Download all docs

Recipe: Branded Voice Content

This recipe generates on-brand copy and speaks it aloud in a consistent voice. A brand carries your visual and verbal identity, a prompt holds the reusable instruction template, a lab writes the text, and a mouth — the text-to-speech voice inside the lab — turns it into audio. Same brand, same voice, every time.

The problem it solves

Generated content drifts. One prompt writes formally, the next casually; one audio clip uses a different voice than the last. Without a single home for “how we sound”, every piece of content reinvents the brand. This recipe pins identity to a brand element and the phrasing to a reusable prompt, so the text the lab writes and the audio the mouth speaks stay consistent across everything you produce.

Elements

ElementRole
brandPer-circle visual and verbal identity for chrome, login, and content surfaces.
promptReusable instruction template (with optional A/B variants).
labGenerates the on-brand text.
mouthText-to-speech voice inside the lab; speaks the generated copy.

Flow

  1. Create a brand. Like other modifiers it is config-only — it carries your identity (the look and the tone) and is attached to surfaces rather than invoked; a circle, an app, or a spa can each attach a brand for its visual identity.
  2. Create a prompt that frames the writing task and folds in your brand’s tone of voice. render it with per-piece variables to preview the final instruction, and keep variants if you want to A/B two phrasings.
  3. Create a lab and generate (or chat) the copy from the rendered prompt. The lab produces the on-brand text.
  4. Speak it. A mouth is a text-to-speech voice that lives inside the lab — list-voices to see the voices configured on it, then synthesize to turn the generated line into audio with a chosen voice_id (PCM gives the lowest latency, ~0.7s to first audio). Manage the voice library with clone-voice and delete-voice.

What this shows

Identity is a durable element, not a paragraph re-pasted into every prompt: the brand modifier holds “how we present” once and is attached wherever it is needed, and the prompt holds “how we phrase it” as a reusable, A/B-testable template. The lab is a compound element — its mouth (voice), alongside its brain and ears, lives inside the lab rather than as a loose peer — so text generation and speech share one home. Change the voice once and every clip changes; change the brand once and every surface follows.

Next pages