This is the MESA Cascade Update (v4.5).
We spent the last two months obsessively studying user interactions and identifying improvement points. Every system that runs the livestream has been rebuilt or refined.
Details below.
Cast Expansion
We grew the cast from 12 to 20 characters.
This process was massively accelerated thanks to our in-house character pipeline. More characters means more content.
Navigation Systems
Navigation systems received a complete revamp.
Character spacing is improved, clustering is reduced, and pathfinding is more reliable. The system was rebuilt from the ground up to support complex animations and character states.
Better navigation means better content.
Output Modes
One stream, many outputs.
Each output mode can be configured for where it's going. This is useful for running TikTok Lives, or just creating short-form content.
Each platform has its own audience and deserves its own experience.
Text-To-Speech
Text-to-speech is now 30% faster, more expressive, and has reduced latency on character voice switches.
One of our TTS provider instances can handle thousands of unique character voices with little additional overhead. Around 10 minutes of reference audio is enough to create accurate voices.
MESA Backend
We added support for local AI models with Ollama.
This is useful for low latency tasks like fetching character memories, triggering emotes, or other interactions in realtime.
Dialogue System
Characters now acknowledge chat more at the beginning and end of scenes. This helps the dialogue segue into requested topics while still providing a balanced experience.
In addition, we've tweaked the system to increase general adherence to topics, something that chat or strong character personalities would tend to override.
Updated Site
- Watch page includes YouTube player and chat for a more focused stream experience.
- New guide page and command builders.
- Refreshed beta signup page and feedback page.
Summary Of Changes
- Cast expanded to 20, scaling toward 100.
- Rebuilt navigation.
- Added dedicated output modes.
- Improved TTS performance and quality.
- Smarter dialogue and faster local model integration.
- New site.
What's Up Next
- Platform-specific livestream experiments, including weekly TikTok event streams and an experience built for X.
- Scaling toward 100 characters to expand the sandbox.
- Continuing to validate the content engine. Once MESA is proven, we invite creators in.