Documentation
Settings and defaults
Choose where to save a setting, and understand which values a new run will use.
- Operators
- Developers
flowchart TB
A[Built-in and global defaults] --> B[Effective project settings]
C[Project context and overrides] --> B
D[Provider connection and speech defaults] --> B
B --> E[New generation request]
F[Cast and block or phrase voices] --> E
G[Selected reviewed speech plan] --> E
E --> H[Run with frozen inputs and settings]Start in the project's settings for the stage you want to change. Global defaults provide starting values; project settings let you make an exception for one job. Simple mode hides some controls. Advanced mode exposes more detail without creating a different project.
Find the relevant section
| Section | What to check |
|---|---|
| Source cleaning | Extraction, OCR and repeated headers or footers. Inspect the extracted text before using it. |
| Transcription | Recognition model, source language, speaker detection and word timing. Recognition and alignment can support different languages. |
| Correction | Language-model profile, instructions and review of proposed corrections. |
| Translation | Source and target language, glossary, context and target subtitle limits. |
| Text and speech preparation | Dialogue recognition, pronunciation changes and segmentation. Rebuild the plan after changing its source. |
| Speech synthesis | Service, exact model, voice/reference, language and model-specific controls. |
| Subtitles | Line length, reading speed, timing and which written revision to export. |
| Audio | Processing and fades. Listen for clipped beginnings and changes in level. |
| RVC | Optional voice conversion after synthesis; requires its own installed service and compatible voice model. |
| Output | File format, bitrate, book metadata, cover and video/caption options. |
How values are resolved
General settings combine built-in defaults, saved global defaults, project context and the project's override. Speech also uses its provider connection and provider defaults. A request can explicitly select a saved run's settings. Speaker casting and individual block overrides are resolved separately; they are not another layer of every project setting.
Changing a global default does not rewrite a completed run. Starting generation freezes the settings, speech plan and source used by that run. Its history remains useful even after the project's settings change.
For voices, a whole-block override takes priority over phrase/cast assignments. Otherwise an explicit phrase voice wins, followed by a named character, source speaker, dialogue default and narrator/session fallback. The generation preview is the place to check what actually resolved.
A short example
Keep an English narrator as the project default. Assign Fred a different voice in Multiple voices, then override one whispered phrase if the model supports it. Preview the affected blocks. Changing the narrator does not necessarily change Fred or an explicitly overridden phrase.
Generate one or two blocks before saving a new setting as a global default. For repeated comparisons, change one control at a time and keep the same text.
MCP and API
Read the section's effective values and current revision before writing. A section PUT replaces its saved override. PATCH merges only at the top level: supplying a nested object replaces that nested value, rather than recursively merging every member. Do not use null as an assumed reset command. Discover the current operation schema and use its reset/default action where provided.
See generation and take selection and the release-pinned generation reference.