Skip to content

Troubleshooting

The popup says the Local AI Live Translate server is not running

Start START_Local_AI_Live_Translate.bat and leave its window open. The popup checks the server at http://127.0.0.1:8000; if something else uses port 8000, close it first.

No models are listed on the Model tab

Start LM Studio's server: LM Studio → Developer tab → start the server. The popup expects it at http://127.0.0.1:1234 (change LM Studio Server URL on the Model tab if yours differs), and lists the models you've downloaded in LM Studio.

Subtitles show the original text, untranslated

The translation model couldn't be reached or didn't answer. Check that LM Studio's server is running and a model is loaded (the Model tab shows its state). There is no online fallback: until the model answers, lines are shown as they were recognised.

Subtitles are slow

Hover over the round-trip time on the status card to see which step takes the time (Live captions):

  • Pause detection — lower the Silence Threshold on the Tuning tab.
  • LLM translation — a model that doesn't fit in your GPU's memory is much slower; see Translation models for smaller ones. On AMD GPUs keep LM Studio's Vulkan runtime, GPU offload at max and flash attention on.
  • Speech recognition — Whisper-Small is the slowest engine; see Speech engines.
Words are misheard, or a language isn't recognised

Set the video language instead of Auto Detect: SenseVoice's automatic detection only knows Chinese, Cantonese, English, Japanese and Korean. For noisy videos, raise the Speech Detection Threshold (Tuning).

Two people get the same speaker number (or one person gets two)

Adjust Speaker Separation on the Tuning tab: raise it if two people share a number, lower it if one person is split in two. See Label speakers.

A ⚠ line appears in the subtitles

It's an online provider's error — for example a wrong API key, a used-up quota or an unknown model name. See Cloud providers.

The app won't start after unzipping

Unzip it to a folder with a short path, such as C:\LocalAI\: Windows limits file paths to 260 characters, and a very deep folder can stop Python from loading.

Still stuck? Open an issue with the server window's messages (leave out anything private).