DocumentationLocal backend
Profiles
Full, Medium and Light: how the local backend fits your graphics card.
The local backend has three profiles. Each one is a set of models sized to fit a certain amount of video memory (VRAM). Bigger models give a better dub but need more memory.
The three profiles
- Full
- Free video memory
- 12 GB or more
- Best for
- Best translation, for videos
- Medium
- Free video memory
- 8 to 12 GB
- Best for
- Smaller translation model, keeps up with live
- Light
- Free video memory
- 6 to 8 GB
- Best for
- Lighter speech recognition
Each profile stays within its limit: Full uses up to 12 GB, Medium up to 8 GB and Light up to 6 GB.
Full reads best but is the slowest to translate. It is the profile for recordings on a free 16 GB card or a 24 GB one. On live streams, Medium usually keeps up better.
Light recognises speech with a smaller model, so it makes more recognition mistakes, and its voices take longer to make.
How the profile is chosen
By default the profile is set to Auto. When the backend starts, it looks at how much video memory is free at that moment and picks the profile that fits.
Other programs on the GPU count. A game, a video editor or another AI app holding a few gigabytes can move you from Full to Medium, or from Medium to Light. Close them before you start a dub if you want the bigger profile.
Pin a profile
You can pick the profile yourself under Profile in the popup. Choose Auto to go back to automatic choice.
When you switch, the popup shows Switching… dubs restart in a few seconds. The backend restarts with the new models and the dub picks up again.
Pick a lighter profile if the popup says the backend is slower than the video or the stream.
Light without an account
Without an account, the local backend runs the Light profile only. Signing in unlocks Medium and Full, on any plan, including Free. Signing in does not move the dub to our server: it still runs on your GPU. See Your account.
Not enough memory
The local backend needs a card with 6 GB or more. When less than 4 GB is free, it warns you and recommends Server, which runs the dub on our GPUs instead.