Every dubbing project uses a speech model to generate the voices. You choose the model for each target language, and you can change it at any time. This guide explains the models available in CAMB.AI Studio and which one to pick.
Which models can I choose from?
MARS8.1-Beta (default): the newest generation of our dubbing model and the default for new projects. It improves pronunciation, rhythm (prosody), accent control and expressiveness with high-pitched voices, and covers more languages. Works best when the voice reference is in the same language and accent as the target.
MARS8-Pro: our proven high-quality model. It keeps the original speaker's voice and emotion, and works well even with short or difficult reference audio. It is the only model where the speaker audio settings (Clean Reference, Acoustic Quality Boost, Stability, Speaker Similarity, Accent Boost, Speaking Rate) take effect.
MARS8-Instruct: "director-level" control over emotion and delivery, for film, TV and creative work. You steer each line with emotion tags such as [happy], [sad], [laughing] or [sighing]. It is slower than the other models.
MARS-X Instruct (Beta): our newest Instruct model, in beta. It is the default model in Agentic Dubbing.
MARS7: the previous generation (legacy). Use it only if you need to match older projects.
Which one should I use?
Most videos: keep the default, MARS8.1-Beta.
You want to fine-tune a speaker's voice with the audio settings (for example noisy source audio, or a voice that doesn't sound like the original): switch to MARS8-Pro.
You need to direct emotion line by line (drama, film, ads): MARS8-Instruct, with emotion tags. See the Emotion Tag Gradation Guide.
You're dubbing with the AI agent: MARS-X Instruct (Beta) is selected by default; you can switch to any other model in the model menu.
Not sure which sounds best for your video? Make a copy of the project (Settings → Create a copy of this project), switch the model in the copy and compare.
How do I change the model in a project?
1. Open your project and click the gear icon at the bottom of the left-hand toolbar.
2. Under Project Model, use the arrows to pick the language you want to change. Each language has its own model.
3. Choose the model from the Model menu and click Update.
4. Regenerate the dialogues to hear the new model. Audio you've already generated stays as it was until you regenerate it, and each regeneration uses credits.
When MARS8-Instruct is selected, the panel also links to Best practice for MARS-Instruct.
Do the speaker audio settings work with every model?
No. The speaker audio settings only work with MARS8-Pro. You'll still see the sliders on projects that use other models, including the default MARS8.1-Beta and MARS8-Instruct, but changing them has no effect there. To use them, switch the language's model to MARS8-Pro and regenerate. See Speaker Settings and the MARS8-Pro best practices.
Where else do I pick a model?
Agentic Dubbing: in the model menu next to the send button. See Agentic Dubbing.
Text-to-Speech: see How to generate Text-to-Speech audio.
API: developers choose the model in each request. See the model guide in the API docs.