@mux/ai 0.10.0 → 0.11.0

package/dist/index.js

@@ -5,7 +5,7 @@ var __export = (target, all) => {
 };
 
 // package.json
-var version = "0.10.0";
+var version = "0.11.0";
 
 // src/env.ts
 import { z } from "zod";
@@ -803,13 +803,16 @@ __export(primitives_exports, {
   getHotspotsForPlaybackId: () => getHotspotsForPlaybackId,
   getHotspotsForVideo: () => getHotspotsForVideo,
   getReadyTextTracks: () => getReadyTextTracks,
+  getShotsForAsset: () => getShotsForAsset,
   getStoryboardUrl: () => getStoryboardUrl,
   getThumbnailUrls: () => getThumbnailUrls,
   parseVTTCues: () => parseVTTCues,
   replaceCueText: () => replaceCueText,
+  requestShotsForAsset: () => requestShotsForAsset,
   secondsToTimestamp: () => secondsToTimestamp,
   splitVttPreambleAndCueBlocks: () => splitVttPreambleAndCueBlocks,
-  vttTimestampToSeconds: () => vttTimestampToSeconds
+  vttTimestampToSeconds: () => vttTimestampToSeconds,
+  waitForShotsForAsset: () => waitForShotsForAsset
 });
 
 // src/lib/providers.ts
@@ -1082,6 +1085,142 @@ async function fetchHotspots(identifierType, id, options) {
   return transformHotspotResponse(response);
 }
 
+// src/primitives/shots.ts
+var DEFAULT_POLL_INTERVAL_MS = 2e3;
+var MIN_POLL_INTERVAL_MS = 1e3;
+var DEFAULT_MAX_ATTEMPTS = 60;
+var SHOTS_ALREADY_REQUESTED_MESSAGE = "shots generation has already been requested";
+function getShotsPath(assetId) {
+  return `/video/v1/assets/${assetId}/shots`;
+}
+function mapManifestShots(shots) {
+  return shots.map((shot, index) => {
+    const { startTime, imageUrl } = shot;
+    if (typeof startTime !== "number" || !Number.isFinite(startTime)) {
+      throw new TypeError(`Invalid shot startTime in shots manifest at index ${index}`);
+    }
+    if (typeof imageUrl !== "string" || imageUrl.length === 0) {
+      throw new TypeError(`Invalid shot imageUrl in shots manifest at index ${index}`);
+    }
+    return {
+      startTime,
+      imageUrl
+    };
+  });
+}
+async function fetchShotsFromManifest(shotsManifestUrl) {
+  const response = await fetch(shotsManifestUrl);
+  if (!response.ok) {
+    throw new Error(
+      `Failed to fetch shots manifest: ${response.status} ${response.statusText}`
+    );
+  }
+  const manifest = await response.json();
+  if (!Array.isArray(manifest.shots)) {
+    throw new TypeError("Invalid shots manifest response: missing shots array");
+  }
+  return mapManifestShots(manifest.shots);
+}
+async function transformShotsResponse(response) {
+  switch (response.data.status) {
+    case "pending":
+      return {
+        status: "pending",
+        createdAt: response.data.created_at
+      };
+    case "errored":
+      return {
+        status: "errored",
+        createdAt: response.data.created_at,
+        error: response.data.error
+      };
+    case "completed":
+      return {
+        status: "completed",
+        createdAt: response.data.created_at,
+        shots: await fetchShotsFromManifest(response.data.shots_manifest_url)
+      };
+    default: {
+      const exhaustiveCheck = response.data;
+      throw new Error(`Unsupported shots response: ${JSON.stringify(exhaustiveCheck)}`);
+    }
+  }
+}
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function isShotsAlreadyRequestedError(error) {
+  const statusCode = error?.status ?? error?.statusCode;
+  const messages = error?.error?.messages;
+  const lowerCaseMessages = messages?.map((message) => message.toLowerCase()) ?? [];
+  const errorMessage = error instanceof Error ? error.message.toLowerCase() : "";
+  return statusCode === 400 && (lowerCaseMessages.some((message) => message.includes(SHOTS_ALREADY_REQUESTED_MESSAGE)) || errorMessage.includes(SHOTS_ALREADY_REQUESTED_MESSAGE));
+}
+async function requestShotsForAsset(assetId, options = {}) {
+  "use step";
+  const { credentials } = options;
+  const muxClient = await getMuxClientFromEnv(credentials);
+  const mux = await muxClient.createClient();
+  const response = await mux.post(
+    getShotsPath(assetId),
+    { body: {} }
+  );
+  const result = await transformShotsResponse(response);
+  if (result.status !== "pending") {
+    throw new Error(
+      `Expected pending status after requesting shots for asset '${assetId}', received '${result.status}'`
+    );
+  }
+  return result;
+}
+async function getShotsForAsset(assetId, options = {}) {
+  "use step";
+  const { credentials } = options;
+  const muxClient = await getMuxClientFromEnv(credentials);
+  const mux = await muxClient.createClient();
+  const response = await mux.get(
+    getShotsPath(assetId)
+  );
+  return await transformShotsResponse(response);
+}
+async function waitForShotsForAsset(assetId, options = {}) {
+  "use step";
+  const {
+    pollIntervalMs = DEFAULT_POLL_INTERVAL_MS,
+    maxAttempts = DEFAULT_MAX_ATTEMPTS,
+    createIfMissing = true,
+    credentials
+  } = options;
+  if (createIfMissing) {
+    try {
+      await requestShotsForAsset(assetId, { credentials });
+    } catch (error) {
+      if (!isShotsAlreadyRequestedError(error)) {
+        throw error;
+      }
+    }
+  }
+  const normalizedMaxAttempts = Math.max(1, maxAttempts);
+  const normalizedPollIntervalMs = Math.max(MIN_POLL_INTERVAL_MS, pollIntervalMs);
+  let lastStatus;
+  for (let attempt = 0; attempt < normalizedMaxAttempts; attempt++) {
