@mux/ai 0.43.0 → 0.44.0

package/dist/workflows/index.js

@@ -3480,8 +3480,32 @@ async function hasBurnedInCaptionsInternal(assetId, options, collectedUsage) {
 
 // src/workflows/chapters.ts
 import { generateText as generateText3, Output as Output3 } from "ai";
-import dedent4 from "dedent";
+import dedent5 from "dedent";
 import { z as z5 } from "zod";
+
+// src/prompts/chapters.ts
+import dedent4 from "dedent";
+function createChapterGuidance({
+  terminologySource = "transcript"
+} = {}) {
+  return {
+    qualityGuidelines: {
+      tag: "quality_guidelines",
+      content: dedent4`
+        - Create chapters at topic shifts or clear transitions
+        - Keep chapter titles concise and descriptive`
+    },
+    titleGuidelines: {
+      tag: "title_guidelines",
+      content: dedent4`
+        - Keep titles concise and descriptive
+        - Avoid filler or generic labels like "Chapter 1"
+        ${terminologySource === "transcript" ? "- Use the transcript's terminology" : "- Use terminology from the provided evidence"}`
+    }
+  };
+}
+
+// src/workflows/chapters.ts
 var chapterSchema = z5.object({
   startTime: z5.number(),
   title: z5.string().max(300)
@@ -3539,6 +3563,7 @@ async function generateChaptersWithAI({
     unexpectedChapterKeys
   };
 }
+var chapterGuidance = createChapterGuidance();
 var chapterSystemPromptBuilder = createPromptBuilder({
   template: {
     role: {
@@ -3547,13 +3572,13 @@ var chapterSystemPromptBuilder = createPromptBuilder({
     },
     context: {
       tag: "context",
-      content: dedent4`
+      content: dedent5`
         You receive a timestamped transcript with lines in the form "[12s] Caption text".
         Use those timestamps as anchors to determine chapter start times in seconds.`
     },
     security: {
       tag: "security",
-      content: dedent4`
+      content: dedent5`
         ${NON_DISCLOSURE_CONSTRAINT}
 
         ${UNTRUSTED_USER_INPUT_NOTICE}
@@ -3562,34 +3587,32 @@ var chapterSystemPromptBuilder = createPromptBuilder({
     },
     constraints: {
       tag: "constraints",
-      content: dedent4`
+      content: dedent5`
         - Only use information present in the transcript
         - Return structured data that matches the requested JSON schema
         - Do not add commentary or extra text outside the JSON
         - When a <language> section is provided, all chapter titles MUST be written in that language`
     },
     qualityGuidelines: {
-      tag: "quality_guidelines",
-      content: dedent4`
-        - Create chapters at topic shifts or clear transitions
-        - Keep chapter titles concise and descriptive
-        - Ensure the first chapter starts at 0 seconds`
+      ...chapterGuidance.qualityGuidelines,
+      content: `${chapterGuidance.qualityGuidelines.content}
+- Ensure the first chapter starts at 0 seconds`
     }
   },
   sectionOrder: ["role", "context", "security", "constraints", "qualityGuidelines"]
 });
 var AUDIO_ONLY_SYSTEM_PROMPT_OVERRIDES = {
   role: "You are an audio editor and transcript analyst specializing in segmenting content into logical chapters.",
-  context: dedent4`
+  context: dedent5`
     You receive a timestamped transcript from audio-only content with lines in the form "[12s] Transcript text".
     Use those timestamps as anchors to determine chapter start times in seconds.`
 };
 var SCENE_CONTEXT_SYSTEM_PROMPT_OVERRIDES = {
-  context: dedent4`
+  context: dedent5`
     You receive a timestamped transcript with lines in the form "[12s] Caption text" and an ordered scene context.
     Transcript timestamps are in seconds. Scene start and end timestamps are in milliseconds.
     Use the scene progression to choose logical chapter groupings and scene starts as chapter boundaries.`,
-  constraints: dedent4`
+  constraints: dedent5`
     - Only use information present in the transcript and scene context
     - Return structured data that matches the requested JSON schema
     - Do not add commentary or extra text outside the JSON
@@ -3603,7 +3626,7 @@ var chaptersPromptBuilder = createPromptBuilder({
     },
     outputFormat: {
       tag: "output_format",
-      content: dedent4`
+      content: dedent5`
         Return valid JSON in this exact shape:
         {
           "chapters": [
@@ -3620,13 +3643,7 @@ var chaptersPromptBuilder = createPromptBuilder({
       content: ""
       // Placeholder - always overridden with dynamic content
     },
-    titleGuidelines: {
-      tag: "title_guidelines",
-      content: dedent4`
-        - Keep titles concise and descriptive
-        - Avoid filler or generic labels like "Chapter 1"
-        - Use the transcript's terminology`
-    }
+    titleGuidelines: chapterGuidance.titleGuidelines
   },
   sectionOrder: ["task", "outputFormat", "chapterGuidelines", "titleGuidelines"]
 });
@@ -3665,7 +3682,7 @@ function buildUserPrompt3({
   } else if (scope) {
     firstChapterGuideline = `- The first chapter must start at ${scope.startTime}s`;
   }
-  const dynamicChapterGuidelines = dedent4`
+  const dynamicChapterGuidelines = dedent5`
     - Create at least ${minChaptersPerHour} and at most ${maxChaptersPerHour} chapters per hour of content
     - Use start times in seconds (not HH:MM:SS)
     ${scope ? `- Analyze only the range from ${scope.startTime}s (inclusive) to ${scope.endTime}s (exclusive)` : ""}
@@ -5323,7 +5340,7 @@ async function generateVideoEmbeddings(assetId, options = {}) {
 
