Overview

Character.AI lets users chat with AI-driven characters (pre-built or user-created) through persistent, multi-turn conversations styled as roleplay or open-ended chat. Using it day to day means working with a small set of structured fields rather than a single free-text prompt box: a character is defined by a Definition (the instruction block covering personality, backstory, quirks, and behavioral rules), an optional set of example dialogues that model tone and pacing, and a Greeting that's the first thing a user reads. If left blank, the user has to open the conversation themselves. A conversation's context carries forward rather than resetting each session.

Group Chat, Scenes, and Streams

Beyond one-on-one chat, Character.AI supports three multi-character formats. Group Chat puts several AI characters and human users in the same room; it rolled out first to c.ai+ subscribers and, as of this writing, works only in the iOS/Android apps, not on web. Scenes place a user inside a pre-built interactive story scenario with a character, and Streams generate a scripted dialogue between two characters around a topic the user picks (a debate, a roast, a vlog) that can be watched rather than actively driven. Scenes, Streams, and Group Chat results can all be shared to an in-app community feed, alongside a broader social feed for sharing character creations.

Memory features

Character.AI's memory system has two parts. Story Memory lets any user pin individual messages (key confessions, plot turns, decisions) so they stay available to the character later in the conversation. Facts automatically extracts details mentioned during chat (appearance, relationships, hobbies) into separate lists for the user's persona, the character, and supporting characters; Facts is limited to paid subscribers, who also get a larger pin allowance and more detailed memory visualization.

AvatarFX and image generation

AvatarFX is Character.AI's video-generation tool: a user supplies a still image and an audio clip, and the system produces a short video where the character's lips, face, and gestures move in sync with that audio. It's built on flow-based diffusion models and Character.AI's own text-to-speech system, and applies safety filters that block generation from photos of real people or public figures and add a watermark to output; it launched to c.ai+ subscribers first, on a waitlist, ahead of a wider rollout. Separately, the Imagine tool turns a moment in a text conversation into a generated image, collected in an in-app gallery that can be filtered by persona, favorited, and shared to the community feed or downloaded.

Technical infrastructure

Character.AI runs its own inference stack rather than reselling access to an outside model provider, and has published technical detail on how it keeps that affordable at scale: techniques like multi-query attention, cross-layer key/value sharing, and int8 quantization to cut GPU memory use per conversation, running across thousands of GPUs to serve a high volume of concurrent chats. This is a meaningfully different approach from character-chat products that route conversations through a third-party LLM API.

Content policy and adult-content positioning

Character.AI's community guidelines explicitly prohibit "illegal sexual content, child exploitation or abuse imagery, grooming, sexual extortion, pornographic content, and nudity," along with non-consensual sexual content and graphic descriptions of sexual acts. Romance and mature storytelling themes are allowed as part of general fiction genres, but content must stay within these limits: the policy does not permit explicit sexual content for any user. Users under 18 face additional restrictions: Character.AI began removing the ability for under-18 accounts to have open-ended chat conversations with characters in late 2025, replacing that experience with more limited, creation-focused features (such as generating videos or stories with characters) alongside expanded age-assurance checks.

Access model and pricing structure

Character.AI's free tier runs mid-chat ads and shares a daily cap on swipe/reroll regenerations across every character on the account; once that cap is hit, additional swipes can be bought with an in-app currency (Charms) that can also be earned through daily tasks like logging in or setting up a persona. The paid c.ai+ subscription removes ads and the swipe cap outright, and adds priority access during peak hours, unlimited voice calls, expanded memory (Facts), a larger pin allowance, and earlier access to new features such as AvatarFX and Group Chat.

Limitations

There's no public developer API: all interaction happens through Character.AI's own apps and website, and any third-party "API" wrapper in circulation is an unofficial, unsupported reverse-engineering project rather than something Character.AI provides. Group Chat and Scenes are currently mobile-only, so desktop web users don't have access to those formats yet, and the free tier's ads and capped swipes are a real day-to-day constraint compared to some competing platforms' flatter free experience.

Important distinctions

Character.AI is frequently discussed alongside CHAI and JanitorAI, which also center on persistent character chat but take different approaches to adult content. Like CHAI, Character.AI runs conversations entirely on its own models rather than letting users choose a backend, which sets both apart from JanitorAI's bring-your-own-model design; where Character.AI differs from CHAI specifically is in its multi-character formats and its investment in video generation (AvatarFX), neither of which CHAI's product currently offers. Unlike JanitorAI, Character.AI does not let users bring their own third-party LLM or API key.

Related glossary concepts

Character Chatbot AI Roleplay Persona Character Memory Long-Term Memory Text-to-Image Content Moderation NSFW Filter Large Language Model (LLM) Group Chat

Related database entries

CHAI JanitorAI Replika Nomi SpicyChat

Sources & further reading