+    const result = await getShotsForAsset(assetId, { credentials });
+    lastStatus = result.status;
+    if (result.status === "completed") {
+      return result;
+    }
+    if (result.status === "errored") {
+      throw new Error(`Shots generation errored for asset '${assetId}'`);
+    }
+    if (attempt < normalizedMaxAttempts - 1) {
+      await sleep(normalizedPollIntervalMs);
+    }
+  }
+  throw new Error(
+    `Timed out waiting for shots for asset '${assetId}' after ${normalizedMaxAttempts} attempts. Last status: ${lastStatus ?? "unknown"}`
+  );
+}
+
 // src/lib/mux-image-url.ts
 var DEFAULT_MUX_IMAGE_ORIGIN = "https://image.mux.com";
 function normalizeMuxImageOrigin(value) {
@@ -1785,23 +1924,33 @@ async function fetchTranscriptForAsset(asset, playbackId, options = {}) {
 // src/workflows/index.ts
 var workflows_exports = {};
 __export(workflows_exports, {
+  DEFAULT_DESCRIPTION_LENGTH: () => DEFAULT_DESCRIPTION_LENGTH,
+  DEFAULT_SUMMARY_KEYWORD_LIMIT: () => DEFAULT_SUMMARY_KEYWORD_LIMIT,
+  DEFAULT_TITLE_LENGTH: () => DEFAULT_TITLE_LENGTH,
   HIVE_SEXUAL_CATEGORIES: () => HIVE_SEXUAL_CATEGORIES,
   HIVE_VIOLENCE_CATEGORIES: () => HIVE_VIOLENCE_CATEGORIES,
-  SUMMARY_KEYWORD_LIMIT: () => SUMMARY_KEYWORD_LIMIT,
   aggregateTokenUsage: () => aggregateTokenUsage,
+  applyOverrideLists: () => applyOverrideLists,
+  applyReplacements: () => applyReplacements,
   askQuestions: () => askQuestions,
+  buildReplacementRegex: () => buildReplacementRegex,
   burnedInCaptionsSchema: () => burnedInCaptionsSchema,
+  censorVttContent: () => censorVttContent,
   chapterSchema: () => chapterSchema,
   chaptersSchema: () => chaptersSchema,
+  createReplacer: () => createReplacer,
+  editCaptions: () => editCaptions,
   generateChapters: () => generateChapters,
   generateEmbeddings: () => generateEmbeddings,
   generateVideoEmbeddings: () => generateVideoEmbeddings,
   getModerationScores: () => getModerationScores,
   getSummaryAndTags: () => getSummaryAndTags,
   hasBurnedInCaptions: () => hasBurnedInCaptions,
+  profanityDetectionSchema: () => profanityDetectionSchema,
   questionAnswerSchema: () => questionAnswerSchema,
   shouldSplitChunkTranslationError: () => shouldSplitChunkTranslationError,
   summarySchema: () => summarySchema,
+  transformCueText: () => transformCueText,
   translateAudio: () => translateAudio,
   translateCaptions: () => translateCaptions,
   translationSchema: () => translationSchema
@@ -2964,6 +3113,429 @@ async function generateChapters(assetId, languageCode, options = {}) {
   };
 }
 
+// src/workflows/edit-captions.ts
+import { generateText as generateText4, Output as Output4 } from "ai";
+import dedent4 from "dedent";
+import { z as z5 } from "zod";
+
+// src/lib/mux-tracks.ts
+async function fetchVttFromMux(vttUrl) {
+  "use step";
+  const vttResponse = await fetch(vttUrl);
+  if (!vttResponse.ok) {
+    throw new Error(`Failed to fetch VTT file: ${vttResponse.statusText}`);
+  }
+  return vttResponse.text();
+}
+async function createTextTrackOnMux(assetId, languageCode, trackName, presignedUrl, credentials) {
+  "use step";
+  const muxClient = await resolveMuxClient(credentials);
+  const mux = await muxClient.createClient();
+  const trackResponse = await mux.video.assets.createTrack(assetId, {
+    type: "text",
+    text_type: "subtitles",
+    language_code: languageCode,
+    name: trackName,
+    url: presignedUrl
+  });
+  if (!trackResponse.id) {
+    throw new Error("Failed to create text track: no track ID returned from Mux");
+  }
+  return trackResponse.id;
+}
+
+// src/lib/storage-adapter.ts
+function requireCredentials(accessKeyId, secretAccessKey) {
+  if (!accessKeyId || !secretAccessKey) {
+    throw new Error(
+      "S3 credentials are required for default storage operations. Provide S3_ACCESS_KEY_ID and S3_SECRET_ACCESS_KEY or pass options.storageAdapter."
+    );
+  }
+  return { accessKeyId, secretAccessKey };
+}
+async function putObjectWithStorageAdapter(input, adapter) {
+  if (adapter) {
+    await adapter.putObject(input);
+    return;
+  }
+  const credentials = requireCredentials(input.accessKeyId, input.secretAccessKey);
+  await putObjectToS3({
+    accessKeyId: credentials.accessKeyId,
+    secretAccessKey: credentials.secretAccessKey,
+    endpoint: input.endpoint,
+    region: input.region,
+    bucket: input.bucket,
+    key: input.key,
+    body: input.body,
+    contentType: input.contentType
+  });
+}
+async function createPresignedGetUrlWithStorageAdapter(input, adapter) {
+  if (adapter) {
+    return adapter.createPresignedGetUrl(input);
+  }
+  const credentials = requireCredentials(input.accessKeyId, input.secretAccessKey);
+  return createPresignedGetUrl({
+    accessKeyId: credentials.accessKeyId,
+    secretAccessKey: credentials.secretAccessKey,
+    endpoint: input.endpoint,
+    region: input.region,
+    bucket: input.bucket,
+    key: input.key,
+    expiresInSeconds: input.expiresInSeconds
+  });
+}
+
+// src/workflows/edit-captions.ts
+var profanityDetectionSchema = z5.object({
+  profanity: z5.array(z5.string()).describe(
+    "Unique profane words or short phrases exactly as they appear in the transcript text. Include each distinct form only once (e.g., if 'fuck' and 'fucking' both appear, list both)."
