diodiogod/TTS-Audio-Suite v4.27.0

v4.27.0 - Initial MOSS-TTS LoRA Training Support

on GitHub

Initial MOSS-TTS LoRA training release

This release adds the first integrated MOSS-TTS LoRA training workflow to TTS Audio Suite.

Added

• Initial MOSS-TTS 8B (Delay) LoRA training through the unified 🎓 Model Training node
• MOSS dataset staging, manifest building, dataset preparation, and training config nodes
• Managed local MOSS LoRA adapter layout under ComfyUI/models/TTS/moss_tts/loras/
• Example workflow for building a clipped speech dataset and training a MOSS LoRA
• Optional 4-bit MOSS LoRA training path for lower-VRAM setups

Improved

• MOSS LoRA inference loading and adapter management
• MOSS direct TTS inference now works without forcing a narrator fallback when no reference is provided
• MOSS documentation, model layout notes, and generated engine comparison metadata

Current scope

This is initial support, not a finished MOSS training UX.

Current integrated training scope is:

• MOSS-TTS 8B (Delay) LoRA training
• Local adapter export for direct reuse in inference
• Functional but still early dataset-building workflow

Still planned later:

• Better dataset review/editing UI
• Broader training support beyond the initial Delay 8B LoRA path
• Additional training polish and resume/workflow refinements