@mevdragon/vidfarm-devcli 0.19.1 → 0.20.1

package/dist/src/cli.js

@@ -285,6 +285,8 @@ Raws (the third library — mine long-form video into a reusable raws store):
     --tracer <name>           Group this session's clips (also the default folder)
     --folder <name>           Save the clip into /raws/<name>
     --name <text>             Human name for the clip
+    --audio-only              Save just the audio of the subrange as an .m4a raw
+                            (needs --start/--end; e.g. lift a copyright-free track)
                             (multi-clip: re-run with a new range + same --tracer)
 
 Speech (TTS/STT) — LOCAL-FIRST on your own AI key; --cloud is the explicit backup:
@@ -300,7 +302,19 @@ Speech (TTS/STT) — LOCAL-FIRST on your own AI key; --cloud is the explicit bac
     --engine local            KEYLESS local Kokoro-82M engine (free, no account) — fixed voice
                               presets (af_heart, am_adam, bf_emma, …), no --style. When NO
                               provider key is found the default byok engine auto-falls back here.
-    --cloud                   BACKUP: saved-key platform job   → POST /api/v1/primitives/audio/speech (+ poll)
+    --cloud                   BACKUP: platform job → POST /api/v1/primitives/audio/speech (+ poll).
+                              Defaults to ElevenLabs on the platform key (wallet-billed); --voice
+                              takes an ElevenLabs voice_id (see: vidfarm voices).
+    --own-key                 (--cloud) use your own ElevenLabs / BYOK key instead of the wallet
+    --list-voices             List ElevenLabs voices and exit (same as: vidfarm voices)
+  music "<prompt>"          Prompt → music track (ElevenLabs).            CLOUD-ONLY (no local engine)
+    --length <sec> | --length-ms <ms>   Duration (default 30s, max 5 min)
+    --out <file>              Output mp3 path (default music-<id>.mp3)
+    --own-key                 Use your own ElevenLabs key instead of the platform key + wallet
+                              (default: platform ElevenLabs, wallet-billed) → POST /api/v1/primitives/music/generate
+  voices                    List ElevenLabs voices for TTS (voice_id + name + labels).
+    --own-key                 List voices on your own saved ElevenLabs key (default: platform account)
+    --limit <n> | --json      Show more / raw JSON
   stt <file|url>            Video or audio → transcript (alias: transcribe).  LOCAL: local ffmpeg demux +
                             Returns BOTH formats: the simple subtitle    your provider key (gemini labels
                             version (plain text + timed SRT cues) and    speakers; openai/openrouter give
@@ -315,8 +329,11 @@ Speech (TTS/STT) — LOCAL-FIRST on your own AI key; --cloud is the explicit bac
                               timestamps (words[] in the .json/--json output), no speaker labels.
                               When NO provider key is found the default byok engine auto-falls
                               back here. --model then picks the whisper model (tiny.en…large-v3).
-    --cloud                   BACKUP: saved-key platform job   → POST /api/v1/primitives/audio/transcribe (+ poll)
+    --cloud                   BACKUP: platform job → POST /api/v1/primitives/audio/transcribe (+ poll).
+                              Defaults to ElevenLabs Scribe on the platform key (wallet-billed;
+                              native diarization + real word timings).
                               (a local file uploads to your temp folder first — 30-day TTL)
+    --own-key                 (--cloud) use your own ElevenLabs / BYOK key instead of the wallet
 
