@mevdragon/vidfarm-devcli 0.19.1 → 0.20.0

package/dist/src/reskin/library-page.js

@@ -350,6 +350,10 @@ const CLIPS_DIALOG = `
           <option value="openrouter">OpenRouter</option>
         </select>
       </div>
+      <label class="rk-clips-toggle" for="rkScanAudioOnly">
+        <input type="checkbox" id="rkScanAudioOnly" name="audio_only" />
+        <span><b>Clip audio only</b> — save each mined moment as an audio-only <code>.m4a</code> (no video), e.g. lifting sounds from a copyright-free audio video.</span>
+      </label>
     </div>
     <p class="rk-hint rk-clips-dialog-note" id="rkScanNote">We'll pull the whole video into your raws library.</p>
     <div class="rk-clips-dialog-actions">
@@ -1182,6 +1186,7 @@ const CLIPS_SCRIPT = `
   var scanNote=root.getElementById('rkScanNote');
   var scanSubmit=root.getElementById('rkClipsScanSubmit');
   var scanClip=root.getElementById('rkScanClip');
+  var scanAudioOnly=root.getElementById('rkScanAudioOnly');
   var scanAdvanced=root.getElementById('rkClipsAdvanced');
   var IMPORT_NOTE="We'll open the Clipper so you can preview the video and save an exact subrange (or the whole thing) to your raws.";
   var CLIP_NOTE='AI tagging runs on your own key (BYOK) — vidfarm only bills the clip-scan compute.';
@@ -1239,6 +1244,7 @@ const CLIPS_SCRIPT = `
       var clip=!!(scanClip&&scanClip.checked);
       var prompt=clip?(scanPrompt&&scanPrompt.value||'').trim():'';
       var provider=clip?(scanProvider&&scanProvider.value||'').trim():'';
+      var audioOnly=clip&&!!(scanAudioOnly&&scanAudioOnly.checked);
       if(!url){ setNote('Enter a video URL.','error'); return; }
       // Whole-video (non-AI) path now opens the Clipper so the user can pick an
       // exact subrange (or save the whole thing) with a live video preview.
@@ -1253,7 +1259,7 @@ const CLIPS_SCRIPT = `
       if(scanSubmit){ scanSubmit.disabled=true; scanSubmit.classList.add('is-loading'); }
 
       var payload={source_url:url};
-      if(clip){ payload.prompt=prompt; if(provider)payload.provider=provider; } else { payload.import_only=true; }
+      if(clip){ payload.prompt=prompt; if(provider)payload.provider=provider; if(audioOnly)payload.audio_only=true; } else { payload.import_only=true; }
       if(tracer)payload.tracer=tracer;
 
       // Fire the request now; hold it and only wire the result AFTER the modal

package/dist/src/reskin/portfolio-page.js

@@ -38,19 +38,19 @@ const SAMPLE = {
 export const PORTFOLIOS = {
     "sunny-studio": {
         slug: "sunny-studio",
-        title: "Sunny Studio — Short-form that sells",
-        tagline: "A boutique video studio turning brands into scroll-stopping short-form. 40+ ads shipped, 12M+ organic views.",
-        about: "We script, shoot, and edit high-retention vertical video for DTC and SaaS brands. Every cut below started as a rough brief and shipped in under a week. Hover to preview, tap any piece to watch it full-screen.",
+        title: "Andrea Villanueva — Short-form that sells",
+        tagline: "A freelance short-form editor turning brands into scroll-stopping video. 40+ ads shipped, 12M+ organic views.",
+        about: "I script, edit, and caption high-retention vertical video for DTC and SaaS brands — solo, hands-on, and fast. Every cut below started as a rough brief and shipped in under a week. Hover to preview, tap any piece to watch it full-screen.",
         author: {
-            name: "Mia Okonkwo",
-            role: "Founder & lead editor · Sunny Studio",
+            name: "Andrea Villanueva",
+            role: "Freelance short-form editor · Private hire",
             avatarUrl: PROFILE_PIC,
-            bio: "Ex-agency editor now running a two-person studio. I obsess over the first 3 seconds, punchy pacing, and captions that carry the story even on mute.",
-            location: "Lisbon · working worldwide",
+            bio: "Ex-agency editor now freelancing solo. I obsess over the first 3 seconds, punchy pacing, and captions that carry the story even on mute.",
+            location: "Philippines · working worldwide",
             availableForHire: true,
             stats: [
                 { label: "Videos shipped", value: "140+" },
-                { label: "Starting from", value: "$75/hr" },
+                { label: "Starting from", value: "$3/hr" },
                 { label: "Client rating", value: "4.9/5" }
             ],
             certifications: [
@@ -267,7 +267,7 @@ export function renderReskinPortfolio(slug, input = {}) {
     if (!data)
         return null;
     const { author, videos } = data;
-    const contactLabel = author.contactLabel || "Check Availability";
+    const contactLabel = author.contactLabel || "Get 1st Video Free";
     const contactHref = author.contactUrl || CTA_URL;
     const statPills = (author.stats && author.stats.length)
         ? `<div class="pf-stats">${author.stats