+  )
+});
+var SYSTEM_PROMPT3 = dedent4`
+  You are a content moderation assistant. Your task is to identify profane, vulgar, or obscene
+  words and phrases in subtitle text. Return ONLY the exact profane words or phrases as they appear
+  in the text. Do not modify, censor, or paraphrase them. Do not include words that are merely
+  informal or slang but not profane. Focus on words that would be bleeped on broadcast television.`;
+function transformCueText(rawVtt, transform) {
+  const lines = rawVtt.split("\n");
+  let inCueText = false;
+  let currentCueStartTime = 0;
+  const transformed = lines.map((line) => {
+    if (line.includes("-->")) {
+      const startTimestamp = line.split("-->")[0].trim();
+      currentCueStartTime = vttTimestampToSeconds(startTimestamp);
+      inCueText = true;
+      return line;
+    }
+    if (line.trim() === "") {
+      inCueText = false;
+      return line;
+    }
+    if (inCueText) {
+      return transform(line, currentCueStartTime);
+    }
+    return line;
+  });
+  return transformed.join("\n");
+}
+function buildReplacementRegex(words) {
+  const filtered = words.filter((w) => w.length > 0);
+  if (filtered.length === 0)
+    return null;
+  filtered.sort((a, b) => b.length - a.length);
+  const escaped = filtered.map((w) => w.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"));
+  const pattern = escaped.join("|");
+  return new RegExp(`\\b(?:${pattern})\\b`, "gi");
+}
+function createReplacer(mode) {
+  switch (mode) {
+    case "blank":
+      return (match) => `[${"_".repeat(match.length)}]`;
+    case "remove":
+      return () => "";
+    case "mask":
+      return (match) => "?".repeat(match.length);
+  }
+}
+function censorVttContent(rawVtt, profanity, mode) {
+  if (profanity.length === 0) {
+    return { censoredVtt: rawVtt, replacements: [] };
+  }
+  const regex = buildReplacementRegex(profanity);
+  if (!regex) {
+    return { censoredVtt: rawVtt, replacements: [] };
+  }
+  const replacer = createReplacer(mode);
+  const replacements = [];
+  const censoredVtt = transformCueText(rawVtt, (line, cueStartTime) => {
+    return line.replace(regex, (match) => {
+      const after = replacer(match);
+      replacements.push({ cueStartTime, before: match, after });
+      return after;
+    });
+  });
+  return { censoredVtt, replacements };
+}
+function applyOverrideLists(detected, alwaysCensor, neverCensor) {
+  const seen = new Set(detected.map((w) => w.toLowerCase()));
+  const merged = [...detected];
+  for (const word of alwaysCensor) {
+    const lower = word.toLowerCase();
+    if (!seen.has(lower)) {
+      seen.add(lower);
+      merged.push(word);
+    }
+  }
+  const neverSet = new Set(neverCensor.map((w) => w.toLowerCase()));
+  return merged.filter((w) => !neverSet.has(w.toLowerCase()));
+}
+function applyReplacements(rawVtt, replacements) {
+  const filtered = replacements.filter((r) => r.find.length > 0);
+  if (filtered.length === 0) {
+    return { editedVtt: rawVtt, replacements: [] };
+  }
+  const records = [];
+  const editedVtt = transformCueText(rawVtt, (line, cueStartTime) => {
+    let result = line;
+    for (const { find, replace } of filtered) {
+      const escaped = find.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+      const regex = new RegExp(`\\b${escaped}\\b`, "g");
+      result = result.replace(regex, (match) => {
+        records.push({ cueStartTime, before: match, after: replace });
+        return replace;
+      });
+    }
+    return result;
+  });
+  return { editedVtt, replacements: records };
+}
+async function identifyProfanityWithAI({
+  plainText,
+  provider,
+  modelId,
+  credentials
+}) {
+  "use step";
+  const model = await createLanguageModelFromConfig(provider, modelId, credentials);
+  const response = await generateText4({
+    model,
+    output: Output4.object({ schema: profanityDetectionSchema }),
+    messages: [
+      {
+        role: "system",
+        content: SYSTEM_PROMPT3
+      },
+      {
+        role: "user",
+        content: `Identify all profane words and phrases in the following subtitle transcript. Return each unique profane word or phrase exactly as it appears in the text.