 Animated captions (word-by-word TikTok/CapCut styles on a pulled/served composition):
   captions generate <dir>   Transcribe the narration & lay down animated  LOCAL: local ffmpeg demux +
@@ -586,6 +603,12 @@ async function main() {
         case "tts":
             await runTtsCommand(rest);
             return;
+        case "music":
+            await runMusicCommand(rest);
+            return;
+        case "voices":
+            await runVoicesCommand(rest);
+            return;
         case "stt":
         case "transcribe":
             await runSttCommand(rest);
@@ -3045,10 +3068,17 @@ async function runTtsCommand(argv) {
             "gemini-key": { type: "string" },
             "openai-key": { type: "string" },
             "openrouter-key": { type: "string" },
+            "own-key": { type: "boolean", default: false },
+            "list-voices": { type: "boolean", default: false },
             "no-wait": { type: "boolean", default: false },
             tracer: { type: "string" }
         }
     });
+    // --list-voices: browse the ElevenLabs voice catalog and exit (cloud call).
+    if (parsed.values["list-voices"]) {
+        await runVoicesCommand(parsed.values["own-key"] ? ["--own-key"] : []);
+        return;
+    }
     const text = (parsed.values.text ?? parsed.positionals.join(" ")).trim();
     if (!text)
         throw new Error('tts requires text: vidfarm tts "Welcome back to the channel" --style "calm, warm narrator".');
@@ -3118,7 +3148,9 @@ async function runTtsCommand(argv) {
         }
         return;
     }
-    // BACKUP: the platform primitive route (saved provider keys, async job).
+    // BACKUP: the platform primitive route (async job). Defaults to high-quality
+    // ElevenLabs on the platform key, billed to the wallet (use_wallet_credits).
+    // --own-key flips to your own saved ElevenLabs / BYOK provider key instead.
     const ctx = commonContext(parsed.values);
     const payload = { text, output_format: format };
     if (voice)
@@ -3129,6 +3161,11 @@ async function runTtsCommand(argv) {
         payload.provider = parsed.values.provider;
     if (parsed.values.model)
         payload.model = parsed.values.model;
+    if (parsed.values["own-key"])
+        payload.use_wallet_credits = false;
+    if (!ctx.json && !parsed.values["own-key"]) {
+        console.log(`${DIM}Using ElevenLabs on the platform key (wallet-billed). Browse voices: vidfarm voices — then pass --voice <voice_id>. Add --own-key to use your own key.${RESET}`);
+    }
     const tracer = parsed.values.tracer ?? `devcli-tts-${Date.now().toString(36)}`;
     const submit = await apiRequest({ method: "POST", host: ctx.host, path: "/api/v1/primitives/audio/speech", auth: ctx.auth, body: { tracer, payload } });
     assertApiOk(submit, "tts");
@@ -3290,6 +3327,112 @@ function emitSttResult(input) {
         console.log(`${GREEN}Wrote ${dest}${RESET}`);
     }
 }
+// Music generation (ElevenLabs) — cloud-only (no local engine). Defaults to the
+// platform key billed to the wallet; --own-key uses your saved ElevenLabs key.
+async function runMusicCommand(argv) {
+    const parsed = parseArgs({
+        args: argv,
+        allowPositionals: true,
+        options: {
+            ...commonOptions(),
+            prompt: { type: "string" },
+            "length-ms": { type: "string" },
+            length: { type: "string" },
+            model: { type: "string" },
+            out: { type: "string" },
+            "own-key": { type: "boolean", default: false },
+            "no-wait": { type: "boolean", default: false },
+            tracer: { type: "string" }
+        }
+    });
+    const prompt = (parsed.values.prompt ?? parsed.positionals.join(" ")).trim();
+    if (!prompt)
+        throw new Error('music requires a prompt: vidfarm music "upbeat lo-fi hip hop beat with jazzy piano" --length 30');
