Roleplay Chat Guide

Actions in the "+" panel

Continue, Rewrite, Impersonate, Translate, and the second stage of Inspiration call a text model. Read aloud calls the configured speech model. Opening the panel and choosing an Inspiration direction are free.

Message-level actions (long-press any message)

• Candidates (swipe): switch left/right among multiple generated versions at the same floor; a candidate picker can show all at once.
• Edit: change message body; chain of thought can also be edited or copied separately.
• Hide / restore: move a message out of the prompt. It stays visible in the UI, but the model can't see it. Restore any time.
• Bookmark / checkpoint: mark key floors and jump back directly from the session list.
• Branch: start another story line from this message.
• View this message prompt: Prompt Inspector shows the actual context this reply received, useful for checking whether worldbook entries injected.
• Other actions: copy, copy as new message, delete this message / delete current candidate / delete everything after this, and multi-select cleanup.

Writing style and identity

• Open Chat → ⋮ → Current status to change AI writing style, Author's Note, Persona, and World lore in one place.
• Persona is your identity card (name + persona), with multiple profiles and switching, injected at the {{user}} position.
• Chat appearance and non-streaming display live under ⋮ → All tools → Presentation. These are display settings and do not create a model call.

Quick Reply · automation · regex

• Quick Reply v2: card-provided or custom shortcut button groups. Tap once to send a preset instruction, with STscript support.
• Automation: four modes: compatible / confirm each / fill only without sending / disabled. Controls how far in-card automation scripts can execute. STscript commands in the input (/send, /trigger, and so on) are intercepted and executed.
• Regex management: view and toggle global and current-card regex scripts, with separate controls for display-layer replacement (beautify) and prompt-feed replacement. If a regex replaces content with empty text, the original text is shown automatically.
• Use ⋮ → Chat settings to switch among Global, This character, and This chat. The narrower layer wins: This chat > This character > Global > built-in default.

Group chat

• Speaking strategy: natural rotation (by story heat) / list order / manual speaker (tap avatar to choose a speaker, or make someone speak immediately) / random draw.
• Character card injection: individual (speaker's card) / merge everyone / merge including disabled members.
• Auto mode: set an interval for group members to respond automatically. "Allow consecutive speech" controls whether the same character can speak twice in a row.
• Member management: add/delete, reorder, temporarily disable. Group chat JSON exported from desktop Tavern imports directly.

Online chat (IM mode)

A per-world switch. When on, chat becomes WeChat-style: replies play as split messages by semantic chunk (with typing-dot bubble and tap-to-skip), typing rhythm has three levels (fast / natural / slow), plus read receipts and automatic time bars after gaps longer than 5 minutes. It is display-only; turning it off leaves chat history unchanged. Good for phone lover / online-friend cards.

Image · read aloud · translate · voice

• Image generation: long-press an AI message and choose Generate image, or use the supported /imagine description command. Each accepted generation may be billed separately.
• Read aloud: read one message or use auto read aloud (new replies are read as they arrive), with an option to read dialogue only; choose voices by model under "Read-aloud voices".
• Translate: translate one message or a selected range.
• Voice input: use the microphone built into Gboard/your keyboard; the old in-app microphone button is gone. Send image is under the + panel and requires a vision-capable chat model.

Session management

Each card can have multiple sessions: create, switch, rename, pin, archive/delete, export, import chat history, and search history.Long-chat summary is under ⋮ → All tools. Summarization may make an extra text-model call; the original archived messages remain local and can be viewed separately.