Highlights
This release adds the complete MOSS audio-generation family and standardizes reference-free voice creation across the suite.
The main additions are:
- MOSS-TTS v1.5 speech generation and LoRA training support
- MOSS-SoundEffect v1 and v2 text-to-sound generation
- a unified Voice Designer for Qwen3-TTS, MOSS-TTS, and OmniVoice
- reusable character persistence through Save Character Voice
MOSS-TTS v1.5
The MOSS engine now exposes the current 8B v1.5 checkpoint alongside the existing MOSS models. This release also extends the integrated MOSS LoRA training workflow so adapters can be trained against v1.5.
Model selection now distinguishes local checkpoints clearly, applies model-specific defaults, and provides cleaner Hugging Face download progress.
MOSS sound effects
Two MOSS sound-effect architectures are supported through the unified 🌩️ Sound Effects workflow:
- MOSS-SoundEffect v1 through the existing MOSS-TTS engine family
- MOSS-SoundEffect v2 through its dedicated 48 kHz diffusion engine
The unified node supports per-segment parameter switching, pauses, chunked timelines, configurable crossfades, deterministic seeds, and generated-audio caching. MOSS v2 additionally supports negative prompts, CFG, flow shift, diffusion-step controls, and durations up to 30 seconds per generated segment.
Unified voice design
The new 🎨 Voice Designer provides one workflow for:
- Qwen3-TTS VoiceDesign
- MOSS VoiceGenerator
- OmniVoice Voice Design mode
The engine owns model, language, and generation settings, while Voice Designer owns the reference text and voice instruction. Incompatible engine modes stop with direct guidance. OmniVoice instructions can be built visually with 📐 Visual Tag Builder.
Generated voices can be previewed immediately or passed to 💾 Save Character Voice for reuse in Character Voices, TTS Text, and TTS SRT workflows.
Character voice workflow improvements
- save generated or imported audio as reusable character voices
- discover newly saved voices immediately
- prioritize user voices when names overlap bundled examples
- use improved waveform trimming and playback controls
- keep older workflows compact with the responsive Character Voices layout
Examples and documentation
This release includes new unified example workflows and cover art for Voice Designer and Sound Effects, plus a dedicated Sound Effects guide and updated model/language documentation.
Migration note
The old Qwen-specific Voice Designer node has been removed. Existing workflows using it must connect a Qwen3-TTS engine configured with the VoiceDesign model to 🎨 Voice Designer instead. Connect its opt_narrator output to 💾 Save Character Voice when persistence is needed.