@nodaro/shared 2.14.0 → 2.16.0

package/dist/index.cjs

@@ -1978,18 +1978,17 @@ var AUDIO_MODELS = {
   "elevenlabs-dialogue": {
     id: "elevenlabs-dialogue",
     kind: "audio",
-    modes: ["tts"],
+    // Its own mode, never "tts": the script shape (inputs[]) doesn't fit the
+    // single-text generate_speech contract, so list_models must never offer
+    // it there. The dialogue-capable MCP verb is `generate_dialogue`.
+    modes: ["dialogue"],
     family: "ElevenLabs",
     label: "ElevenLabs Dialogue v3",
     series: "ElevenLabs",
-    description: "Multi-speaker dialogue TTS \u2014 give it a script, it voices each role.",
+    description: "Multi-speaker dialogue via the direct ElevenLabs API \u2014 give it a script, it voices each role (any voice: premade, library, or cloned).",
     useCases: ["tts", "dialogue", "multi-speaker"],
-    pricing: [{ identifier: "elevenlabs-dialogue", credits: 25, note: "per 1K chars" }],
-    // Driven only via the dialogue/character-voice path (multi-speaker script
-    // shape), NOT the single-text generate_speech verb. Hide from MCP
-    // list_models so generate_speech (TTS_PROVIDERS) can't advertise it and
-    // then 400. Re-expose if a dialogue-capable MCP verb is added.
-    mcpHidden: true
+    features: ["audio-tags", "voice-cloning"],
+    pricing: [{ identifier: "elevenlabs-dialogue", credits: 25, note: "per 1K chars" }]
   },
   // ── ElevenLabs voice utilities ──
   "voice-clone": {
@@ -2548,10 +2547,10 @@ var MAX_TTS_CHARS_BY_PROVIDER = {
   // == eleven_flash_v2_5 (functionally equivalent)
   "elevenlabs-multilingual": 1e4,
   // eleven_multilingual_v2
-  "elevenlabs-v3": 3e3,
-  // conservative (official 5000 / API-reported 3000)
-  "elevenlabs-dialogue": 2e3
-  // text-to-dialogue recommended per-request max
+  "elevenlabs-v3": 5e3,
+  // official cap (probed: 5,200 chars accepted; keep the clamp)
+  "elevenlabs-dialogue": 5e3
+  // total across lines; ≤2,000 recommended for best quality
 };
 function getMaxTtsChars(provider) {
   return provider && MAX_TTS_CHARS_BY_PROVIDER[provider] || TTS_TEXT_MAX;
@@ -9917,6 +9916,65 @@ function findLocationMentionTokens(prompt, knownLocationSlugs) {
   return tokens;
 }
 