+
+<transcript>
+${plainText}
+</transcript>`
+      }
+    ]
+  });
+  return {
+    profanity: response.output.profanity,
+    usage: {
+      inputTokens: response.usage.inputTokens,
+      outputTokens: response.usage.outputTokens,
+      totalTokens: response.usage.totalTokens,
+      reasoningTokens: response.usage.reasoningTokens,
+      cachedInputTokens: response.usage.cachedInputTokens
+    }
+  };
+}
+async function uploadEditedVttToS3({
+  editedVtt,
+  assetId,
+  trackId,
+  s3Endpoint,
+  s3Region,
+  s3Bucket,
+  storageAdapter,
+  s3SignedUrlExpirySeconds
+}) {
+  "use step";
+  const s3AccessKeyId = env_default.S3_ACCESS_KEY_ID;
+  const s3SecretAccessKey = env_default.S3_SECRET_ACCESS_KEY;
+  const vttKey = `edited/${assetId}/${trackId}-edited-${Date.now()}.vtt`;
+  await putObjectWithStorageAdapter({
+    accessKeyId: s3AccessKeyId,
+    secretAccessKey: s3SecretAccessKey,
+    endpoint: s3Endpoint,
+    region: s3Region,
+    bucket: s3Bucket,
+    key: vttKey,
+    body: editedVtt,
+    contentType: "text/vtt"
+  }, storageAdapter);
+  return createPresignedGetUrlWithStorageAdapter({
+    accessKeyId: s3AccessKeyId,
+    secretAccessKey: s3SecretAccessKey,
+    endpoint: s3Endpoint,
+    region: s3Region,
+    bucket: s3Bucket,
+    key: vttKey,
+    expiresInSeconds: s3SignedUrlExpirySeconds ?? 86400
+  }, storageAdapter);
+}
+async function deleteTrackOnMux(assetId, trackId, credentials) {
+  "use step";
+  const muxClient = await resolveMuxClient(credentials);
+  const mux = await muxClient.createClient();
+  await mux.video.assets.deleteTrack(assetId, trackId);
+}
+async function editCaptions(assetId, trackId, options) {
+  "use workflow";
+  const {
+    provider,
+    model,
+    autoCensorProfanity: autoCensorOption,
+    replacements: replacementsOption,
+    deleteOriginalTrack,
+    uploadToMux: uploadToMuxOption,
+    s3Endpoint: providedS3Endpoint,
+    s3Region: providedS3Region,
+    s3Bucket: providedS3Bucket,
+    trackNameSuffix,
+    storageAdapter,
+    credentials
+  } = options;
+  const hasAutoCensor = !!autoCensorOption;
+  const hasReplacements = !!replacementsOption && replacementsOption.length > 0;
+  if (!hasAutoCensor && !hasReplacements) {
+    throw new Error("At least one of autoCensorProfanity or replacements must be provided.");
+  }
+  if (autoCensorOption && !provider) {
+    throw new Error("provider is required when using autoCensorProfanity.");
+  }
+  const deleteOriginal = deleteOriginalTrack !== false;
+  const uploadToMux = uploadToMuxOption !== false;
+  const s3Endpoint = providedS3Endpoint ?? env_default.S3_ENDPOINT;
+  const s3Region = providedS3Region ?? env_default.S3_REGION ?? "auto";
+  const s3Bucket = providedS3Bucket ?? env_default.S3_BUCKET;
+  const s3AccessKeyId = env_default.S3_ACCESS_KEY_ID;
+  const s3SecretAccessKey = env_default.S3_SECRET_ACCESS_KEY;
+  if (uploadToMux && (!s3Endpoint || !s3Bucket || !storageAdapter && (!s3AccessKeyId || !s3SecretAccessKey))) {
+    throw new Error(
+      "Storage configuration is required for uploading to Mux. Provide s3Endpoint and s3Bucket. If no storageAdapter is supplied, also provide s3AccessKeyId and s3SecretAccessKey in options or set S3_ENDPOINT, S3_BUCKET, S3_ACCESS_KEY_ID, and S3_SECRET_ACCESS_KEY environment variables."
+    );
+  }
+  const { asset: assetData, playbackId, policy } = await getPlaybackIdForAsset(assetId, credentials);
+  const assetDurationSeconds = getAssetDurationSecondsFromAsset(assetData);
+  const signingContext = await resolveMuxSigningContext(credentials);
+  if (policy === "signed" && !signingContext) {
+    throw new Error(
+      "Signed playback ID requires signing credentials. Set MUX_SIGNING_KEY and MUX_PRIVATE_KEY environment variables."
+    );
+  }
+  const readyTextTracks = getReadyTextTracks(assetData);
+  const sourceTrack = readyTextTracks.find((t) => t.id === trackId);
+  if (!sourceTrack) {
+    const availableTrackIds = readyTextTracks.map((t) => t.id).filter(Boolean).join(", ");
+    throw new Error(
+      `Track '${trackId}' not found or not ready on asset '${assetId}'. Available track IDs: ${availableTrackIds || "none"}`
+    );
+  }
+  const vttUrl = await buildTranscriptUrl(playbackId, trackId, policy === "signed", credentials);
+  let vttContent;
+  try {
+    vttContent = await fetchVttFromMux(vttUrl);
+  } catch (error) {
+    throw new Error(`Failed to fetch VTT content: ${error instanceof Error ? error.message : "Unknown error"}`);
+  }
+  let editedVtt = vttContent;
+  let totalReplacementCount = 0;
+  let autoCensorResult;
+  let usage;
+  if (autoCensorOption) {
+    const { mode = "blank", alwaysCensor = [], neverCensor = [] } = autoCensorOption;
+    const plainText = extractTextFromVTT(vttContent);
+    if (!plainText.trim()) {
+      throw new Error("Track transcript is empty; nothing to censor.");
+    }
+    const modelConfig = resolveLanguageModelConfig({
+      ...options,
+      provider,
+      model
+    });
+    let detectedProfanity;
+    try {
+      const result = await identifyProfanityWithAI({
+        plainText,
+        provider: modelConfig.provider,
+        modelId: modelConfig.modelId,
+        credentials
+      });
+      detectedProfanity = result.profanity;
+      usage = result.usage;
+    } catch (error) {
+      throw new Error(`Failed to detect profanity with ${modelConfig.provider}: ${error instanceof Error ? error.message : "Unknown error"}`);
+    }
+    const finalProfanity = applyOverrideLists(detectedProfanity, alwaysCensor, neverCensor);
+    const { censoredVtt, replacements: censorReplacements } = censorVttContent(editedVtt, finalProfanity, mode);
+    editedVtt = censoredVtt;
+    totalReplacementCount += censorReplacements.length;
+    autoCensorResult = { replacements: censorReplacements };
+  }
+  let replacementsResult;
+  if (replacementsOption && replacementsOption.length > 0) {
+    const { editedVtt: afterReplacements, replacements: staticReplacements } = applyReplacements(editedVtt, replacementsOption);