 // src/workflows/engagement-insights.ts
 import { generateText as generateText5, Output as Output5 } from "ai";
-import dedent5 from "dedent";
+import dedent6 from "dedent";
 import { z as z7 } from "zod";
 
 // src/primitives/heatmap.ts
@@ -5592,7 +5609,7 @@ var AUDIO_ONLY_SYSTEM_PROMPT2 = promptDedent`
     - Connect engagement data to content features
   </quality_guidelines>
 `;
-var INSIGHT_GUIDELINES = dedent5`
+var INSIGHT_GUIDELINES = dedent6`
   Provide explanatory insights that describe WHY moments are engaging.
   Focus on observable patterns and viewer behavior correlation.
   Example: "The cooking demonstration at 2:15 has 3x average engagement
@@ -5610,7 +5627,7 @@ var engagementInsightsPromptBuilder = createPromptBuilder({
     },
     outputFormat: {
       tag: "output_format",
-      content: dedent5`
+      content: dedent6`
         Return valid JSON matching the provided schema.
         Include insights for each hotspot and overall engagement trends.
         Each momentInsight must include the hotspotIndex matching the input.
@@ -5618,7 +5635,7 @@ var engagementInsightsPromptBuilder = createPromptBuilder({
     },
     visualContext: {
       tag: "visual_context",
-      content: dedent5`
+      content: dedent6`
         Images are provided for each hotspot timestamp.
         A storyboard overview image may also be included showing the full video timeline.
         Use visual evidence from these images to support your engagement analysis.
@@ -5632,14 +5649,14 @@ function buildUserPrompt4(hotspots, transcriptSegments, heatmap, isAudioOnly) {
     const transcript = transcriptSegments[idx]?.text || "(no transcript)";
     const startTimestamp = secondsToTimestamp(h.startMs / 1e3);
     const endTimestamp = secondsToTimestamp(h.endMs / 1e3);
-    return dedent5`
+    return dedent6`
         Hotspot ${idx} (hotspotIndex: ${idx}):
         - Time Range: ${startTimestamp} - ${endTimestamp}
         - Engagement Score: ${h.score.toFixed(2)} (0=low, 1=high)
         - Transcript: ${JSON.stringify(transcript)}
       `;
   }).join("\n\n");
-  const heatmapData = dedent5`
+  const heatmapData = dedent6`
     Engagement Heatmap (${heatmap.length} values, each representing 1/${heatmap.length}th of the video, higher = more re-watches):
     [${heatmap.map((v) => v.toFixed(2)).join(", ")}]
   `;
@@ -7458,7 +7475,6 @@ async function getModerationScores(assetId, options = {}) {
 
 // src/workflows/summarization.ts
 import { generateText as generateText7, Output as Output7 } from "ai";
-import dedent6 from "dedent";
 import { z as z9 } from "zod";
 