package/dist/src/reskin/settings-page.js

@@ -278,7 +278,7 @@ export function renderReskinSettings(input) {
     const keyRows = input.providerKeys.length
         ? input.providerKeys.map((k, i) => {
             const id = `rk-pk-${i}`;
-            const tint = k.provider === "openai" ? "rk-tile-green" : k.provider === "gemini" ? "rk-tile-sky" : "rk-tile-violet";
+            const tint = k.provider === "openai" ? "rk-tile-green" : k.provider === "gemini" ? "rk-tile-sky" : k.provider === "elevenlabs" ? "rk-tile-amber" : "rk-tile-violet";
             const status = k.status || "active";
             const lastUsed = k.last_used_at ? `used ${rkEscape(fmtDate(k.last_used_at))}` : "never used";
             return `<div class="rk-item">
@@ -318,8 +318,8 @@ export function renderReskinSettings(input) {
     <div class="rk-card">
       <div class="rk-card-head">
         <span class="rk-eyebrow">AI keys</span>
-        <h2>Bring your own model keys</h2>
-        <p class="rk-muted">Stored per workspace. Used for clip tagging, recaption, and Swipe — never billed by vidfarm.</p>
+        <h2>Bring your own AI Keys</h2>
+        <p class="rk-muted">Stored per workspace. OpenAI, Gemini, or OpenRouter power clip tagging, recaption, image/video generation, and Swipe — BYOK, never billed by vidfarm. <b>ElevenLabs is audio-only</b> (music, voiceover, transcription) and does <b>not</b> count as a qualified AI key — you still need one of the others for the core features. With no ElevenLabs key, music &amp; speech run on vidfarm's key and bill your wallet (<code>use_wallet_credits</code>).</p>
       </div>
       <div class="rk-list rk-set-keys">${keyRows}</div>
       <form class="rk-set-addkey" method="post" action="/settings/provider-keys">
@@ -328,7 +328,7 @@ export function renderReskinSettings(input) {
           <div class="rk-field">
             <label class="rk-label" for="rk-prov">Provider</label>
             <select class="rk-select" id="rk-prov" name="provider">
-              <option value="openai">OpenAI</option><option value="gemini">Gemini</option><option value="openrouter">OpenRouter</option><option value="anthropic">Anthropic</option>
+              <option value="openai">OpenAI</option><option value="gemini">Gemini</option><option value="openrouter">OpenRouter</option><option value="elevenlabs">ElevenLabs (audio only)</option>
             </select>
           </div>
           <div class="rk-field">

package/dist/src/reskin/theme.js

@@ -635,6 +635,7 @@ export const RESKIN_CSS = `
 .rk-tile-sky{background:var(--rk-sky-tint);color:var(--rk-sky)}
 .rk-tile-violet{background:var(--rk-violet-tint);color:var(--rk-violet)}
 .rk-tile-green{background:var(--rk-green-tint);color:#16a34a}
+.rk-tile-amber{background:var(--rk-gold-tint2);color:var(--rk-gold-700)}
 .rk-tile-coral{background:var(--rk-coral-tint);color:var(--rk-coral)}
 .rk-tile-ink{background:var(--rk-n-100);color:var(--rk-ink)}
 