+    const ctx = commonContext(parsed.values);
+    let musicLengthMs;
+    const lengthMs = parsed.values["length-ms"]?.trim();
+    const lengthSec = parsed.values.length?.trim();
+    if (lengthMs)
+        musicLengthMs = Math.round(Number(lengthMs));
+    else if (lengthSec)
+        musicLengthMs = Math.round(Number(lengthSec) * 1000);
+    const payload = { prompt };
+    if (musicLengthMs && Number.isFinite(musicLengthMs))
+        payload.music_length_ms = musicLengthMs;
+    if (parsed.values.model)
+        payload.model = parsed.values.model;
+    if (parsed.values["own-key"])
+        payload.use_wallet_credits = false;
+    if (!ctx.json && !parsed.values["own-key"]) {
+        console.log(`${DIM}Composing with ElevenLabs on the platform key (wallet-billed). Add --own-key to use your own ElevenLabs key.${RESET}`);
+    }
+    const tracer = parsed.values.tracer ?? `devcli-music-${Date.now().toString(36)}`;
+    const submit = await apiRequest({ method: "POST", host: ctx.host, path: "/api/v1/primitives/music/generate", auth: ctx.auth, body: { tracer, payload } });
+    assertApiOk(submit, "music");
+    const jobId = submit.json?.job_id;
+    if (parsed.values["no-wait"] || !jobId) {
+        if (!ctx.json && jobId)
+            console.log(`${DIM}Queued ${jobId} (tracer ${tracer}). Poll: vidfarm api GET /api/v1/user/me/jobs/${jobId} — or drop --no-wait to auto-poll.${RESET}`);
+        emitResult(submit, ctx.json);
+        return;
+    }
+    if (!ctx.json)
+        console.log(`${DIM}Generating music (${jobId})… polling every 5s.${RESET}`);
+    const job = await pollPrimitiveJob(ctx, jobId);
+    const mediaUrl = resolveJobMediaUrl(job);
+    if (!mediaUrl) {
+        if (ctx.json) {
+            printJson({ ok: false, job_id: jobId, status: String(job?.status ?? ""), job });
+        }
+        else {
+            console.log(`${RED}Music ${String(job?.status ?? "did not finish")} — no audio URL.${RESET}`);
+            printJson(job);
+        }
+        process.exitCode = 1;
+        return;
+    }
+    const outPath = path.resolve(process.cwd(), parsed.values.out ?? `music-${jobId}.mp3`);
+    await downloadUrlToFile(mediaUrl, outPath);
+    if (ctx.json) {
+        printJson({ ok: true, job_id: jobId, tracer, audio_url: mediaUrl, out: outPath });
+    }
+    else {
+        console.log(`${GREEN}Wrote ${outPath}${RESET} ${DIM}(${mediaUrl})${RESET}`);
+        console.log(`${DIM}Place it as its own <audio> layer under narration (~0.1–0.2 volume).${RESET}`);
+    }
+}
+// List ElevenLabs voices for TTS. --own-key lists the voices on your own saved
+// ElevenLabs key; default lists the platform account's voices.
+async function runVoicesCommand(argv) {
+    const parsed = parseArgs({
+        args: argv,
+        allowPositionals: true,
+        options: {
+            ...commonOptions(),
+            "own-key": { type: "boolean", default: false },
+            limit: { type: "string" }
+        }
+    });
+    const ctx = commonContext(parsed.values);
+    const query = parsed.values["own-key"] ? "?use_wallet_credits=false" : "";
+    const res = await apiRequest({ method: "GET", host: ctx.host, path: `/api/v1/primitives/audio/voices${query}`, auth: ctx.auth });
+    assertApiOk(res, "voices");
+    const body = (res.json ?? {});
+    const voices = Array.isArray(body.voices) ? body.voices : [];
+    if (ctx.json) {
+        printJson(body);
+        return;
+    }
+    console.log(`${DIM}${voices.length} ElevenLabs voices (${body.scope ?? "platform"} account). Browse more: ${body.voice_library_url ?? "https://elevenlabs.io/app/voice-library"}${RESET}`);
+    const limit = Number(parsed.values.limit ?? "40") || 40;
+    for (const voice of voices.slice(0, limit)) {
+        const labels = voice.labels && typeof voice.labels === "object" ? Object.values(voice.labels).join(", ") : "";