+    editedVtt = afterReplacements;
+    totalReplacementCount += staticReplacements.length;
+    replacementsResult = { replacements: staticReplacements };
+  }
+  const usageWithMetadata = usage ? {
+    ...usage,
+    metadata: {
+      assetDurationSeconds
+    }
+  } : void 0;
+  if (!uploadToMux) {
+    return {
+      assetId,
+      trackId,
+      originalVtt: vttContent,
+      editedVtt,
+      totalReplacementCount,
+      autoCensorProfanity: autoCensorResult,
+      replacements: replacementsResult,
+      usage: usageWithMetadata
+    };
+  }
+  let presignedUrl;
+  try {
+    presignedUrl = await uploadEditedVttToS3({
+      editedVtt,
+      assetId,
+      trackId,
+      s3Endpoint,
+      s3Region,
+      s3Bucket,
+      storageAdapter,
+      s3SignedUrlExpirySeconds: options.s3SignedUrlExpirySeconds
+    });
+  } catch (error) {
+    throw new Error(`Failed to upload VTT to S3: ${error instanceof Error ? error.message : "Unknown error"}`);
+  }
+  let uploadedTrackId;
+  try {
+    const languageCode = sourceTrack.language_code || "en";
+    const suffix = trackNameSuffix ?? "edited";
+    const trackName = `${sourceTrack.name || "Subtitles"} (${suffix})`;
+    uploadedTrackId = await createTextTrackOnMux(
+      assetId,
+      languageCode,
+      trackName,
+      presignedUrl,
+      credentials
+    );
+  } catch (error) {
+    console.warn(`Failed to add track to Mux asset: ${error instanceof Error ? error.message : "Unknown error"}`);
+  }
+  if (deleteOriginal && uploadedTrackId) {
+    try {
+      await deleteTrackOnMux(assetId, trackId, credentials);
+    } catch (error) {
+      console.warn(`Failed to delete original track: ${error instanceof Error ? error.message : "Unknown error"}`);
+    }
+  }
+  return {
+    assetId,
+    trackId,
+    originalVtt: vttContent,
+    editedVtt,
+    totalReplacementCount,
+    autoCensorProfanity: autoCensorResult,
+    replacements: replacementsResult,
+    uploadedTrackId,
+    presignedUrl,
+    usage: usageWithMetadata
+  };
+}
+
 // src/workflows/embeddings.ts
 import { embed } from "ai";
 function averageEmbeddings(embeddings) {
@@ -3611,16 +4183,18 @@ async function getModerationScores(assetId, options = {}) {
 }
 
 // src/workflows/summarization.ts
-import { generateText as generateText4, Output as Output4 } from "ai";
-import dedent4 from "dedent";
-import { z as z5 } from "zod";
-var SUMMARY_KEYWORD_LIMIT = 10;
-var summarySchema = z5.object({
-  keywords: z5.array(z5.string()),
-  title: z5.string(),
-  description: z5.string()
+import { generateText as generateText5, Output as Output5 } from "ai";
+import dedent5 from "dedent";
+import { z as z6 } from "zod";
+var DEFAULT_SUMMARY_KEYWORD_LIMIT = 10;
+var DEFAULT_TITLE_LENGTH = 10;
+var DEFAULT_DESCRIPTION_LENGTH = 50;
+var summarySchema = z6.object({
+  keywords: z6.array(z6.string()),
+  title: z6.string(),
+  description: z6.string()
 }).strict();
-var SUMMARY_OUTPUT = Output4.object({
+var SUMMARY_OUTPUT = Output5.object({
   name: "summary_metadata",
   description: "Structured summary with title, description, and keywords.",
   schema: summarySchema
@@ -3631,10 +4205,49 @@ var TONE_INSTRUCTIONS = {
   playful: "Channel your inner diva! Answer with maximum sass, wit, and playful attitude. Don't hold back - be cheeky, clever, and delightfully snarky. Make it pop!",
   professional: "Provide a professional, executive-level analysis suitable for business reporting."
 };
+var DESCRIPTION_LENGTH_THRESHOLD_SMALL = 25;
+var DESCRIPTION_LENGTH_THRESHOLD_LARGE = 100;
+function buildDescriptionGuidance(wordCount, contentType) {
+  if (wordCount < DESCRIPTION_LENGTH_THRESHOLD_SMALL) {
+    if (contentType === "video") {
+      return dedent5`A brief summary of the video in approximately ${wordCount} words.
+        Focus on the single most important subject or action.
+        Write in present tense.`;
+    }
+    return dedent5`A brief summary of the audio content in approximately ${wordCount} words.
+      Focus on the single most important topic or theme.
+      Write in present tense.`;
+  }
+  if (wordCount > DESCRIPTION_LENGTH_THRESHOLD_LARGE) {
+    if (contentType === "video") {
+      return dedent5`A detailed summary that describes what happens across the video.
+        Aim for approximately ${wordCount} words, and you may use multiple sentences.
+        Be thorough: cover subjects, actions, setting, progression, and any notable details visible across frames.
+        Write in present tense. Be specific about observable details rather than making assumptions.
+        If the transcript provides dialogue or narration, incorporate key points but prioritize visual content.`;
+    }
+    return dedent5`A detailed summary that describes the audio content.
+      Aim for approximately ${wordCount} words, and you may use multiple sentences.
+      Be thorough: cover topics, speakers, themes, progression, and any notable insights.
+      Write in present tense. Be specific about what is discussed or presented rather than making assumptions.
+      Focus on the spoken content and any key insights, dialogue, or narrative elements.`;
+  }
+  if (contentType === "video") {
+    return dedent5`A summary that describes what happens across the video.
+      Aim for approximately ${wordCount} words, and you may use multiple sentences.
+      Cover the main subjects, actions, setting, and any notable progression visible across frames.
+      Write in present tense. Be specific about observable details rather than making assumptions.
+      If the transcript provides dialogue or narration, incorporate key points but prioritize visual content.`;
+  }
+  return dedent5`A summary that describes the audio content.
+    Aim for approximately ${wordCount} words, and you may use multiple sentences.
+    Cover the main topics, speakers, themes, and any notable progression in the discussion or narration.
+    Write in present tense. Be specific about what is discussed or presented rather than making assumptions.