package/dist/src/services/billing.js

@@ -168,6 +168,11 @@ function resolveCostCenterSlug(input) {
             // primitive (BYOK image gen + chroma key): platform compute/convenience.
             || explicitSlug === "greenscreen_chroma_key"
             || explicitSlug === "media_overlay"
+            // Platform ElevenLabs audio (music / TTS / STT) run on the wallet when the
+            // caller opts into use_wallet_credits (the audio-primitive default).
+            || explicitSlug === "elevenlabs_music"
+            || explicitSlug === "elevenlabs_tts"
+            || explicitSlug === "elevenlabs_stt"
             || explicitSlug === "primitive_media_lambda"
             || explicitSlug === "job_runner_lambda"
             || explicitSlug === "step_functions_standard"

package/dist/src/services/clip-curation/ffmpeg.js

@@ -256,6 +256,54 @@ export async function extractClip(input) {
     }
     return ok;
 }
+/**
+ * Trim EXACTLY [start, start+dur] of the source's AUDIO into a standalone
+ * `.m4a` (AAC) file — the audio-only twin of {@link extractClip}. The video
+ * track is dropped (`-vn`) so the result is a small, editor-placeable audio raw
+ * (e.g. clipping the soundtrack out of a copyright-free YouTube video). Returns
+ * false if the source has no audio track (ffmpeg exits non-zero with no output).
+ */
+export async function extractAudioClip(input) {
+    mkdirSync(path.dirname(input.outPath), { recursive: true });
+    const ffmpeg = await resolveFfmpeg();
+    const { start_time_sec: start, duration_sec: dur } = input.scene;
+    const { code, stderr } = await run(ffmpeg, [
+        "-y",
+        "-ss", start.toFixed(3),
+        "-i", input.videoPath,
+        "-t", dur.toFixed(3),
+        "-vn",
+        "-c:a", "aac", "-b:a", "192k",
+        "-movflags", "+faststart",
+        input.outPath
+    ]);
+    const ok = code === 0 && existsSync(input.outPath);
+    if (!ok)
+        input.onError?.(`ffmpeg exit ${code}: ${stderr.slice(-500).replace(/\s+/g, " ").trim()}`);
+    return ok;
+}
+/**
+ * A static waveform PNG for an audio clip — its "thumbnail" so an audio-only raw
+ * still shows something in the raws gallery (there is no video frame to grab).
+ * Best-effort; returns false if the source has no audio to plot.
+ */
+export async function extractWaveformThumbnail(input) {
+    mkdirSync(path.dirname(input.outPath), { recursive: true });
+    const ffmpeg = await resolveFfmpeg();
+    const { start_time_sec: start, duration_sec: dur } = input.scene;
+    const w = input.width ?? 320;
+    const h = input.height ?? 200;
+    const { code } = await run(ffmpeg, [
+        "-y",
+        "-ss", start.toFixed(3),
+        "-i", input.videoPath,
+        "-t", dur.toFixed(3),
+        "-filter_complex", `showwavespic=s=${w}x${h}:colors=#e2a63a`,
+        "-frames:v", "1",
+        input.outPath
+    ]);
+    return code === 0 && existsSync(input.outPath);
+}
 /** Single representative thumbnail for the scene. */
 export async function extractThumbnail(input) {
     mkdirSync(path.dirname(input.outPath), { recursive: true });

package/dist/src/services/clip-curation/hunt.js

@@ -395,6 +395,8 @@ export function normalizeHuntSpec(raw) {
         spec.source_url = rec.source_url.trim();
     if (typeof rec.folder_path === "string" && rec.folder_path.trim())
         spec.folder_path = rec.folder_path.trim().slice(0, 500);
+    if (rec.audio_only === true || rec.audioOnly === true)
+        spec.audio_only = true;
     return spec;
 }
 // ── misc ────────────────────────────────────────────────────────────────────