+        console.log(`${GREEN}${voice.voice_id}${RESET}  ${voice.name}${labels ? ` ${DIM}(${labels})${RESET}` : ""}`);
+    }
+    if (voices.length > limit)
+        console.log(`${DIM}… ${voices.length - limit} more (raise with --limit or use --json).${RESET}`);
+    console.log(`${DIM}Use one: vidfarm tts "Hello" --cloud --voice <voice_id>${RESET}`);
+}
 async function runSttCommand(argv) {
     const parsed = parseArgs({
         args: argv,
@@ -3307,6 +3450,7 @@ async function runSttCommand(argv) {
             "gemini-key": { type: "string" },
             "openai-key": { type: "string" },
             "openrouter-key": { type: "string" },
+            "own-key": { type: "boolean", default: false },
             "no-wait": { type: "boolean", default: false },
             tracer: { type: "string" },
             folder: { type: "string" }
@@ -3405,6 +3549,11 @@ async function runSttCommand(argv) {
         payload.provider = parsed.values.provider;
     if (parsed.values.model)
         payload.model = parsed.values.model;
+    // Cloud STT defaults to ElevenLabs Scribe on the platform key (native
+    // diarization + real word timings), wallet-billed. --own-key uses your own
+    // ElevenLabs / BYOK key instead.
+    if (parsed.values["own-key"])
+        payload.use_wallet_credits = false;
     const tracer = parsed.values.tracer ?? `devcli-stt-${Date.now().toString(36)}`;
     const submit = await apiRequest({ method: "POST", host: ctx.host, path: "/api/v1/primitives/audio/transcribe", auth: ctx.auth, body: { tracer, payload } });
     assertApiOk(submit, "stt");
@@ -3953,7 +4102,8 @@ async function runClipperCommand(argv) {
             out: { type: "string" },
             tracer: { type: "string" },
             folder: { type: "string" },
-            name: { type: "string" }
+            name: { type: "string" },
+            "audio-only": { type: "boolean", default: false }
         }
     });
     const source = parsed.positionals[0];
@@ -3970,6 +4120,10 @@ async function runClipperCommand(argv) {
     if (!wholeVideo && (!hasStart || !hasEnd)) {
         throw new Error("clipper needs BOTH --start and --end for an exact subrange, or NEITHER to import the whole video.");
     }
+    const audioOnly = Boolean(parsed.values["audio-only"]);
+    if (audioOnly && wholeVideo) {
+        throw new Error("--audio-only needs an exact subrange (--start <t> --end <t>). Whole-video import keeps the video track.");
+    }
     const body = {};
     if (parsed.values.tracer)
         body.tracer = String(parsed.values.tracer);
@@ -3977,6 +4131,8 @@ async function runClipperCommand(argv) {
         body.folder_path = String(parsed.values.folder);
     if (parsed.values.name)
         body.name = String(parsed.values.name);
+    if (audioOnly)
+        body.audio_only = true;
     // Resolve the source into the body once: a URL rides as source_url; a local
     // file is staged into the temp store and referenced by temp_file_id.
     if (/^https?:\/\//i.test(source)) {
@@ -4037,7 +4193,8 @@ async function runClipperCommand(argv) {
     }
     const j = res.json;
     const dest = j?.folder_path ? `/raws/${j.folder_path}` : "your raws library";
-    console.log(`${GREEN}Clipped ${(j?.duration_sec ?? (endSec - startSec)).toFixed?.(2) ?? j?.duration_sec}s → ${dest} (clip ${j?.clip_id ?? "?"}).${RESET}`);
+    const kindWord = (j?.media_kind === "audio" || audioOnly) ? "audio" : "clip";
+    console.log(`${GREEN}Clipped ${kindWord} ${(j?.duration_sec ?? (endSec - startSec)).toFixed?.(2) ?? j?.duration_sec}s → ${dest} (clip ${j?.clip_id ?? "?"}).${RESET}`);
     if (j?.view_url)
         console.log(`${DIM}Preview: ${new URL(j.view_url, ctx.host).toString()}${RESET}`);
     console.log(`${DIM}Grab another range from the same source with the same --tracer/--folder.${RESET}`);

