@miraiclip/narration

Studio voiceover for a screen recording, in four plain steps you can inspect and edit between:

import { applyCommands } from "@miraiclip/core";
import { openAiSpeechGenerator } from "@miraiclip/audio-sources";
import { scriptFromMarkers, rewriteScript, synthesizeNarration, alignNarration } from "@miraiclip/narration";

const doc = project.getState().doc;
let script = scriptFromMarkers(doc, { transcript });                     // 1. a line per step
script = await rewriteScript(chatModel, doc, script, { style: "friendly" }); // 2. optional: clean it up
const takes = await synthesizeNarration(openAiSpeechGenerator({ apiKey }), script, { store }); // 3. speak it
const plan = alignNarration(doc, script, takes, { clipIds: ["screen"], linkedClipIds: ["camera"] }); // 4. fit it
applyCommands(project, plan.commands);

• alignNarration is deterministic. Each line gets its step's slot; a line that runs long gets a freeze frame at the end of its step (or a slow-down with fit: "slow"), and with speedUp a step with little to say plays faster. Every adjustment is an ordinary clip, speed or move on the timeline, and placements lists what happened to each line, so sync problems are visible and fixable.
• Any chat model for the rewrite (@miraiclip/assistant models fit), with a per-line word budget from the step's length.
• Any voice generator from @miraiclip/audio-sources (ElevenLabs, OpenAI, a server proxy). Lengths are read from the audio itself; unchanged lines are reused on re-generation.

Docs: https://docs.miraiclip.com/