 // src/lib/markdown-normalization.ts
@@ -7468,124 +7484,128 @@ function normalizeMarkdownDescription(description) {
   return description.replace(ESCAPED_LINE_BREAK_REGEX, "\n").replace(/([^\n]) {2,}([*-]) (?=\S)/g, "$1\n$2 ").replace(/([.!?:;]) ([*-]) (?=\S)/g, "$1\n$2 ").replace(MULTIPLE_BLANK_LINES_REGEX, "\n\n");
 }
 
-// src/workflows/summarization.ts
+// src/prompts/summarization.ts
+import dedent7 from "dedent";
 var DEFAULT_SUMMARY_KEYWORD_LIMIT = 10;
 var DEFAULT_TITLE_LENGTH = 10;
 var DEFAULT_DESCRIPTION_LENGTH = 50;
-var summarySchema = z9.object({
-  keywords: z9.array(z9.string().max(200)),
-  title: z9.string().max(500),
-  description: z9.string().max(1e4)
-});
-function buildSummarySchema(descriptionLength) {
-  const descriptionMaxChars = Math.max(2e3, descriptionLength * 15);
-  return z9.object({
-    keywords: z9.array(z9.string().max(200)),
-    title: z9.string().max(500),
-    description: z9.string().max(descriptionMaxChars)
-  });
-}
-var VALID_TONES = ["neutral", "playful", "professional"];
-var TONE_INSTRUCTIONS = {
-  neutral: "Provide a clear, straightforward analysis.",
-  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 dedent6`A brief summary of the video in no more than ${wordCount} words. Shorter is fine.
+      return dedent7`A brief summary of the video in no more than ${wordCount} words. Shorter is fine.
         Focus on the single most important subject or action.
         Write in present tense.`;
     }
-    return dedent6`A brief summary of the audio content in no more than ${wordCount} words. Shorter is fine.
+    return dedent7`A brief summary of the audio content in no more than ${wordCount} words. Shorter is fine.
       Focus on the single most important topic or theme.
       Write in present tense.`;
   }
   if (wordCount > DESCRIPTION_LENGTH_THRESHOLD_LARGE) {
     if (contentType === "video") {
-      return dedent6`A detailed summary that describes what happens across the video.
+      return dedent7`A detailed summary that describes what happens across the video.
         Never exceed ${wordCount} words, but shorter is perfectly fine. 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 dedent6`A detailed summary that describes the audio content.
+    return dedent7`A detailed summary that describes the audio content.
       Never exceed ${wordCount} words, but shorter is perfectly fine. 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 dedent6`A summary that describes what happens across the video.
+    return dedent7`A summary that describes what happens across the video.
       Never exceed ${wordCount} words, but shorter is perfectly fine. 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 dedent6`A summary that describes the audio content.
+  return dedent7`A summary that describes the audio content.
     Never exceed ${wordCount} words, but shorter is perfectly fine. 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 buildSceneAwareDescriptionGuidance(wordCount) {
+function buildSceneAwareDescriptionGuidance(wordCount, hasStoryboard) {
+  const storyboardGuidance = hasStoryboard ? "Use the storyboard as direct visual evidence and the scene context to preserve progression." : "Use the ordered scene context as grounded evidence of the content and its progression.";
   if (wordCount < DESCRIPTION_LENGTH_THRESHOLD_SMALL) {
-    return dedent6`A brief summary of the video in no more than ${wordCount} words. Shorter is fine.
+    return dedent7`A brief summary of the video in no more than ${wordCount} words. Shorter is fine.
       Focus on the single most important subject or action across the ordered scenes.
-      Use the storyboard as direct visual evidence and the scene context to preserve progression.
+      ${storyboardGuidance}
       Write in present tense.`;
   }
   if (wordCount > DESCRIPTION_LENGTH_THRESHOLD_LARGE) {
-    return dedent6`A detailed summary that describes what happens across the ordered scenes.
+    return dedent7`A detailed summary that describes what happens across the ordered scenes.
       Never exceed ${wordCount} words, but shorter is perfectly fine. You may use multiple sentences.
       Be thorough: cover subjects, actions, setting, progression, and notable details supported across scenes.
-      Use the storyboard as direct visual evidence and the scene context to preserve progression.
+      ${storyboardGuidance}
       Write in present tense. If the transcript provides dialogue or narration, incorporate key points while preserving scene progression.`;
   }
-  return dedent6`A summary that describes what happens across the ordered scenes.
+  return dedent7`A summary that describes what happens across the ordered scenes.