+    Focus on the spoken content and any key insights, dialogue, or narrative elements.`;
+}
 function createSummarizationBuilder({ titleLength, descriptionLength, tagCount } = {}) {
-  const titleBrevity = titleLength != null ? `Aim for approximately ${titleLength} characters.` : "Aim for brevity - typically under 10 words.";
-  const descConstraint = descriptionLength != null ? `approximately ${descriptionLength} characters` : "2-4 sentences";
-  const keywordLimit = tagCount ?? SUMMARY_KEYWORD_LIMIT;
+  const titleBrevity = `Aim for approximately ${titleLength ?? DEFAULT_TITLE_LENGTH} words.`;
+  const keywordLimit = tagCount ?? DEFAULT_SUMMARY_KEYWORD_LIMIT;
   return createPromptBuilder({
     template: {
       task: {
@@ -3643,7 +4256,7 @@ function createSummarizationBuilder({ titleLength, descriptionLength, tagCount }
       },
       title: {
         tag: "title_requirements",
-        content: dedent4`
+        content: dedent5`
           A short, compelling headline that immediately communicates the subject or action.
           ${titleBrevity} Think of how a news headline or video card title would read.
           Start with the primary subject, action, or topic - never begin with "A video of" or similar phrasing.
@@ -3651,15 +4264,11 @@ function createSummarizationBuilder({ titleLength, descriptionLength, tagCount }
       },
       description: {
         tag: "description_requirements",
-        content: dedent4`
-          A concise summary (${descConstraint}) that describes what happens across the video.
-          Cover the main subjects, actions, setting, and any notable progression visible across frames.
-          Write in present tense. Be specific about observable details rather than making assumptions.
-          If the transcript provides dialogue or narration, incorporate key points but prioritize visual content.`
+        content: buildDescriptionGuidance(descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH, "video")
       },
       keywords: {
         tag: "keywords_requirements",
-        content: dedent4`
+        content: dedent5`
           Specific, searchable terms (up to ${keywordLimit}) that capture:
           - Primary subjects (people, animals, objects)
           - Actions and activities being performed
@@ -3671,7 +4280,7 @@ function createSummarizationBuilder({ titleLength, descriptionLength, tagCount }
       },
       qualityGuidelines: {
         tag: "quality_guidelines",
-        content: dedent4`
+        content: dedent5`
           - Examine all frames to understand the full context and progression
           - Be precise: "golden retriever" is better than "dog" when identifiable
           - Capture the narrative: what begins, develops, and concludes
@@ -3682,9 +4291,8 @@ function createSummarizationBuilder({ titleLength, descriptionLength, tagCount }
   });
 }
 function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {}) {
-  const titleBrevity = titleLength != null ? `Aim for approximately ${titleLength} characters.` : "Aim for brevity - typically under 10 words.";
-  const descConstraint = descriptionLength != null ? `approximately ${descriptionLength} characters` : "2-4 sentences";
-  const keywordLimit = tagCount ?? SUMMARY_KEYWORD_LIMIT;
+  const titleBrevity = `Aim for approximately ${titleLength ?? DEFAULT_TITLE_LENGTH} words.`;
+  const keywordLimit = tagCount ?? DEFAULT_SUMMARY_KEYWORD_LIMIT;
   return createPromptBuilder({
     template: {
       task: {
@@ -3693,7 +4301,7 @@ function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {
       },
       title: {
         tag: "title_requirements",
-        content: dedent4`
+        content: dedent5`
           A short, compelling headline that immediately communicates the subject or topic.
           ${titleBrevity} Think of how a podcast title or audio description would read.
           Start with the primary subject, action, or topic - never begin with "An audio of" or similar phrasing.
@@ -3701,15 +4309,11 @@ function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {
       },
       description: {
         tag: "description_requirements",
-        content: dedent4`
-          A concise summary (${descConstraint}) that describes the audio content.
-          Cover the main topics, speakers, themes, and any notable progression in the discussion or narration.
-          Write in present tense. Be specific about what is discussed or presented rather than making assumptions.
-          Focus on the spoken content and any key insights, dialogue, or narrative elements.`
+        content: buildDescriptionGuidance(descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH, "audio")
       },
       keywords: {
         tag: "keywords_requirements",
-        content: dedent4`
+        content: dedent5`
           Specific, searchable terms (up to ${keywordLimit}) that capture:
           - Primary topics and themes
           - Speakers or presenters (if named)
@@ -3721,7 +4325,7 @@ function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {
       },
       qualityGuidelines: {
         tag: "quality_guidelines",
-        content: dedent4`
+        content: dedent5`
           - Analyze the full transcript to understand context and themes
           - Be precise: use specific terminology when mentioned
           - Capture the narrative: what is introduced, discussed, and concluded
@@ -3731,7 +4335,7 @@ function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {
     sectionOrder: ["task", "title", "description", "keywords", "qualityGuidelines"]
   });
 }
-var SYSTEM_PROMPT3 = dedent4`
+var SYSTEM_PROMPT4 = dedent5`
   <role>
     You are a video content analyst specializing in storyboard interpretation and multimodal analysis.
   </role>
@@ -3788,7 +4392,7 @@ var SYSTEM_PROMPT3 = dedent4`
 
     Write as if describing reality, not describing a recording of reality.
   </language_guidelines>`;
-var AUDIO_ONLY_SYSTEM_PROMPT = dedent4`
+var AUDIO_ONLY_SYSTEM_PROMPT = dedent5`
   <role>
     You are an audio content analyst specializing in transcript analysis and metadata generation.
   </role>
@@ -3856,6 +4460,11 @@ function buildUserPrompt4({
   const contextSections = [createToneSection(TONE_INSTRUCTIONS[tone])];
   if (languageName) {
     contextSections.push(createLanguageSection(languageName));
+  } else {
+    contextSections.push({
+      tag: "language",
+      content: "Respond in English. Never switch languages to satisfy length constraints."