package/dist/src/services/clip-curation/index.js

@@ -7,7 +7,7 @@ export { BUILTIN_PRESETS, findBuiltinPreset } from "./presets.js";
 export { ClipModelClient, ClipGeminiClient, ClipEmbeddingClient, EMBEDDING_DIM, EMBEDDING_MODEL, EMBEDDING_MODELS, TAG_MODEL_BY_TIER, buildClipEmbeddingText, buildQueryPrompt, buildTaggingPrompt, normalize } from "./gemini.js";
 export { LocalAgentClipClient, detectLocalAgent, extractJsonObject } from "./local-agent.js";
 export { AVOID_TEXT_GUIDANCE, DEFAULT_CLIP_ASPECT, DEFAULT_CLIP_DURATION_BAND, DEFAULT_CLIPS_PER_10_MIN, applyHuntSpecDefaults, buildEffectiveGuidance, capScenesEvenlyToCount, cropFilterFor, effectiveHuntDurationSec, fitScenesToDurationBand, normalizeAspect, normalizeCropFocus, normalizeHuntSpec, normalizeWindows, parseClipHuntPrompt, parseTimecode, parseTimeRanges, resolveDurationBand, resolveTargetClipCount } from "./hunt.js";
-export { detectScenes, extractClip, extractKeyframes, extractSceneAudio, extractThumbnail, hasFfmpeg, probeVideo, resolveFfmpeg, resolveFfprobe } from "./ffmpeg.js";
+export { detectScenes, extractAudioClip, extractClip, extractKeyframes, extractSceneAudio, extractThumbnail, extractWaveformThumbnail, hasFfmpeg, probeVideo, resolveFfmpeg, resolveFfprobe } from "./ffmpeg.js";
 export { cosineSimilarity, matchesCriteria, rankHits, scoreClip, searchClips, structuredMatchRatio } from "./query.js";
 export { estimateScanCostFromDuration, estimateScanCostFromScenes, formatCostEstimate } from "./cost.js";
 export { processSceneToClip, scanVideo } from "./scan.js";

package/dist/src/services/clip-curation/scan.js

@@ -8,7 +8,7 @@ import path from "node:path";
 import { createIdV7 } from "../../lib/ids.js";
 import { ACTIVE_TAXONOMY_VERSION } from "./taxonomy.js";
 import { buildClipEmbeddingText } from "./gemini.js";
-import { detectScenes, extractClip, extractKeyframes, extractSceneAudio, extractThumbnail, probeVideo } from "./ffmpeg.js";
+import { detectScenes, extractAudioClip, extractClip, extractKeyframes, extractSceneAudio, extractThumbnail, extractWaveformThumbnail, probeVideo } from "./ffmpeg.js";
 import { capScenesEvenlyToCount, cropFilterFor, fitScenesToDurationBand, normalizeWindows } from "./hunt.js";
 import { refineScenesWithGuidance } from "./refine.js";
 /** Turn one scene into a Clip. Throws only on unrecoverable extraction failure. */
@@ -46,22 +46,34 @@ export async function processSceneToClip(input) {
         scene,
         guidancePrompt: ctx.guidancePrompt
     });
-    // 4. Cut the clip + thumbnail (cropped to the target aspect when requested).
-    const clipLocal = path.join(ctx.clipsDir, `${clipId}.mp4`);
-    const thumbLocal = path.join(ctx.thumbsDir, `${clipId}.jpg`);
+    // 4. Cut the clip + thumbnail. Audio-only hunts emit an .m4a (no video) with a
+    //    waveform PNG thumbnail; ordinary hunts emit an .mp4 cropped to the target
+    //    aspect when requested.
+    const audioOnly = ctx.audioOnly === true;
+    const clipLocal = path.join(ctx.clipsDir, `${clipId}.${audioOnly ? "m4a" : "mp4"}`);
+    const thumbLocal = path.join(ctx.thumbsDir, `${clipId}.${audioOnly ? "png" : "jpg"}`);
     let extractDetail = "";
-    const clipOk = await extractClip({
-        videoPath,
-        scene,
-        outPath: clipLocal,
-        reencode: ctx.reencodeClips !== false,
-        cropFilter,
-        onError: (detail) => (extractDetail = detail)
-    });
-    if (!clipOk) {
-        throw new Error(`ffmpeg failed to extract clip for scene @${scene.start_time_sec}s${extractDetail ? ` — ${extractDetail}` : ""}`);
+    if (audioOnly) {
+        const audioOk = await extractAudioClip({ videoPath, scene, outPath: clipLocal, onError: (detail) => (extractDetail = detail) });
+        if (!audioOk) {
+            throw new Error(`ffmpeg failed to extract audio for scene @${scene.start_time_sec}s${extractDetail ? ` — ${extractDetail}` : ""}`);
+        }
+        await extractWaveformThumbnail({ videoPath, scene, outPath: thumbLocal }).catch(() => false);
+    }
+    else {
+        const clipOk = await extractClip({
+            videoPath,
+            scene,
+            outPath: clipLocal,
+            reencode: ctx.reencodeClips !== false,
+            cropFilter,
+            onError: (detail) => (extractDetail = detail)
+        });
+        if (!clipOk) {
+            throw new Error(`ffmpeg failed to extract clip for scene @${scene.start_time_sec}s${extractDetail ? ` — ${extractDetail}` : ""}`);
+        }
+        await extractThumbnail({ videoPath, scene, outPath: thumbLocal, cropFilter });
     }
-    await extractThumbnail({ videoPath, scene, outPath: thumbLocal, cropFilter });
     // 5. Embed the document text for semantic search (skipped when the client can't embed).
     let embedding;
     if (ctx.client.canEmbed) {
@@ -84,7 +96,8 @@ export async function processSceneToClip(input) {
         tier: ctx.client.tier,
         provider: ctx.client.provider,
         embedding_model: ctx.client.embeddingModelId ?? undefined,
-        aspect: cropFilter && ctx.crop ? ctx.crop.aspect : undefined,
+        aspect: !audioOnly && cropFilter && ctx.crop ? ctx.crop.aspect : undefined,
+        media_kind: audioOnly ? "audio" : undefined,
         owner_id: input.ownerId,
         created_at: new Date().toISOString()
     };