     Never exceed ${wordCount} words, but shorter is perfectly fine. You may use multiple sentences.
     Cover the main subjects, actions, setting, and notable progression supported across scenes.
-    Use the storyboard as direct visual evidence and the scene context to preserve progression.
+    ${storyboardGuidance}
     Write in present tense. If the transcript provides dialogue or narration, incorporate key points while preserving scene progression.`;
 }
-function createSummarizationBuilder({ titleLength, descriptionLength, tagCount } = {}, { hasSceneContext = false } = {}) {
+function createSummarizationGuidance({
+  mediaType = "video",
+  titleLength,
+  descriptionLength,
+  tagCount,
+  hasSceneContext = false,
+  hasStoryboard = true
+} = {}) {
   const titleLimit = titleLength ?? DEFAULT_TITLE_LENGTH;
+  const descriptionLimit = descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH;
   const keywordLimit = tagCount ?? DEFAULT_SUMMARY_KEYWORD_LIMIT;
-  const task = hasSceneContext ? "Analyze the storyboard frames and ordered scene context together, then generate metadata that captures the full progression of the video content." : "Analyze the storyboard frames and generate metadata that captures the essence of the video content.";
-  const qualityGuidelines = hasSceneContext ? dedent6`
-      - Follow scene_index order to understand what begins, develops, and concludes
-      - Use scene narratives and concepts as grounded evidence, preserving consistent terminology across the output
-      - Use the storyboard as direct visual evidence for whole-asset details
-      - Balance brevity with informativeness` : dedent6`
-      - 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
-      - Balance brevity with informativeness`;
-  return createPromptBuilder({
-    template: {
-      task: {
-        tag: "task",
-        content: task
-      },
-      title: {
-        tag: "title_requirements",
-        content: dedent6`
+  const isAudioOnly = mediaType === "audio";
+  const sceneAware = !isAudioOnly && hasSceneContext;
+  const qualityGuidelines = isAudioOnly ? dedent7`
+      - 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
+      - Balance brevity with informativeness` : sceneAware ? dedent7`
+        - Follow scene_index order to understand what begins, develops, and concludes
+        - Use scene narratives and concepts as grounded evidence, preserving consistent terminology across the output
+        ${hasStoryboard ? "- Use the storyboard as direct visual evidence for whole-asset details" : "- Ground whole-asset details in the provided scene context and transcript when available"}
+        - Balance brevity with informativeness` : dedent7`
+        - 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
+        - Balance brevity with informativeness`;
+  return {
+    title: {
+      tag: "title_requirements",
+      content: isAudioOnly ? dedent7`
+          A concise, label-style title — not a sentence or description.
+          Never exceed ${titleLimit} words, but shorter is better.
+          Think of how a podcast episode title or playlist entry would read — e.g. "Weekly News Roundup" or "Interview with Dr. Smith".
+          Start with the primary subject or topic. Never begin with "An audio of" or similar phrasing.
+          Use specific nouns over lengthy descriptions. Avoid clauses, conjunctions, or narrative structure.` : dedent7`
           A concise, label-style title — not a sentence or description.
           Never exceed ${titleLimit} words, but shorter is better.
           Think of how a video card title, playlist entry, or file name would read — e.g. "Predator: Badlands Trailer" or "Chef Prepares Holiday Feast".
           Start with the primary subject or topic. Never begin with "A video of" or similar phrasing.
           Use specific nouns over lengthy descriptions. Avoid clauses, conjunctions, or narrative structure.`
-      },
-      description: {
-        tag: "description_requirements",
-        content: hasSceneContext ? buildSceneAwareDescriptionGuidance(descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH) : buildDescriptionGuidance(descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH, "video")
-      },
-      keywords: {
-        tag: "keywords_requirements",
-        content: dedent6`
+    },
+    description: {
+      tag: "description_requirements",
+      content: sceneAware ? buildSceneAwareDescriptionGuidance(descriptionLimit, hasStoryboard) : buildDescriptionGuidance(descriptionLimit, mediaType)
+    },
+    keywords: {
+      tag: "keywords_requirements",
+      content: isAudioOnly ? dedent7`