+    });
   }
   if (transcriptText) {
     const format = isCleanTranscript ? "plain text" : "WebVTT";
@@ -3868,7 +4477,7 @@ function buildUserPrompt4({
 async function analyzeStoryboard2(imageDataUrl, provider, modelId, userPrompt, systemPrompt, credentials) {
   "use step";
   const model = await createLanguageModelFromConfig(provider, modelId, credentials);
-  const response = await generateText4({
+  const response = await generateText5({
     model,
     output: SUMMARY_OUTPUT,
     messages: [
@@ -3903,7 +4512,7 @@ async function analyzeStoryboard2(imageDataUrl, provider, modelId, userPrompt, s
 async function analyzeAudioOnly(provider, modelId, userPrompt, systemPrompt, credentials) {
   "use step";
   const model = await createLanguageModelFromConfig(provider, modelId, credentials);
-  const response = await generateText4({
+  const response = await generateText5({
     model,
     output: SUMMARY_OUTPUT,
     messages: [
@@ -3932,7 +4541,7 @@ async function analyzeAudioOnly(provider, modelId, userPrompt, systemPrompt, cre
     }
   };
 }
-function normalizeKeywords(keywords, limit = SUMMARY_KEYWORD_LIMIT) {
+function normalizeKeywords(keywords, limit = DEFAULT_SUMMARY_KEYWORD_LIMIT) {
   if (!Array.isArray(keywords) || keywords.length === 0) {
     return [];
   }
@@ -4019,7 +4628,7 @@ async function getSummaryAndTags(assetId, options) {
   });
   let analysisResponse;
   let imageUrl;
-  const systemPrompt = isAudioOnly ? AUDIO_ONLY_SYSTEM_PROMPT : SYSTEM_PROMPT3;
+  const systemPrompt = isAudioOnly ? AUDIO_ONLY_SYSTEM_PROMPT : SYSTEM_PROMPT4;
   try {
     if (isAudioOnly) {
       analysisResponse = await analyzeAudioOnly(
@@ -4072,7 +4681,7 @@ async function getSummaryAndTags(assetId, options) {
     assetId,
     title: analysisResponse.result.title,
     description: analysisResponse.result.description,
-    tags: normalizeKeywords(analysisResponse.result.keywords, tagCount ?? SUMMARY_KEYWORD_LIMIT),
+    tags: normalizeKeywords(analysisResponse.result.keywords, tagCount ?? DEFAULT_SUMMARY_KEYWORD_LIMIT),
     storyboardUrl: imageUrl,
     // undefined for audio-only assets
     usage: {
@@ -4085,52 +4694,10 @@ async function getSummaryAndTags(assetId, options) {
   };
 }
 
-// src/lib/storage-adapter.ts
-function requireCredentials(accessKeyId, secretAccessKey) {
-  if (!accessKeyId || !secretAccessKey) {
-    throw new Error(
-      "S3 credentials are required for default storage operations. Provide S3_ACCESS_KEY_ID and S3_SECRET_ACCESS_KEY or pass options.storageAdapter."
-    );
-  }
-  return { accessKeyId, secretAccessKey };
-}
-async function putObjectWithStorageAdapter(input, adapter) {
-  if (adapter) {
-    await adapter.putObject(input);
-    return;
-  }
-  const credentials = requireCredentials(input.accessKeyId, input.secretAccessKey);
-  await putObjectToS3({
-    accessKeyId: credentials.accessKeyId,
-    secretAccessKey: credentials.secretAccessKey,
-    endpoint: input.endpoint,
-    region: input.region,
-    bucket: input.bucket,
-    key: input.key,
-    body: input.body,
-    contentType: input.contentType
-  });
-}
-async function createPresignedGetUrlWithStorageAdapter(input, adapter) {
-  if (adapter) {
-    return adapter.createPresignedGetUrl(input);
-  }
-  const credentials = requireCredentials(input.accessKeyId, input.secretAccessKey);
-  return createPresignedGetUrl({
-    accessKeyId: credentials.accessKeyId,
-    secretAccessKey: credentials.secretAccessKey,
-    endpoint: input.endpoint,
-    region: input.region,
-    bucket: input.bucket,
-    key: input.key,
-    expiresInSeconds: input.expiresInSeconds
-  });
-}
-
 // src/workflows/translate-audio.ts
 var STATIC_RENDITION_POLL_INTERVAL_MS = 5e3;
 var STATIC_RENDITION_MAX_ATTEMPTS = 36;
-async function sleep(ms) {
+async function sleep2(ms) {
   "use step";
   await new Promise((resolve) => setTimeout(resolve, ms));
 }
@@ -4196,7 +4763,7 @@ async function waitForAudioStaticRendition({
     console.warn(`\u2139\uFE0F Static rendition already ${status}. Waiting for it to finish...`);
   }
   for (let attempt = 1; attempt <= STATIC_RENDITION_MAX_ATTEMPTS; attempt++) {
-    await sleep(STATIC_RENDITION_POLL_INTERVAL_MS);
+    await sleep2(STATIC_RENDITION_POLL_INTERVAL_MS);
     currentAsset = await mux.video.assets.retrieve(assetId);
     if (hasReadyAudioStaticRendition(currentAsset)) {
       return currentAsset;
@@ -4303,7 +4870,8 @@ async function uploadDubbedAudioToS3({
   s3Endpoint,
   s3Region,
   s3Bucket,
-  storageAdapter
+  storageAdapter,
+  s3SignedUrlExpirySeconds
 }) {
   "use step";
   const s3AccessKeyId = env_default.S3_ACCESS_KEY_ID;
@@ -4326,10 +4894,11 @@ async function uploadDubbedAudioToS3({
     region: s3Region,
     bucket: s3Bucket,
     key: audioKey,
-    expiresInSeconds: 3600
+    expiresInSeconds: s3SignedUrlExpirySeconds ?? 86400
   }, storageAdapter);
+  const expiryHours = Math.round((s3SignedUrlExpirySeconds ?? 86400) / 3600);
   console.warn(`\u2705 Audio uploaded successfully to: ${audioKey}`);
-  console.warn(`\u{1F517} Generated presigned URL (expires in 1 hour)`);
+  console.warn(`\u{1F517} Generated presigned URL (expires in ${expiryHours} hour${expiryHours === 1 ? "" : "s"})`);
   return presignedUrl;
 }
 async function createAudioTrackOnMux(assetId, languageCode, presignedUrl, credentials) {
@@ -4428,7 +4997,7 @@ async function translateAudio(assetId, toLanguageCode, options = {}) {
   const maxPollAttempts = 180;
   let targetLanguages = [];
   while (dubbingStatus === "dubbing" && pollAttempts < maxPollAttempts) {
-    await sleep(1e4);
+    await sleep2(1e4);
     pollAttempts++;
     try {
       const statusResult = await checkElevenLabsDubbingStatus({
@@ -4493,7 +5062,8 @@ async function translateAudio(assetId, toLanguageCode, options = {}) {
       s3Endpoint,
       s3Region,
       s3Bucket,
-      storageAdapter: effectiveStorageAdapter
+      storageAdapter: effectiveStorageAdapter,
+      s3SignedUrlExpirySeconds: options.s3SignedUrlExpirySeconds
     });
   } catch (error) {
     throw new Error(`Failed to upload audio to S3: ${error instanceof Error ? error.message : "Unknown error"}`);
@@ -4531,24 +5101,24 @@ async function translateAudio(assetId, toLanguageCode, options = {}) {
 // src/workflows/translate-captions.ts
 import {
   APICallError,
-  generateText as generateText5,
+  generateText as generateText6,
   NoObjectGeneratedError,
-  Output as Output5,
+  Output as Output6,
   RetryError,
   TypeValidationError
 } from "ai";
-import dedent5 from "dedent";
-import { z as z6 } from "zod";
-var translationSchema = z6.object({
-  translation: z6.string()
+import dedent6 from "dedent";
+import { z as z7 } from "zod";
+var translationSchema = z7.object({
+  translation: z7.string()
 });
-var SYSTEM_PROMPT4 = dedent5`
+var SYSTEM_PROMPT5 = dedent6`
   You are a subtitle translation expert. Translate VTT subtitle files to the target language specified by the user.