package/dist/src/config.js

@@ -66,6 +66,18 @@ const schema = z.object({
     OPENROUTER_API_KEY: z.string().optional(),
     GEMINI_API_KEY: z.string().optional(),
     PERPLEXITY_API_KEY: z.string().optional(),
+    // Platform ElevenLabs key — the fallback used for music / TTS / STT primitives
+    // when the caller opts into `use_wallet_credits` (the default) or has no usable
+    // BYOK audio key. Usage on this key is billed to the customer's vidfarm wallet.
+    ELEVENLABS_API_KEY: z.string().optional(),
+    // Base provider costs for ElevenLabs primitives billed to the wallet (the
+    // billing lambda applies the standard markup on top). Defaults track published
+    // ElevenLabs list pricing: TTS multilingual v2/v3 $0.10 / 1K chars (flash/turbo
+    // is $0.05), Music $0.15 / minute, Scribe STT $0.22 / hour ≈ $0.003667 / min.
+    // Override per-environment as pricing shifts.
+    ELEVENLABS_MUSIC_USD_PER_MIN: z.coerce.number().min(0).default(0.15),
+    ELEVENLABS_TTS_USD_PER_1K_CHARS: z.coerce.number().min(0).default(0.10),
+    ELEVENLABS_STT_USD_PER_MIN: z.coerce.number().min(0).default(0.003667),
     GHOSTCUT_KEY: z.string().optional(),
     GHOSTCUT_SECRET: z.string().optional(),
     GHOSTCUT_USD_PER_30S: z.coerce.number().min(0).default(0.10),

package/dist/src/devcli/clips.js

@@ -44,6 +44,7 @@ export const CLIPS_HELP = `vidfarm raws — build & search a local raws library
     --yes                          Skip the >$1 confirmation prompt
     --frames <n>                   Keyframes per scene for tagging (1-3, default 2)
     --no-audio                     Skip per-scene audio (Gemini only; no transcripts; cheaper)
+    --audio-only                   Save each mined clip as an audio-only .m4a (no video); waveform thumb
     --concurrency <n>              Parallel scenes (default 3; agent default 2)
     --threshold <0..1>             Scene-cut sensitivity (default 0.3; lower = more cuts)
     --min-scene <sec>              Merge scenes shorter than this (default 0.6)
@@ -133,6 +134,7 @@ async function runScan(argv) {
             "allow-text": { type: "boolean" },
             "max-clips": { type: "string" },
             "no-refine": { type: "boolean" },
+            "audio-only": { type: "boolean" },
             cloud: { type: "boolean" },
             url: { type: "string" },
             tracer: { type: "string" },
@@ -197,7 +199,8 @@ async function runScan(argv) {
             aspect,
             cropFocus,
             avoidText,
-            maxClips
+            maxClips,
+            audioOnly: Boolean(values["audio-only"])
         });
     }
     const videoArg = positionals[0];
@@ -336,6 +339,7 @@ async function runScan(argv) {
         guidancePrompt: guidance,
         refineSegmentation: !values["no-refine"],
         reencodeClips: true,
+        audioOnly: Boolean(values["audio-only"]),
         crop: aspect && aspect !== "original" ? { aspect, focus: cropFocus } : undefined
     };
     console.log(`[raws] scanning ${scenes.length} scenes (${provider}/${activeClient.tagModelId}, ${concurrency}x) …`);
@@ -466,7 +470,8 @@ async function runScanCloud(input) {
                 // Send explicitly (incl. false) so --allow-text can override the
                 // server-side avoid-text default.
                 avoid_text: input.avoidText,
-                ...(input.maxClips ? { max_clips: input.maxClips } : {})
+                ...(input.maxClips ? { max_clips: input.maxClips } : {}),
+                ...(input.audioOnly ? { audio_only: true } : {})
             }
         })
     });

package/dist/src/domain.js

@@ -1,2 +1,5 @@
+// Providers that count as a qualified AI key for text/image/video generation
+// (everything except the audio-only elevenlabs key).
+export const QUALIFIED_AI_PROVIDER_TYPES = ["openai", "gemini", "openrouter", "perplexity"];
 export const CUSTOMER_PLAN_TIERS = ["client", "standalone", "agency"];
 //# sourceMappingURL=domain.js.map

