Character: identity, model & voice

The settings that decide who the character is, what it looks like, how it sounds, and the rules and moods it brings to every conversation.

Open a character from Characters with Open studio. The editor shows the character on a stage. The right dock holds six setting groups — Identity & model, Language & voice, Mind & knowledge, Behavior, Scene & motion and Channels, with the separate Teacher mode entry below them. Click a group to open its first tab in a panel; the group's other tabs sit in a strip at the top of the panel. The left dock holds the appearance tools.

This page covers the Identity & model and Language & voice groups, plus the Guardrails and Emotion tabs of the Behavior group. The AI model, documents, personality, memory and tools are on Mind & knowledge; publishing and channels are on Faces, phone & publishing.

Identity — who it is

Identity & model › Identity holds the character's own words.

Writing a persona that works

Write it as a description of a person, not a specification of a robot. "You are a helpful assistant" produces a generic answer machine; "You have run the front desk of this hotel for eleven years, you know every regular by name, and you would rather solve a problem than explain a policy" produces a character.

Wake word

Choosing Wake word reveals four extra settings. The character then ignores what it hears until one of your phrases appears in it — the right choice for a noisy showroom or a device left running all day. The check runs on the server, so Unreal, web and kiosk clients all follow the same rule.

• Wake phrases (one per line) — give people several natural ways to say it.
• Case sensitive — leave off, so "Hey Alara" and "hey alara" both count.
• Hide the wake phrase from the LLM — leave on, so "hey alara what time do you close" reaches the AI as "what time do you close".
• Cooldown (s) — the minimum time between two triggers. The default 1.5 seconds stops a single greeting firing twice.

Model & Face — what it looks like

How you change the look depends on where the character's model comes from.

• Character Studio library characters (the realistic 3D characters; a new character opens here, on Templates) — use the left dock: Templates, Face & head, Hair, Brows & lashes, Eyes, Skin, Makeup, Teeth, Body and Outfit & accessories. Each tool opens Character Studio over the stage. Choosing the active tool again, or opening a right-dock group, closes it and saves the look ("Saving the look…").
• Character packs — the bar at the top left of the stage switches between 3D (the Character Studio library), Cartoon pack (cartoon characters built from parts) and LiveFace 2D (a talking 2D portrait).
• Your own model — open Identity & model › Model & Face. Current avatar shows the model in use (Edit appearance opens the left dock's Face & head). Create is the modular builder with its Presets; finish with Use avatar or Save changes. Upload takes a GLB file of up to 100 MB (Choose a GLB file, then Use avatar).

The character editor guide covers the stage, the scene and animation tools, and testing. Custom avatars lists what an uploaded model needs.

Model settings

Identity & model › Model settings answers two questions for your own model: what kind of model it is, and which shape in it corresponds to which facial movement. For a Character Studio library character a read-only Library character card takes the place of the type, source and mapping cards — the library draws it with its own face rig and lip sync, so there is nothing to map. Its Edit the look button opens the appearance tools.

1. Choose the Avatar Type that matches where the model came from. Types with a ready mapping show Preset available.
2. Upload the model under Model Source rather than pasting a URL. Uploading lets the studio read the list of shapes inside the file, so the mapping dropdowns are filled in for you. Maximum 100 MB.
3. If a Native facial rig card appears, the model carries its own calibrated facial driver and needs no mapping.
4. Otherwise press Apply preset if one exists for the type, or Auto-detect to match names automatically.
5. Check the Jaw and Mouth & Lips groups. A name the model does not contain is shown in red as "(not in model)".
6. Try it with Test conversation and raise Global Gain if the mouth movement looks too subtle.

Emotion → Face Morph Intensity sets how strongly the character's mood shows on its face (default ×0.70). Mouth shapes are held back automatically while it speaks, so emotion never fights lip sync.

