December 2nd, 2025

Improved

[251202] Improvements to How Setting Notes and Memories Are Applied

We’ve improved how setting notes and memories are applied.

1️⃣ Improved Application and Writing Guide for Setting Notes

We’ve adjusted the system so that setting notes related to the current conversation are applied more frequently and more accurately. Even when there are multiple notes on similar topics, the note that best matches the current conversation will now be referenced first.

Along with this, we’ve added a brief writing guide to the [Setting Notes] section in the character creation form.

  • Please keep each note short and concise, and focus on a single topic per note.

  • For core settings that should always apply, write them briefly as well in the character’s Basic Settings/Story section so they can be reflected more reliably.

  • Avoid overlapping content between notes, and try organizing them by categories such as characters/relationships/worldbuilding. This will improve both search and application efficiency.

If you refer to this guide when writing setting notes, the output will more closely match your intentions.

2️⃣ Improved Handling of Memories That Conflict with Settings

We’ve improved the system so it can better remember new facts learned during the conversation, even if they somewhat conflict with existing settings. Example) If there is a setting such as “The character doesn’t know where the gym is,” we have mitigated the issue where the character kept forgetting the gym’s location even after learning it.

3️⃣ Enhanced Recall of Memories Similar to the Current Conversation

We’ve adjusted the system to recall more past memories whose context is similar to the current conversation than before.

4️⃣ Reduced Tendency to Make Up Nonexistent Facts

We’ve adjusted the system to reduce how often characters invent things that don’t actually exist or are not defined in the settings.

We’ve made these improvements so you can enjoy a more satisfying chat experience, but please understand that in certain situations, behavior may still differ from your intentions due to model quality limitations. We will continue to improve this, and if you encounter related issues, please share specific examples with us via [Contact Us] so we can refer to them.

Thank you, as always. 🙇‍♀️