LLM-powered subtitle translation for donghua & anime — built to protect the things that break.
Website · Download · Documentation
Polygluttony translates .ass subtitle files with the AI model you connect, while guarding against the ways machine translation usually goes wrong. Connect a model, open a folder and press one button: Polygluttony builds a glossary of names and terms, then translates, and you can watch every batch land as it happens.
- Line markers & partial-failure recovery — every line is tracked, so when a model drops, merges, or reorders lines, Polygluttony detects exactly where it broke and salvages the correct prefix instead of failing the whole batch.
- Drift detection — after each episode, a five-signal weighted check and an AI spot-check catch translations wandering off the source, and only the lines that drifted are retranslated.
- Your typesetting stays put — override tags like
{\pos}and{\an8}are lifted out before translating and put back after, and styles, timings and script info are copied from your file. Only the dialogue is translated. - Cross-episode glossary — for Chinese sources, a six-category glossary, with auto-detected world type (xianxia / wuxia / historical / modern), keeps names and terms consistent across a whole season. It's built as part of translating and grows with each new episode; edit it in place, with undo and earlier versions to go back to.
- Verification, not a score — every file checks its own work and lists the lines that need a look, never a number. Fix them in the episode's own window: retranslate a line, edit it, or correct the name in the glossary.
- Your choice of model — Anthropic, OpenAI, Google, a local model through Ollama or LM Studio, or any OpenAI- or Anthropic-compatible endpoint. One model for everything, or a different one per step.
Get the latest build from the releases page:
| OS | File |
|---|---|
| Windows 10/11 (x64) | Polygluttony-vX.Y.Z-windows.exe (installer, no admin rights needed) |
| macOS (Apple Silicon) | Polygluttony-vX.Y.Z-macos.zip |
| Linux x86_64, any distribution | Polygluttony-vX.Y.Z-linux.AppImage |
The builds aren't signed with a paid developer certificate, so the first launch takes one extra step on Windows and macOS: see Platform notes.
Everything lives in the wiki. New here? Start with the first two.
| Page | Covers |
|---|---|
| 🚀 Getting started | Install, connect a model, translate your first episode |
| 💻 Platform notes | The first launch on Windows, macOS and Linux |
| 🧭 How it works | What happens between opening a folder and getting translated files |
| 🔌 Connecting a model | Providers, API keys, local models, one model per step, tuning |
| ⚖️ Choosing a model | Which model for which job, and what it costs you |
| 📁 Folders & files | What gets read, what gets written, what happens to your tags |
| 🌐 Languages | The 15 languages, and what's different about Chinese → English |
| 📖 The glossary | Keeping names and terms consistent across episodes |
| Running a translation and watching an episode live | |
| 🎭 Tones | Standard, xianxia, wuxia, comedic and funny |
| ✅ Reading the results | What "Needs a look" means, and what to do about it |
| ✏️ Custom prompts | Changing exactly what the AI is told |
| 🛡️ What it protects | The ways AI translation goes wrong, and how they're caught |
| 🔒 Privacy & your data | What gets sent where, and where everything is stored |
| 🩺 Troubleshooting | Fixes for the usual problems |
| 🗑️ Uninstalling | Removing the app and everything it stored |
| 🛠️ Building from source | Run, test and package Polygluttony yourself |