Unreal plugin switches

The UE Plugin Settings card is read by the Unreal plugin only:

• Face Morph Driver — drives facial expression from the emotion engine. Leave it off if your Blueprint already has a face AnimBP, or the two will fight.
• Game Actions (MoveTo / Follow / …) — gives the character movement it can decide to use (move to, follow, stop, wait). For game NPCs; leave off for a kiosk. See Actions.
• Conversational Gestures — lets the character choose gestures (wave, nod, point…) while it speaks.
• Teacher Mode (classroom board) — gives it a classroom board for formulas, notes and quizzes.
• Interrupt (on new input) — on by default: a new typed message or mic press stops the current answer.
• Auto Barge-In (voice detection) — off by default: the character stops when someone starts speaking aloud. Turn it on only if the microphone is far from the speaker.

The group's last tab, LiveFace 2D, turns a portrait photo into a talking head. It is a plan feature and is covered in Faces, phone & publishing.

Language & Voice — how it sounds

Language & voice › Language & Voice sets the voice. The languages the character speaks are chosen on Identity › Languages.

Keys saved in My workspace › Settings › API Keys become the default for all your characters. For MiniMax and OpenAI TTS, Override for this character sets a separate key for this one (Manage in Settings opens the workspace keys); when a workspace owner or admin saves such a key, it also replaces the workspace default for that provider. For ElevenLabs the tab has no key field: the workspace key, when one is saved, is used for every character, and Manage my key in Settings on the Voice connection card opens the workspace keys. With a key in place, Fetch voices or Refresh voices loads the list under Voice Selection.

Create your voice

With ElevenLabs selected, Create your voice makes a voice from your own recording. It is available from the Starter plan up; on Free the card offers View paid plans.

1. Enter a Voice name.
2. Add an Audio recording: MP3, WAV, M4A, OGG or WebM, up to 20 MB, one speaker, no background music.
3. Confirm that it is your voice, or that you have the speaker's permission, and press Create voice.

With the included IAMX voice, the new voice belongs to your workspace, up to the number of personal voices shown on the card. With your own ElevenLabs key, it is added to your ElevenLabs account and your provider's limits and charges apply. A voice that needs provider verification waits under Voices awaiting verification and counts toward your limit until you remove it. Delete my voice removes a personal voice you no longer use.

Choose a speech model, then test the whole reply

For ElevenLabs and OpenAI TTS, TTS Model trades response speed against quality. Compare models with the same voice and sentence.

The tab starts on Flash v2.5, which the card recommends for a live NPC. Eleven v4 Turbo is the newest real-time model, at about 150 ms to the first audio. Eleven v4 has the highest quality but takes about a second to start speaking, so it suits narration and long text better than live conversation.

OpenAI TTS offers GPT-4o Mini TTS (the default), TTS-1 and TTS-1 HD.

Expressiveness

For ElevenLabs, Voice Delivery offers starting points — Natural conversation, Warm and friendly, Calm and clear, or Custom — use character speaking style. A profile sets Stability, Style Intensity and Speed for you.

Prosody fine-tunes delivery: Stability (low is dynamic and emotional, high is even and predictable), Similarity (how closely a cloned voice is followed), Style Intensity (0 is neutral, 1 is exaggerated), Speed (ElevenLabs only) and Speaker Boost. These are ElevenLabs-style settings; other providers may ignore them.

If the character uses real-time voice

When Core AI is set to Gemini Live (realtime) or OpenAI Realtime (voice), this tab changes completely: there is no separate speech engine, because the model produces the audio itself. Pick a voice from the Gemini Live Voice or OpenAI Realtime Voice grid, filtered by All / Female / Male. A real-time model cannot answer on WhatsApp.

Guardrails — the rules

Behavior › Guardrails holds what the character must not do or say, in plain language, one rule per line. With Enable guardrails on, IAMX adds them to the AI's instructions with high priority, in standard and real-time voice conversations alike. They guide the model; they do not guarantee that every reply follows every rule.

• Phrase every line as a prohibition: "Do not…"
• Click the Examples (click → add): lines for the common ones — political neutrality, no medical advice, never repeating account, ID or card numbers, no response to abusive language, no alcohol advice to under-18s.
• Stay within the counter above the box (7,500 characters). When it turns red, saving fails.

Emotion

Behavior › Emotion sets the character's mood. The Plutchik Emotion Wheel is the starting state across eight emotions — Joy, Trust, Fear, Surprise, Sadness, Disgust, Anger and Anticipation. During a conversation every visitor message shifts it, and the state feeds both the character's tone and its facial expression.

Where to go next