Two providers. Switch at generation time.
Configure OpenAI, Claude, or both in Settings. When both are set up, the provider picker appears inline above the Generate button — flip between them on a per-draft basis. The cost estimate updates live based on the selected provider and model, so you know what you're about to spend before you spend it.
- OpenAIGPT-4o by default. Configure model in Settings → AI Writing.
- AnthropicClaude Sonnet by default. Same configuration pattern.
- SwitchingPer-generation, via the inline picker. No setting trip required.
- CostLive estimate before generation. flexVox charges nothing; you pay your provider.
What's actually in the prompt.
The flexVox prompt engine is not a single template. It builds a structured request from many sources and tunes phrasing to whichever provider you picked. Below: every piece of context that travels with a script-generation call.
- Brief Your episode description in plain language. Drives plot, pacing, and characters' relationships to the topic.
- Format & tone Two-host conversation, audio drama scene, newscast, narration — the rough shape of what you want. Tone influences register.
- Speaker count & narrator mode Full Cast, Single Narrator, or Narrator + Cast. Changes how the writer assigns lines.
- Character profiles For each speaker: role, age, bio, personality, speaking style. Produces distinct, consistent voices in the draft.
- Reference materials Optional text you upload — articles, research, outlines. Used as system context, not copied verbatim.
- Tag instructions Detailed v3 expression-tag guidance: stacking, emotional arcs, natural-language custom tags, ElevenLabs SFX best practices.
- SFX & music guidance
How to write a useful prompt, when to use
@underlay, ambient-loop best practices, audio terminology. - Show context For episodes belonging to a production: cast bible, episode template, show tone, narrator mode. Inherited automatically.
The point of all this context isn't to make the writer cleverer. It's to make the draft land in your project ready to cast — tagged in your style, scoped to your cast, with audio cues that will actually generate cleanly.
Character profiles double as casting context.
A character profile attached to a speaker is used twice. First, as context when the AI writer drafts a script — so ALEX's lines actually sound like ALEX. Second, as input to the Auto-Cast suggester, so ALEX's voice match is informed by the same personality fields you already filled in.
- RoleTheir function in the show — host, guest, narrator, antagonist.
- AgeNumber or range. Used by both writer and voice casting.
- BiographyShort paragraph. Where they are in life.
- PersonalityFree-text. The notes you'd give a casting agent.
- Speaking styleCadence, vocabulary, register. Hushed, declarative, hesitant, deadpan.
Reference materials — context without plagiarism.
Upload text-based references — an article you want the conversation to ground in, a research outline, a transcript of a meeting. The content is sent as system context. The writer reads it, but is explicitly instructed not to copy it verbatim. The output is a script informed by the reference, not a script that quotes it.
Prefer a different AI? Copy the prompt.
The prompt sheet has a Copy to Clipboard button. It copies the full, pre-formatted prompt — every piece of context above, assembled into a single instruction. Paste it into ChatGPT, Claude, Gemini, or anything else you trust. Paste the result back into flexVox to land it in the project.
The escape hatch is the point. flexVox doesn't lock you to a provider; it locks you to a workflow that any provider can fit into.
Enhance an existing script with tags.
You don't have to write with the AI to use it. If you already have a script — yours, a collaborator's, a publicly-licensed source — open Enhance Expression Tags. flexVox processes every speech turn, proposes tag additions in a review sheet, and applies the ones you accept. The on-device path uses Apple Intelligence; the cloud path falls back to OpenAI or Claude.