package/dist/src/editor-chat.js

@@ -21,6 +21,10 @@ export function defaultEditorChatModel(provider) {
             return "qwen/qwen3.6-plus";
         case "perplexity":
             return "sonar";
+        default:
+            // Non-text providers (e.g. audio-only elevenlabs) never reach editor chat;
+            // return a safe text default rather than undefined.
+            return "gpt-5.4-mini";
     }
 }
 export function buildTemplateEditorChatSystemPrompt(input) {
@@ -75,7 +79,7 @@ export function buildTemplateEditorChatSystemPrompt(input) {
         "FINE TIMELINE CONTROL — you have trackpad-level verbs; use them instead of rewriting the whole document for small changes: (1) set_layer_keyframes authors a custom, script-free CSS @keyframes animation on ONE layer that previews AND renders — pass layer_key + keyframes:[{offset(0..1), opacity?, translate_x?, translate_y?, scale?, rotate?}, …] (>=2 stops), optional keyframe_easing (linear/ease/ease-in/ease-out/ease-in-out/cubic-bezier(...)) and keyframe_duration (seconds, default = clip duration). translate_x/translate_y are percentages of the layer's OWN box. Example fly-up-and-fade-in: keyframes=[{offset:0,opacity:0,translate_y:40},{offset:1,opacity:1,translate_y:0}]. This is the durable way to hand-author motion beyond the Ken Burns / transition / caption presets — the JS runtime adapters (anime.js/GSAP/Lottie) are devcli-only. (2) nudge_layers shifts one or more layers by a RELATIVE delta_start (seconds) and/or delta_track (lanes) — pass layer_key or layer_keys; grouped clips move together. (3) ripple_edit inserts (delta_start>0) or closes (delta_start<0) time at at_time and shifts every downstream clip so the rest of the timeline stays in sync — use it to make room before an insert or to close a gap you can see in timeline_gaps. (4) trim_layer moves ONE edge to a time: edge='start' (left; advances the in-point, and for video/audio pushes the media start so the visible frame is unchanged) or edge='end' (right; changes duration), with to_time in composition seconds. (5) set_layer_zindex restacks: z_order='front'|'back'|'forward'|'backward' (stacking == track index; higher draws on top) or an explicit track. All of these auto-resolve track collisions and note it in the summary; read the NEXT editor_context to confirm the new start/duration/track/animation landed.",
         "If the user asks to export, render, publish, or produce the final MP4 for this composition (phrases like 'export it', 'render this', 'kick off the export', 'make the mp4'), call editor_action with action_type=export_composition and a brief explanation. This is the same action as the editor's Export button — it queues the composition on production AWS Lambda. Do NOT try to hit any /api/v1/compositions/:id/export route via http_request for this; use the editor_action tool. After calling export_composition, tell the user the export has started and that the finished MP4 URL will appear once the render succeeds; do not fabricate a URL or claim success until a later turn shows last_export_url populated in editor_context.",
         "CONFIRM BEFORE RENDERING — DO NOT auto-render. Rendering the final MP4 is a slow, costly, screen-taking action, so by DEFAULT you must NOT call export_composition on your own initiative. After you finish an edit (a swap, a re-theme, a scene rebuild), STOP and let the user review the live preview, then ASK a short question like 'Happy with the preview? Want me to render the final MP4?' and wait for a yes. Only fire export_composition when EITHER (a) the user explicitly asks to render/export/'make the mp4' in that turn, or (b) the user has told you up front to just render without asking (e.g. 'edit and render it', 'no need to check, just export'). If in doubt, ask — never surprise the user with a render. Never render merely because you judged the edit 'done'.",