package/dist/src/services/elevenlabs.js

@@ -0,0 +1,222 @@
+// ElevenLabs audio provider (music, TTS, STT) — shared by the ProviderService
+// wallet/BYOK router and the primitive handlers. ElevenLabs is a SPECIAL
+// audio-only provider: it never counts as a qualified text/image/video key.
+// Two ways a call reaches here:
+//   1. Platform key (config.ELEVENLABS_API_KEY) + wallet billing — the default
+//      "use_wallet_credits" path, so good speech/music work with zero setup.
+//   2. The customer's own saved ElevenLabs key (use_wallet_credits: false) — no
+//      wallet charge; identity/voices are scoped to THEIR ElevenLabs account.
+//
+// Keep the raw request/response shapes here — one implementation for cloud and
+// devcli, mirroring services/speech.ts for the openai/gemini/openrouter family.
+import { ProviderAuthError, rateLimitOrQuotaError } from "./provider-errors.js";
+import { pcm16MonoToWav } from "./speech.js";
+const ELEVENLABS_BASE_URL = "https://api.elevenlabs.io/v1";
+// Sensible defaults so a bare `music`/`tts` call just works.
+export const DEFAULT_ELEVENLABS_TTS_MODEL = "eleven_multilingual_v2";
+export const DEFAULT_ELEVENLABS_STT_MODEL = "scribe_v1";
+export const DEFAULT_ELEVENLABS_MUSIC_MODEL = "music_v2";
+// George — a warm, versatile narrative voice. Callers can override with any
+// voice_id from `elevenLabsListVoices` (platform or their own account).
+export const DEFAULT_ELEVENLABS_VOICE_ID = "JBFqnCBsd6RMkjVDRZzb";
+export const ELEVENLABS_VOICE_LIBRARY_URL = "https://elevenlabs.io/app/voice-library";
+// A tiny friendly-name → voice_id map so agents/users can say "sarah" instead of
+// an opaque id. Unknown values are passed through untouched (treated as a
+// voice_id), and the real catalog is always `elevenLabsListVoices`.
+const ELEVENLABS_VOICE_PRESETS = {
+    george: "JBFqnCBsd6RMkjVDRZzb",
+    sarah: "EXAVITQu4vr4xnSDxMaL",
+    daniel: "onwK4e9ZLuTAKqWW03F9",
+    charlotte: "XB0fDUnXU5powFXDhCwa",
+    rachel: "21m00Tcm4TlvDq8ikWAM"
+};
+/** Resolve a caller-supplied voice (friendly preset name or raw id) to a voice_id. */
+export function resolveElevenLabsVoiceId(voice) {
+    const raw = voice?.trim();
+    if (!raw) {
+        return DEFAULT_ELEVENLABS_VOICE_ID;
+    }
+    return ELEVENLABS_VOICE_PRESETS[raw.toLowerCase()] ?? raw;
+}
+function authHeaders(apiKey) {
+    return { "xi-api-key": apiKey };
+}
+async function throwForStatus(provider, operation, response) {
+    if (response.status === 401 || response.status === 403) {
+        throw new ProviderAuthError("elevenlabs authentication failed");
+    }
+    if (response.status === 429) {
+        throw await rateLimitOrQuotaError(provider, response);
+    }
+    const details = await response.text().catch(() => "");
+    throw new Error(`elevenlabs ${operation} returned ${response.status}${details ? `: ${details.slice(0, 400)}` : ""}`);
+}
+/** Cheap credential probe — the ElevenLabs equivalent of a chat-completion validate. */
+export async function validateElevenLabsKey(apiKey) {
+    const response = await fetch(`${ELEVENLABS_BASE_URL}/user`, { headers: authHeaders(apiKey) });
+    if (response.ok) {
+        return;
+    }
+    await throwForStatus("elevenlabs", "key validation", response);
+}
+/** List voices available to whichever account owns `apiKey` (platform or the user's). */
+export async function elevenLabsListVoices(apiKey) {
+    const response = await fetch(`${ELEVENLABS_BASE_URL}/voices`, { headers: authHeaders(apiKey) });
+    if (!response.ok) {
