Hello everyone, Voxta v1.1.0 is now available.
This build brings a new Stage View for your avatar-based chats. When using characters with transparent backgrounds, this creates a visual novel-style layout for your multi-character conversations.
We've also added the AssemblyAI module as a new Speech-To-Text alternative.
As usual, find other improvements and bug fixes in the list below.
✨ New Features & Updates
Stage View: A new visual novel-style chat view for avatar-based chats.
AssemblyAI Module: An alternative Speech-To-Text service.
Script Snippets: Added script snippets to the scenario script editor.
Experimental Voxta Cloud Sign In: Adds an experimental sign-in option for Voxta Cloud.
Custom Python Constraints: Allow specifying custom Python constraints (requirements.user.txt and constraints.user.txt).
⚙️ UI and Workflow Improvements
Modal Memory Book Editor: You can now edit memory books in a modal window.
Collapsible Chat View Selector: The drawer for the chat view selector is now collapsible.
Right Panel Service Search: You can now search in the service settings from the right-hand panel.
Large Voice Selector: Added a large variant for the voice selector UI.
Asset List Refresh: Added a button to refresh the assets list.
Copy Multiple Tags: You can now copy multiple tags at once.
Auto-hide Resources in Collections: Resources are now automatically hidden when you add them to a collection.
Diagnostics Tab: You can now middle-click on the diagnostics menu to open it in a new tab.
UI Polish:
Added a gradient to the mobile chat message container.
Prevented layout shift in the assistant message bubble.
Improved memory usage.
Various other styling changes.
ComfyUI: The Workflows directory is now created automatically on install.
Voxta Cloud: Improved error messages for when Voxta Cloud models cannot be loaded.
🛠️ Key Fixes & Technical Enhancements
Prompting:
Improved chat history formatting in the instruct template.
Added Granite 4.0 Prompt Formatting Template.
Improved the Mistral V7 Tekken prompt formatting template.
Prompting now uses "event" for assistant/companion and "story" for roleplay/storytelling.
Fixed double bos (beginning of-stream) tokens for Mistral prompts and local hosted models.
Image Gen: Isolated the image gen prompt to prevent polluting the chat history.
Scenario Intro: Fixed cases where the scenario Intro field was not being used.
Chat Messages: Fixed the "regenerate from here" option missing from messages.
UI Animations: Fixed the cross-fade animation for transparent images.
Windows SDK: Fixed DPI scaling for screen capture.
Chatterbox: Fixed flash-attn being pinned to Windows-only builds.
Number Formatting: Applied invariant culture to fix number formatting issues in some regions.
ExLlamav3: Updated to version 0.0.11.
Installer: Corrected the file icon association for .voxtpkg files.
🧑💻 Voxta SDK
The SDK now targets .NET 10.
Added a simple LLM inference API for Chat Augmentations.
We've also created a lot of new characters for you to play with! You can find them in our Discord channel. https://discord.gg/voxta
We appreciate your continued support and look forward to your feedback on these updates.