-        "MEDIA SWAP — TARGET A REAL LAYER KEY, NOT A SCENE NAME. To swap a clip's media with set_layer_media, the layer_key MUST be an EXACT key (or its slug) from editor_context.layers — that array lists every targetable layer with its key, slug, kind (video/image/audio), and current src. Read it and copy the real key of the video/image layer you mean; do NOT invent a semantic name (like 'growth_examples' or 'mistakes_reflection') unless that exact string appears as a layer key/slug in editor_context, and never target a scene label/group when you mean the video inside it. If set_layer_media returns 'No change applied' or 'Layer not found', the error now lists the actual media layers — re-read it and retry with one of those exact keys. NEVER report a swap as done until the NEXT editor_context shows the layer's src changed.",
+        "MEDIA SWAP — TARGET A REAL LAYER KEY, NOT A SCENE NAME. To swap a clip's media with set_layer_media, the layer_key MUST be an EXACT key (or its slug) from editor_context.layers — that array lists every targetable layer with its key, slug, kind (video/image/audio), and current src. Read it and copy the real key of the video/image layer you mean; do NOT invent a semantic name (like 'growth_examples' or 'mistakes_reflection') unless that exact string appears as a layer key/slug in editor_context, and never target a scene label/group when you mean the video inside it. If set_layer_media returns 'No change applied' or 'Layer not found', the error now lists the actual media layers — re-read it and retry with one of those exact keys. This applies to EVERY set_layer_media field (src, object_fit, object_position, volume, muted, ken_burns, transitions), not just src: the layer_key you used to add a layer (e.g. add_layer layer_key='vf-fishing-snippet-1') becomes an 'element_'-prefixed key on the timeline, so target the real key/slug shown in editor_context, not the bare name you passed earlier. NEVER report a swap or fit/position change as done while any editor_action in this turn came back with a ⚠ / 'Layer not found' error — those calls did NOT apply; re-read the listed keys, retry against a real one, and only confirm success once the action returns ✓ (and, for src, the NEXT editor_context shows the layer changed). Do not claim you changed layers you never successfully targeted.",
         "GETTING THE MEDIA URL IS EASY — USE browse_files. When the user says 'use my raws' / 'it's in my file directory' / 'the fishing clips', you do NOT need them to paste a URL. Call browse_files action=search (or list path='/raws') to find the clip, then action=read with its file_id — the read result's view_url IS the public, ready-to-use media URL. Drop that view_url straight into set_layer_media src (or add_layer src). Do not stall asking the user for a URL you can look up yourself, and never pass a placeholder like 'please-ignore' as src.",
         "The editor_context block includes last_export_url, last_export_title, last_export_status, and last_export_render_id when a render has succeeded in this session. Treat last_export_url as the current 'finished Export MP4' for this composition. If the user asks to approve, package, share, or schedule the exported video without providing an explicit URL, use last_export_url as the primary MP4 media (kind=video, role=primary) for POST /api/v1/approved/posts. If last_export_url is null, tell the user to Export first (or offer to call editor_action with action_type=export_composition on their behalf) instead of guessing a URL.",
         "Each layer in editor_context includes element_id (data-hf-id), optional slug (data-hf-slug), and optional note (data-hf-note). layer_key on any editor_action may be EITHER the element_id or the slug — prefer the slug when the user (or a viral DNA blurb) refers to a layer by its human name like 'the hero_image'. Use set_layer_identity to add/change a layer's slug or note. Slugs are the stable AI-friendly handle: when planning viral DNA notes, reference the element by its slug (e.g., 'raise the {{hero_image}} to full canvas at 4s') so future turns can resolve it deterministically.",