+        await throwForStatus("elevenlabs", "list voices", response);
+    }
+    const data = (await response.json());
+    return (data.voices ?? []).map((voice) => ({
+        voice_id: String(voice.voice_id ?? ""),
+        name: String(voice.name ?? "Unnamed"),
+        category: typeof voice.category === "string" ? voice.category : null,
+        description: typeof voice.description === "string" && voice.description.trim() ? voice.description.trim() : null,
+        labels: voice.labels && typeof voice.labels === "object"
+            ? Object.fromEntries(Object.entries(voice.labels).map(([key, value]) => [key, String(value)]))
+            : {},
+        preview_url: typeof voice.preview_url === "string" ? voice.preview_url : null
+    })).filter((voice) => voice.voice_id);
+}
+/** ElevenLabs text-to-speech. `voice` is a voice_id (or a friendly preset name). */
+export async function elevenLabsTextToSpeech(input) {
+    const voiceId = resolveElevenLabsVoiceId(input.voice);
+    const model = input.model?.trim() || DEFAULT_ELEVENLABS_TTS_MODEL;
+    const wantWav = input.responseFormat === "wav";
+    // ElevenLabs has no direct "wav" output — request raw PCM and wrap it in a WAV
+    // header, mirroring the gemini/openrouter PCM bridge in services/speech.ts.
+    const outputFormat = wantWav ? "pcm_44100" : "mp3_44100_128";
+    const voiceSettings = {};
+    if (input.stability != null)
+        voiceSettings.stability = input.stability;
+    if (input.similarityBoost != null)
+        voiceSettings.similarity_boost = input.similarityBoost;
+    if (input.style != null)
+        voiceSettings.style = input.style;
+    if (input.speed != null)
+        voiceSettings.speed = input.speed;
+    const response = await fetch(`${ELEVENLABS_BASE_URL}/text-to-speech/${encodeURIComponent(voiceId)}?output_format=${outputFormat}`, {
+        method: "POST",
+        headers: { ...authHeaders(input.apiKey), "Content-Type": "application/json" },
+        body: JSON.stringify({
+            text: input.text,
+            model_id: model,
+            ...(input.languageCode ? { language_code: input.languageCode } : {}),
+            ...(Object.keys(voiceSettings).length ? { voice_settings: voiceSettings } : {})
+        })
+    });
+    if (!response.ok) {
+        await throwForStatus("elevenlabs", "text-to-speech", response);
+    }
+    const raw = Buffer.from(await response.arrayBuffer());
+    const bytes = wantWav ? pcm16MonoToWav(raw, 44100) : raw;
+    return {
+        bytes,
+        contentType: wantWav ? "audio/wav" : "audio/mpeg",
+        characterCount: input.text.length,
+        usage: { inputTokens: input.text.length, outputTokens: 0, costUsd: 0 }
+    };
+}
+/** ElevenLabs music generation from a prompt or a hand-built composition plan. */
+export async function elevenLabsMusicCompose(input) {
+    const model = input.model?.trim() || DEFAULT_ELEVENLABS_MUSIC_MODEL;
+    const outputFormat = input.outputFormat?.trim() || "mp3_44100_128";
+    const body = { model_id: model };
+    if (input.compositionPlan) {
+        body.composition_plan = input.compositionPlan;
+    }
+    else {
+        body.prompt = input.prompt;
+        body.music_length_ms = input.musicLengthMs ?? 30000;
+    }
+    const response = await fetch(`${ELEVENLABS_BASE_URL}/music?output_format=${encodeURIComponent(outputFormat)}`, {
+        method: "POST",
+        headers: { ...authHeaders(input.apiKey), "Content-Type": "application/json" },
+        body: JSON.stringify(body)
+    });
+    if (!response.ok) {