+// src/image-mention-slug.ts
+var IMAGE_SLUG_PATTERN = /^[a-z][a-z0-9-]*$/;
+function imageMentionSlug(name) {
+  return name.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/-+/g, "-").replace(/^-|-$/g, "");
+}
+function parseImageMentionToken(text) {
+  if (!text.startsWith("@")) return null;
+  let rest = text.slice(1);
+  if (rest.length === 0 || !/^[a-z]/.test(rest)) return null;
+  let lockField = {};
+  if (rest.endsWith("~nolock")) {
+    rest = rest.slice(0, -"~nolock".length);
+    lockField = { lock: false };
+  } else if (rest.endsWith("~lock")) {
+    rest = rest.slice(0, -"~lock".length);
+    lockField = { lock: true };
+  }
+  const parts = rest.split(":");
+  if (parts.length < 2 || parts.length > 3) return null;
+  const [imageSlug, indexStr, third] = parts;
+  if (!IMAGE_SLUG_PATTERN.test(imageSlug)) return null;
+  if (!/^\d+$/.test(indexStr)) return null;
+  const imageIndex = parseInt(indexStr, 10);
+  if (!Number.isInteger(imageIndex) || imageIndex < 1) return null;
+  if (parts.length === 2) return { imageSlug, imageIndex, ...lockField };
+  if (!IMAGE_SLUG_PATTERN.test(third)) return null;
+  return { imageSlug, imageIndex, role: third, ...lockField };
+}
+function findImageMentionTokens(prompt, knownImageSlugs) {
+  const tokens = [];
+  const regex = /(?:^|[^a-zA-Z0-9])(@[a-z][a-z0-9-]*:\d+(?::[a-z][a-z0-9-]*)?(?:~(?:no)?lock(?![a-z0-9-]))?)(?![:a-z0-9-])/g;
+  const knownSet = new Set(knownImageSlugs);
+  for (const match of prompt.matchAll(regex)) {
+    const token = match[1];
+    const offset = (match.index ?? 0) + (match[0].length - token.length);
+    if (/^\/[a-z]/.test(prompt.slice(offset + token.length))) continue;
+    const parsed = parseImageMentionToken(token);
+    if (parsed && knownSet.has(parsed.imageSlug)) {
+      tokens.push({ token, ...parsed, offset });
+    }
+  }
+  return tokens;
+}
+function imageMentionSlugForRef(r) {
+  if (r.source !== "wired-image" && r.source !== "manual") return null;
+  if (r.isExtraRef === true) return null;
+  if (!r.url || !r.defaultName) return null;
+  const slug = imageMentionSlug(r.defaultName);
+  return IMAGE_SLUG_PATTERN.test(slug) ? slug : null;
+}
+function knownImageSlugsFromRefs(refs) {
+  const out = /* @__PURE__ */ new Set();
+  for (const r of refs) {
+    const slug = imageMentionSlugForRef(r);
+    if (slug) out.add(slug);
+  }
+  return [...out];
+}
+
 // src/to-connected-references.ts
 function toConnectedReference(entity) {
   if (entity.kind === "character") {
@@ -9932,6 +9990,15 @@ function toConnectedReference(entity) {
       variantDisplayName: entity.variant ?? "canonical"
     };
   }
+  if (entity.kind === "image") {
+    return {
+      id: entity.id,
+      defaultName: entity.name,
+      source: "wired-image",
+      url: entity.url ?? "",
+      ...entity.description ? { description: entity.description } : {}
+    };
+  }
   if (entity.kind === "creature") {
     return {
       id: entity.id,
@@ -12818,6 +12885,7 @@ var MODE_TO_NODE = {
   "dubbing": "dubbing",
   "forced-alignment": "forced-alignment",
   "stt": "transcribe",
+  "dialogue": "text-to-dialogue",
   "video-analysis": "video-analysis"
 };
 function modelToNodeTarget(modelId) {
@@ -13217,6 +13285,15 @@ var windowSceneBase = zod.z.object({
   /** Effects on this shot's PICTURE. Absent ⇒ a clean image. */
   effects: zod.z.array(zod.z.enum(VIDEO_ANALYSIS_VISUAL_EFFECTS)).optional(),
   transitionOut: zod.z.enum(VIDEO_ANALYSIS_TRANSITIONS).optional(),
+  // Round 4 (2026-08-30, recast shot craft rev 1.5 Appendix F.8): the edit into
+  // the next scene in the analyser's OWN WORDS — a few words of what it saw
+  // ("fast blurred pan left", "fade to black", "zoom-blur warp into the tunnel")
+  // — or absent when the take continues through the change or no edit is seen
+  // (nothing asserted: the video model decides from story and refs). The closed
+  // `transitionOut` vocabulary above is LEGACY from here on: kept so stored
+  // blueprints still parse, never emitted again. Readers take
+  // `transition ?? transitionOut`.
+  transition: zod.z.string().trim().min(1).max(120).optional(),
   // Array of concurrent layers (music + speech + sfx together); [] = silence.
   audio: zod.z.array(audioLayerSchema),
   /** slotId → variationId for slots wearing a NON-default look in this scene
@@ -14226,6 +14303,7 @@ exports.extractVideoDurationFromNode = extractVideoDurationFromNode;
 exports.fieldKeyFromHandle = fieldKeyFromHandle;
 exports.filterCloneNodes = filterCloneNodes;
 exports.findCharacterMentionTokens = findCharacterMentionTokens;
+exports.findImageMentionTokens = findImageMentionTokens;
 exports.findLocationMentionTokens = findLocationMentionTokens;
 exports.findSeedance2AudioOverLimit = findSeedance2AudioOverLimit;
 exports.firstSightExtraRole = firstSightExtraRole;
@@ -14280,6 +14358,8 @@ exports.hasContiguousSegmentDurations = hasContiguousSegmentDurations;
 exports.hasFeature = hasFeature;
 exports.hexToRgbaArray = hexToRgbaArray;
 exports.humanizeSlotSid = humanizeSlotSid;
+exports.imageMentionSlug = imageMentionSlug;
+exports.imageMentionSlugForRef = imageMentionSlugForRef;
 exports.imageReferenceLimit = imageReferenceLimit;
 exports.inferMusicVideo = inferMusicVideo;
 exports.isAggregateableType = isAggregateableType;
@@ -14305,6 +14385,7 @@ exports.isVeoProvider = isVeoProvider;
 exports.isVideoAnalysisMixedTier = isVideoAnalysisMixedTier;
 exports.isVideoAnalysisTier = isVideoAnalysisTier;
 exports.jsonResultToList = jsonResultToList;
+exports.knownImageSlugsFromRefs = knownImageSlugsFromRefs;
 exports.listBoardTemplates = listBoardTemplates;
 exports.listModels = listModels;
 exports.listSlotSids = listSlotSids;
@@ -14339,6 +14420,7 @@ exports.parseAttributedDialogue = parseAttributedDialogue;
 exports.parseCharacterMentionToken = parseCharacterMentionToken;
 exports.parseGroupHandle = parseGroupHandle;
 exports.parseHandleId = parseHandleId;
+exports.parseImageMentionToken = parseImageMentionToken;
 exports.parseListExpression = parseListExpression;
 exports.parseLocationMentionToken = parseLocationMentionToken;
 exports.parseNodePresetExport = parseNodePresetExport;