package/dist/src/frontend/file-directory.js

@@ -501,20 +501,23 @@ export function createDirectoryExplorer(mount, opts = {}) {
             return;
         selbar.innerHTML = "";
         selbar.appendChild(el("span", "rk-dir-selcount", n + " selected"));
-        if (mode === "attach") {
-            const att = el("button", "rk-dir-selbtn", "Attach");
-            att.type = "button";
-            att.addEventListener("click", () => { emitAttach(Array.from(selection.values())); clearSelection(); });
-            selbar.appendChild(att);
-        }
-        const cp = el("button", "rk-dir-selbtn", "Copy paths");
+        const picked = () => Array.from(selection.values());
+        // Primary action stays on the bar; everything else tucks into a "⋯" menu so
+        // the toolbar never overflows in the narrow editor dock.
+        const cp = el("button", "rk-dir-selbtn is-primary", "Copy paths");
         cp.type = "button";
-        cp.addEventListener("click", () => copyPath(Array.from(selection.values()).map((s) => s.path).join("\n")));
-        const clr = el("button", "rk-dir-selbtn is-ghost", "Clear");
-        clr.type = "button";
-        clr.addEventListener("click", clearSelection);
+        cp.addEventListener("click", () => copyPath(picked().map((s) => s.path).join("\n")));
         selbar.appendChild(cp);
-        selbar.appendChild(clr);
+        const more = [];
+        if (mode === "attach")
+            more.push({ label: "Attach", run: () => { emitAttach(picked()); clearSelection(); } });
+        more.push({ label: "Copy paths", run: () => copyPath(picked().map((s) => s.path).join("\n")) });
+        more.push({ label: "Clear selection", run: clearSelection });
+        const menuBtn = el("button", "rk-dir-selbtn is-ghost rk-dir-selmore", "⋯");
+        menuBtn.type = "button";
+        menuBtn.setAttribute("aria-label", "More actions");
+        menuBtn.addEventListener("click", (e) => { e.stopPropagation(); openMenuAt(menuBtn, more); });
+        selbar.appendChild(menuBtn);
     }
     function checkbox(it) {
         const box = el("input", "rk-dir-check");
@@ -719,15 +722,27 @@ export function createDirectoryExplorer(mount, opts = {}) {
             return;
         uploading = true;
         const folder = currentFolder();
+        const root = currentRoot();
         const arr = Array.from(list);
         let done = 0, failed = 0;
+        // The server may re-home an upload (e.g. /temp files land in a dated
+        // subfolder). Track where the bytes actually landed so we can reveal them —
+        // otherwise the refresh shows the folder we uploaded FROM, which looks empty.
+        let landedFolder = null;
         dropText.textContent = "Uploading " + arr.length + " file(s)…";
+        const finish = () => {
+            uploading = false;
+            if (failed)
+                toast(failed + " of " + arr.length + " failed");
+            const target = root && landedFolder != null ? buildDirectoryPath(root, landedFolder) : null;
+            if (target && target !== currentPath)
+                navigate(target); // reveal the uploads
+            else
+                load(true);
+        };
         const next = (i) => {
             if (i >= arr.length) {
-                uploading = false;
-                if (failed)
-                    toast(failed + " of " + arr.length + " failed");
-                load(true);
+                finish();
                 return;
             }
             const fd = new FormData();
@@ -735,7 +750,13 @@ export function createDirectoryExplorer(mount, opts = {}) {
             fd.append("folder_path", folder);
             fetch(endpoint, { method: "POST", credentials: "same-origin", body: fd })
                 .then((r) => { if (!r.ok)
-                throw new Error("http " + r.status); done++; })
+                throw new Error("http " + r.status); return r.json().catch(() => null); })
+                .then((j) => {
+                done++;
+                const fp = j && ((j.file && j.file.folderPath) ?? (j.attachment && j.attachment.folderPath));
+                if (typeof fp === "string" && landedFolder == null)
+                    landedFolder = fp;
+            })
                 .catch(() => { failed++; })
                 .then(() => next(i + 1));
         };
@@ -943,7 +964,9 @@ export function openDirectoryPicker(opts = {}) {
         mode: "standalone",
         showSearch: true,
         showMultiSelect: false,
-        showUpload: false, // the picker only selects; it doesn't upload
+        // Opt-in upload: when allowed, the explorer shows its drop zone so users can
+        // add a file to the currently-open folder (Files/Temp), then click it to pick.
+        showUpload: opts.allowUpload === true,
         // In file mode a name-click resolves the pick; folders still navigate.
         primaryClick: select === "file" ? "attach" : "copyPath",
         onNavigate: (p) => syncFooter(p),

package/dist/src/primitive-context.js

@@ -198,6 +198,20 @@ export function createPrimitiveJobContext(input) {
                     workerId: input.workerId,
                     ...request
                 });
+            },
+            async generateMusic(request) {
+                return input.providers.generateMusic({
+                    customerId: input.customer.id,
+                    jobId: input.job.id,
+                    workerId: input.workerId,
+                    ...request
+                });
+            },
+            async listElevenLabsVoices(request) {
+                return input.providers.listElevenLabsVoices({
+                    customerId: input.customer.id,
+                    ...request
+                });
             }
         },
         hyperframes: {