Highlights
This release adds two major model updates:
- complete DramaBox expressive TTS integration
- official ChatterBox 23-Lang V3 checkpoint support
- improved quotation highlighting in the Multiline TTS Tag Editor
DramaBox
DramaBox joins TTS Audio Suite as an English expressive TTS engine for both TTS Text and TTS SRT.
DramaBox uses scene-style prompting:
- dialogue inside quotation marks
- stage directions outside quotation marks
- laughter, sighs, pauses, and delivery transitions
- optional reference-audio voice cloning without a required transcription
- reference-free generation with an empty narrator
The engine includes:
- CFG, STG, negative prompting, and CFG rescaling
- native generation-duration targeting
- configurable reference duration
- prompt templates and automatic dialogue quoting
- per-segment parameter switching
- character switching and pause tags
- optional Perth watermarking
- generated-audio caching
Native SRT duration targeting
DramaBox can receive each subtitle’s target duration directly during generation.
The selected SRT timing mode still applies final correction when necessary, but native targeting can substantially reduce waveform stretching and compression.
The SRT Timing Report shows native duration accuracy and marks near-silent DramaBox segments directly beside the affected subtitle.
Memory and performance controls
DramaBox includes three memory strategies:
- Fast keeps the major components available for repeated generation
- Staged releases temporary components between generation stages
- Sequential transfers major components as they are needed
The engine also supports the official LTX FP8 weight-storage path and optional torch.compile.
Staged, sequential, FP8, and compile remain experimental. DramaBox is a large CUDA model, and generation can be sensitive to reference audio, requested duration, guidance settings, and seed.
ChatterBox 23-Lang V3
The ChatterBox 23-Lang engine now supports selecting the official V1, V2, or V3 checkpoint.
V3 uses the existing ChatterBox integration, including:
- TTS Text and TTS SRT
- character switching
- pause tags
- parameter switching
- generated-audio caching
The V3 path skips the legacy alignment analyzer and removes its final degraded token artifact.
The official V3 checkpoint currently does not show a reliable response to the exaggeration control.
Multiline TTS Tag Editor
The editor now subtly highlights complete straight and curly quoted spans.
This makes dialogue easier to distinguish from surrounding scene directions, especially in DramaBox prompts.
Unmatched quotes are left unchanged, and quoted values inside parameter or inline tags retain their existing tag colors.
Workflow and documentation
This release includes:
- a new DramaBox Integration example workflow
- a matching workflow cover
- a complete DramaBox prompting guide
- updated parameter-switching documentation
- updated engine comparisons, model sources, folder layouts, and installation metadata
Important notes
- DramaBox is officially English-only.
- DramaBox requires NVIDIA CUDA and has substantial VRAM requirements.
- DramaBox model weights use the conditional LTX-2 Community License.
- Perth watermarking is optional and disabled by default in TTS Audio Suite.