SoulScript v1.1.0 Release

SoulScript v1.1.0 is now available.
This update focuses on voice consistency, generation control, and core architecture improvements that make the pipeline more stable and predictable.

This is the most stable version of SoulScript so far.

Major Features

Voice Consistency System

Voices now use a voice-level generation seed to keep delivery consistent across all dialogue lines and across different projects.

Every voice stores a default generation profile which is used automatically when generating lines.

This significantly improves voice stability across large scripts.

Line Variation System

The Validate tab now supports safe experimentation with line variations.

New actions include:

Regenerate
Regenerates the line using the current voice settings.

Try Variation
Creates a temporary variation without replacing the approved line.

Accept Variation
Replaces the approved line with the new variation.

Discard Variation
Deletes the preview and restores the original line.

Reset to Voice Default
Removes any line-specific override and regenerates using the voice's default settings.

This allows users to refine difficult lines without losing previously generated audio.

Non-Destructive Validation Workflow

Line regeneration during validation is now non-destructive.

Variations are generated to a temporary preview location and only replace the approved audio when explicitly accepted.

This prevents accidental loss of good takes.

Architecture Improvements

GenerationProfile System

SoulScript now uses an engine-agnostic GenerationProfile abstraction.

This stores generation parameters such as:

• seed

• style modifiers

Profiles exist at two levels:

Voice Default Profile
Used for all lines generated with that voice.

Line Override Profile
Created when a user accepts a variation for a specific line.

Resolution order:

1. Line override

2. Voice default

This architecture allows the system to remain engine-agnostic and prepares SoulScript for additional engine integrations.

Engine-Agnostic Generation Pipeline

Generation profile resolution is now centralized and shared across:

• batch generation

• single-line regeneration

• validation workflow

This ensures consistent behavior across the entire generation pipeline.

UI Improvements

Model Loading Indicator

A new model loading indicator now appears in the main status bar whenever an AI model is loading.

This provides clear feedback during:

• model switching

• voice design

• validation regeneration

GPU Monitor Improvements

The GPU monitor widget has been updated:

• Removed the incorrect Peak usage display.

• Added VRAM usage display.

• Improved overlay rendering so the graph uses the full available space.

Stability Improvements

• Fixed UI freeze when regenerating lines while models load.

• Improved thread handling during validation regeneration.

• Added safe preview generation for line variations.

• Improved generation pipeline reliability.

Internal Improvements

• Added generation profile resolver shared across the pipeline.

• Voice library schema updated to support default generation profiles.

• Project configuration now supports per-line generation overrides.

• Added migration support for older voice libraries.

Upgrade Notes

Existing voices and projects are automatically migrated.

• Voices without generation profiles will receive a default seed.

• Projects without line overrides continue working unchanged.

No user action is required.

Final Notes

This release focuses on consistency and reliability, which are critical for large visual novel projects.

The new generation profile system lays the groundwork for future engine integrations such as:

• Unity

• Ink

• Yarn