Product news, creator spotlights, AI research, and reflections from the team building Character AI.
Today marks the biggest update in Character AI's history. Version 2.0 brings three capabilities our community has asked for since day one — long-term memory, emotional continuity, and real-time voice conversations.
Most language models respond to what you say. Our personas respond to what you mean. Here's the training approach that makes that possible.
Rafael Torres explains how creating an AI version of his book character Isabelle changed the way he writes — and the way readers experience fiction.
Designing AI that handles sensitive topics responsibly is one of the hardest problems in our field. Here's how we've approached it — and what we've learned.
Scenes lets creators build shared narrative environments where users choose which persona to interact with inside a rich, pre-set fictional world.
We developed an internal benchmark called EQ-Score to evaluate the emotional quality of AI responses. Here's what it measures and why we built it.
We crossed a milestone this month that once felt impossible. Here's an honest reflection on what we got right, what we got wrong, and where we're headed.
Keiko Yamamoto's persona "Hana" has had over 20 million conversations. She shares the painstaking process of writing her personality, backstory, and voice.
We're publishing our quarterly transparency report covering reports received, actions taken, and what we learned from edge cases our systems didn't handle well.
Charms are micro-personality modifiers that creators and users can apply to personas to fine-tune tone, communication style, and interaction tendencies.