   You may receive either a full VTT file or a chunk from a larger VTT.
   Preserve all timestamps, cue ordering, and VTT formatting exactly as they appear.
   Return JSON with a single key "translation" containing the translated VTT content.
 `;
-var CUE_TRANSLATION_SYSTEM_PROMPT = dedent5`
+var CUE_TRANSLATION_SYSTEM_PROMPT = dedent6`
   You are a subtitle translation expert.
   You will receive a sequence of subtitle cues extracted from a VTT file.
   Translate the cues to the requested target language while preserving their original order.
@@ -4710,14 +5280,6 @@ function buildTranslationChunkRequests(vttContent, assetDurationSeconds, chunkin
     )
   };
 }
-async function fetchVttFromMux(vttUrl) {
-  "use step";
-  const vttResponse = await fetch(vttUrl);
-  if (!vttResponse.ok) {
-    throw new Error(`Failed to fetch VTT file: ${vttResponse.statusText}`);
-  }
-  return vttResponse.text();
-}
 async function translateVttWithAI({
   vttContent,
   fromLanguageCode,
@@ -4728,13 +5290,13 @@ async function translateVttWithAI({
 }) {
   "use step";
   const model = await createLanguageModelFromConfig(provider, modelId, credentials);
-  const response = await generateText5({
+  const response = await generateText6({
     model,
-    output: Output5.object({ schema: translationSchema }),
+    output: Output6.object({ schema: translationSchema }),
     messages: [
       {
         role: "system",
-        content: SYSTEM_PROMPT4
+        content: SYSTEM_PROMPT5
       },
       {
         role: "user",
@@ -4765,8 +5327,8 @@ async function translateCueChunkWithAI({
 }) {
   "use step";
   const model = await createLanguageModelFromConfig(provider, modelId, credentials);
-  const schema = z6.object({
-    translations: z6.array(z6.string().min(1)).length(cues.length)
+  const schema = z7.object({
+    translations: z7.array(z7.string().min(1)).length(cues.length)
   });
   const cuePayload = cues.map((cue, index) => ({
     index,
@@ -4774,9 +5336,9 @@ async function translateCueChunkWithAI({
     endTime: cue.endTime,
     text: cue.text
   }));
-  const response = await generateText5({
+  const response = await generateText6({
     model,
-    output: Output5.object({ schema }),
+    output: Output6.object({ schema }),
     messages: [
       {
         role: "system",
@@ -4933,7 +5495,8 @@ async function uploadVttToS3({
   s3Endpoint,
   s3Region,
   s3Bucket,
-  storageAdapter
+  storageAdapter,
+  s3SignedUrlExpirySeconds
 }) {
   "use step";
   const s3AccessKeyId = env_default.S3_ACCESS_KEY_ID;
@@ -4956,25 +5519,9 @@ async function uploadVttToS3({
     region: s3Region,
     bucket: s3Bucket,
     key: vttKey,
-    expiresInSeconds: 3600
+    expiresInSeconds: s3SignedUrlExpirySeconds ?? 86400
   }, storageAdapter);
 }
-async function createTextTrackOnMux(assetId, languageCode, trackName, presignedUrl, credentials) {
-  "use step";
-  const muxClient = await resolveMuxClient(credentials);
-  const mux = await muxClient.createClient();
-  const trackResponse = await mux.video.assets.createTrack(assetId, {
-    type: "text",
-    text_type: "subtitles",
-    language_code: languageCode,
-    name: trackName,
-    url: presignedUrl
-  });
-  if (!trackResponse.id) {
-    throw new Error("Failed to create text track: no track ID returned from Mux");
-  }
-  return trackResponse.id;
-}
 async function translateCaptions(assetId, fromLanguageCode, toLanguageCode, options) {
   "use workflow";
   const {
@@ -5092,7 +5639,8 @@ async function translateCaptions(assetId, fromLanguageCode, toLanguageCode, opti
       s3Endpoint,
       s3Region,
       s3Bucket,
-      storageAdapter: effectiveStorageAdapter
+      storageAdapter: effectiveStorageAdapter,
+      s3SignedUrlExpirySeconds: options.s3SignedUrlExpirySeconds
     });
   } catch (error) {
     throw new Error(`Failed to upload VTT to S3: ${error instanceof Error ? error.message : "Unknown error"}`);