+          Specific, searchable terms (up to ${keywordLimit}) that capture:
+          - Primary topics and themes
+          - Speakers or presenters (if named)
+          - Key concepts and terminology
+          - Content type (interview, lecture, music, etc.)
+          - Genre or style (if applicable)
+          Prefer concrete nouns and relevant terms over abstract concepts.
+          Use lowercase. Avoid redundant or overly generic terms like "audio" or "content".` : dedent7`
           Specific, searchable terms (up to ${keywordLimit}) that capture:
           - Primary subjects (people, animals, objects)
           - Actions and activities being performed
@@ -7594,57 +7614,51 @@ function createSummarizationBuilder({ titleLength, descriptionLength, tagCount }
           - Style or genre (if applicable)
           Prefer concrete nouns and action verbs over abstract concepts.
           Use lowercase. Avoid redundant or overly generic terms like "video" or "content".`
+    },
+    qualityGuidelines: { tag: "quality_guidelines", content: qualityGuidelines }
+  };
+}
+
+// src/workflows/summarization.ts
+var summarySchema = z9.object({
+  keywords: z9.array(z9.string().max(200)),
+  title: z9.string().max(500),
+  description: z9.string().max(1e4)
+});
+function buildSummarySchema(descriptionLength) {
+  const descriptionMaxChars = Math.max(2e3, descriptionLength * 15);
+  return z9.object({
+    keywords: z9.array(z9.string().max(200)),
+    title: z9.string().max(500),
+    description: z9.string().max(descriptionMaxChars)
+  });
+}
+var VALID_TONES = ["neutral", "playful", "professional"];
+var TONE_INSTRUCTIONS = {
+  neutral: "Provide a clear, straightforward analysis.",
+  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."
+};
+function createSummarizationBuilder(constraints = {}, { hasSceneContext = false } = {}) {
+  return createPromptBuilder({
+    template: {
+      task: {
+        tag: "task",
+        content: hasSceneContext ? "Analyze the storyboard frames and ordered scene context together, then generate metadata that captures the full progression of the video content." : "Analyze the storyboard frames and generate metadata that captures the essence of the video content."
       },
-      qualityGuidelines: {
-        tag: "quality_guidelines",
-        content: qualityGuidelines
-      }
+      ...createSummarizationGuidance({ ...constraints, hasSceneContext })
     },
     sectionOrder: ["task", "title", "description", "keywords", "qualityGuidelines"]
   });
 }
-function createAudioOnlyBuilder({ titleLength, descriptionLength, tagCount } = {}) {
-  const titleLimit = titleLength ?? DEFAULT_TITLE_LENGTH;
-  const keywordLimit = tagCount ?? DEFAULT_SUMMARY_KEYWORD_LIMIT;
+function createAudioOnlyBuilder(constraints = {}) {
   return createPromptBuilder({
     template: {
       task: {
         tag: "task",
         content: "Analyze the transcript and generate metadata that captures the essence of the audio content."
       },
-      title: {
-        tag: "title_requirements",
-        content: dedent6`
-          A concise, label-style title — not a sentence or description.
-          Never exceed ${titleLimit} words, but shorter is better.
-          Think of how a podcast episode title or playlist entry would read — e.g. "Weekly News Roundup" or "Interview with Dr. Smith".
-          Start with the primary subject or topic. Never begin with "An audio of" or similar phrasing.
-          Use specific nouns over lengthy descriptions. Avoid clauses, conjunctions, or narrative structure.`
-      },
-      description: {
-        tag: "description_requirements",
-        content: buildDescriptionGuidance(descriptionLength ?? DEFAULT_DESCRIPTION_LENGTH, "audio")
-      },
-      keywords: {
-        tag: "keywords_requirements",
-        content: dedent6`
-          Specific, searchable terms (up to ${keywordLimit}) that capture:
-          - Primary topics and themes
-          - Speakers or presenters (if named)
-          - Key concepts and terminology
-          - Content type (interview, lecture, music, etc.)
-          - Genre or style (if applicable)
-          Prefer concrete nouns and relevant terms over abstract concepts.
-          Use lowercase. Avoid redundant or overly generic terms like "audio" or "content".`
-      },
-      qualityGuidelines: {
-        tag: "quality_guidelines",
-        content: dedent6`
-          - 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
-          - Balance brevity with informativeness`
-      }
+      ...createSummarizationGuidance({ ...constraints, mediaType: "audio" })
     },
     sectionOrder: ["task", "title", "description", "keywords", "qualityGuidelines"]
   });