Narration, scene art and ambience

Three optional layers sit on top of the text: a voice that reads the Dungeon Master aloud, pictures painted for scenes and faces, and a bed of ambience and music. Each depends on a service the server has, and each fails soft: when the service is missing the layer stays off, or a placeholder stands in, and the session carries on.

Narration

With Narration on in the campaign's Setup tab, every Dungeon Master passage is read aloud in the campaign's narrator voice, picked from the voices the speech server lists (Kokoro offers ten; Heart, a warm female voice, is its default). The first words arrive about a second after the passage is sent and the rest streams in. Each listener has their own mute, volume and replay in the voice dock, and a read-aloud button sits on every passage. Audio is generated once per passage and cached on the server; a passage that could not be read aloud says why.

Every speaker can have a voice. Game settings has a Voices panel: the party, the cast and each kind of monster take a voice, players set their own character's, and Cast the others fills in the rest from the sheet, the NPC's traits and the story. A quoted line is read in the voice of whoever the prose says spoke it.

Narration needs a speech service the server can reach: Kokoro, or any OpenAI-compatible speech server (OpenAI's own included), chosen in the admin panel under Speech, on a self-hosted server and on a device world alike. The switch simply does nothing without one.

Faces and theatre inserts

The Faces setting puts a theatre insert over the scene when an NPC speaks: their portrait and name in a frame. The presentation setting chooses between plain and theatre for the whole table.

Scene art, portraits and maps

With an image backend the Dungeon Master paints inline scene art as the story moves, location pictures for new places, and character portraits at creation. At a table playing in another language, the picture prompt is rewritten into English first, since image models read English best. Every picture can also be uploaded by hand wherever the AI would paint one; without either, painted portraits for each race and class stand in. On an app-hosted world the OpenAI key that narrates also paints; on a self-hosted server the administrator chooses ComfyUI, an OpenAI-compatible image model, or a FLUX worker. Pictures are queued behind narration on a shared GPU, so a scene picture can arrive a little after its paragraph. The Setup tab sets pictures on or off, their shape, and fast or slow quality. Battle maps themselves are painted on your device from terrain and never need an image backend.

An old or interrupted request shows Image tool requested with a Generate button, so nothing is lost when a backend was busy.

Ambience and music

A catalogue of 78 cues in three layers (a room bed, music, and stings) is driven by the Dungeon Master, by a human DM from the console, or inferred from the scene when Follows the scene is on; a place or a fight's kind of scene picks its sound. Each listener has a volume, and ambience ducks under narration.

The Sound control in the table header opens the Sound panel: what is playing now, the Room and Music rows with several takes per cue, Next to skip to another take, a level for each layer, and tile pickers for whoever steers the story.

The sound library is open-licensed recordings (public domain, Creative Commons and OpenGameArt), each credited on the server's licences page. A self-hosted server set up from the source has it already. The desktop and Android apps leave it out of the installer: an admin at a table with no library sees Download the sound library in the Sound panel (or the admin panel's Sound library card), a 76 MB download kept across app updates, and the table hears the first cue when it lands. Players joining someone else's world hear the host's library.

Where the switches are