CielChan v1.6.8 has been released:
Slightly reduced the total download size of CielChan.
Reverted the base model back to Llama 3.2 3B, as the Gemma model was reportedly causing issues for some users.
Fixed an issue introduced in v1.6.7 where Watch Together System Audio could fail to detect multilingual text in certain edge cases.
Substantially increased overall generation speeds. Recent optimizations made for the Gemma model have also improved performance when using the Llama model, and have increased the available context window.
Fixed an issue where System Audio could become delayed when both System Audio and Observations were enabled outside of Watch Together mode. System Audio can now reliably be used to let CielChan listen to your computer without requiring Watch Together to be enabled.
Fixed an issue from the v1.6.7 update where additional Chatterbox options would not appear when required.
Slightly increased volume output when using Kokoro TTS.
Fixed an issue where avatar models loaded specifically from the template options in Settings could fail to reload in future sessions, causing the avatar to revert to the default model.
Other small various fixes and adjustments.