+        await throwForStatus("elevenlabs", "music generation", response);
+    }
+    return {
+        bytes: Buffer.from(await response.arrayBuffer()),
+        contentType: "audio/mpeg",
+        musicLengthMs: input.musicLengthMs ?? 30000
+    };
+}
+/** ElevenLabs speech-to-text (Scribe) — native diarization + word timestamps. */
+export async function elevenLabsSpeechToText(input) {
+    const model = input.model?.trim() || DEFAULT_ELEVENLABS_STT_MODEL;
+    const form = new FormData();
+    form.set("file", new Blob([Buffer.from(input.audio)], { type: input.contentType ?? "audio/mpeg" }), "audio");
+    form.set("model_id", model);
+    form.set("diarize", String(input.diarize !== false));
+    form.set("timestamps_granularity", "word");
+    if (input.languageCode) {
+        form.set("language_code", input.languageCode);
+    }
+    if (input.numSpeakers != null) {
+        form.set("num_speakers", String(input.numSpeakers));
+    }
+    const response = await fetch(`${ELEVENLABS_BASE_URL}/speech-to-text`, {
+        method: "POST",
+        headers: authHeaders(input.apiKey),
+        body: form
+    });
+    if (!response.ok) {
+        await throwForStatus("elevenlabs", "speech-to-text", response);
+    }
+    const data = (await response.json());
+    const words = Array.isArray(data.words) ? data.words : [];
+    const segments = groupElevenLabsWordsIntoSegments(words);
+    const speakers = new Set(segments.map((segment) => segment.speaker));
+    const audioDurationSec = words.reduce((max, word) => Math.max(max, Number(word.end ?? 0)), 0) || null;
+    return {
+        text: String(data.text ?? "").trim(),
+        language: typeof data.language_code === "string" && data.language_code.trim() ? data.language_code.trim() : null,
+        segments,
+        diarization: speakers.size > 1 ? "diarized" : "none",
+        usage: { inputTokens: 0, outputTokens: 0, costUsd: 0 },
+        audioDurationSec
+    };
+}
+// Fold ElevenLabs' flat word stream (each tagged with a speaker_id) into
+// speaker-turn segments carrying word-level timings — the shape STT callers and
+// the caption builder already consume.
+function groupElevenLabsWordsIntoSegments(words) {
+    const segments = [];
+    let current = null;
+    let currentWords = [];
+    const flush = () => {
+        if (current) {
+            current.text = current.text.trim();
+            if (currentWords.length) {
+                current.words = currentWords;
+            }
+            if (current.text) {
+                segments.push(current);
+            }
+        }
+        current = null;
+        currentWords = [];
+    };
+    for (const word of words) {
+        if (word.type && word.type !== "word" && word.type !== "spacing") {
+            continue;
+        }
+        const speaker = word.speaker_id?.trim() || "speaker_1";
+        const text = String(word.text ?? "");
+        if (!current || current.speaker !== speaker) {
+            flush();
+            current = { speaker, start_sec: Number(word.start ?? 0), end_sec: Number(word.end ?? 0), text: "" };
+            currentWords = [];
+        }
+        current.text += text;
+        current.end_sec = Number(word.end ?? current.end_sec ?? 0);
+        const trimmed = text.trim();
+        if (trimmed && Number.isFinite(Number(word.start)) && Number.isFinite(Number(word.end))) {
+            currentWords.push({ text: trimmed, start_sec: Number(word.start), end_sec: Number(word.end) });
+        }
+    }
+    flush();
+    return segments;
+}
+//# sourceMappingURL=elevenlabs.js.map