, and Command Line (seconv).
- New Apply advanced effects tool that generates cinematic and creative ASSA override-tag animations (typewriter, karaoke, bounce-in, neon, glitch, rainbow, starfield, rain, snow, fireflies, and more) with real-time video preview.
- Hide layer — individual ASSA layers can now be hidden in the preview to focus on the lines you are working on.
- ASSA filtering — filter and search lines in the ASSA grid by style, actor, layer, or tag content.
- Font collector — shows the fonts a script uses and whether they are available, copies them to a folder, and can embed them in the subtitle with unused glyphs trimmed (see Styles and Attachments).
- More advanced effects: Word flip 3D, Lower third and Cinematic title.
- Added EBU-TT-D (read and write) — the TTML distribution profile used by European broadcasters (BBC iPlayer, ARD/ZDF, NPO) and HbbTV.
- Added IMSC-Rosetta Timed Text subtitle format support.
- New IMSC 1.1 image profile export — a single self-contained TTML file with base64 PNG subtitles (
smpte:image) for streaming and broadcast image-subtitle delivery, from File → Export and from the image/binary subtitle editor.
- New DVD sup (MuxMan/Scenarist) image-based export — the classic DVD-Video subpicture
.sup that DVD authoring tools import.
- New Import CSV/XLSX with custom columns window for spreadsheets that don’t fit the standard layout — pick which columns map to start, end, text, etc.
- Teletext / EBU STL — an alignment dialog, a TT grid column, a teletext color picker with a No color option, and the position the file carries is used when previewing on the video (also for TTML and PAC).
- New exports: Final Cut Pro XML Captions, DaVinci Resolve marker EDL (import too), Audacity/Tenacity labels, BDN/xml 8-bit with palette-indexed PNGs, and the Full frame image option is back.
- New imports: Adobe Premiere Pro Markers panel CSV, and XSUB subtitles inside
.avi files.
- D-Cinema interop properties dialog, and the option to remove blank lines when opening a subtitle is back from SE 4.
- Abbreviation lists for 30 more languages.
Command Line (seconv)
The seconv headless converter now lives in the main Subtitle Edit repository — it builds, ships, and updates in lockstep with the desktop app.
- Polished terminal UI — colored output with progress per file, summary tables, and a
--json mode for CI pipelines and scripting.
- Cross-platform — runs on Windows, Linux, and macOS with only the .NET runtime; no display or GUI required, suitable for servers and Docker.
- Broader feature set — additional time and cleanup operations, OCR engine selection (Tesseract / nOCR / Binary OCR / Ollama / PaddleOCR), container input from
.mkv / .mp4 / .mcc, info and lint subcommands for inspection, custom output templates, and POSIX-style flag names (legacy SE 4.x flags still work).
.avi/XSUB input, rule selection for --remove-formatting, and keeping the source file’s date/time on output.
See Command Line (seconv) for usage and examples.
Where to Look Next