@sogni-ai/sogni-intelligence-client 3.24.2 → 3.24.4

package/dist-esm/public-skill-runtime/index.js

@@ -1805,10 +1805,19 @@ export function extractQuotedDialogueSegments(prompt) {
         matches.push(dialogue.trim());
         return ' ';
     });
-    const pattern = /"([^"]{1,800})"/g;
-    let match;
-    while ((match = pattern.exec(untaggedPrompt)) !== null) {
-        matches.push(match[1]);
+    const quotePatterns = [
+        /"([^"\r\n]{1,800})"/g,
+        /“([^”\r\n]{1,800})”/g,
+        /‘([^’\r\n]{1,800})’/g,
+        /«([^»\r\n]{1,800})»/g,
+        /「([^」\r\n]{1,800})」/g,
+        /『([^』\r\n]{1,800})』/g,
+    ];
+    for (const pattern of quotePatterns) {
+        let match;
+        while ((match = pattern.exec(untaggedPrompt)) !== null) {
+            matches.push(match[1]);
+        }
     }
     return matches;
 }
@@ -4765,8 +4774,14 @@ function splitStoryboardTableSections(text) {
                         ? ''
                         : normalizeStoryboardDialogue(extractQuotedDialogueSegments(dialogueCell || audioCell)[0] || dialogueCell || audioCell)
                     : '';
-        const visibleText = normalizeStoryboardVisibleText(extractStoryboardField(audioCell, ['Visible text', 'On-screen text', 'Onscreen text', 'Text', 'CTA'])
-            || (visibleTextHeaderIndex >= 0 ? compactStoryboardLine(visibleTextCell) : ''));
+        const visibleTextHeader = headers?.[visibleTextHeaderIndex] || '';
+        const visibleTextHeaderIsMixedWithAudioOrVoice = visibleTextHeaderIndex >= 0
+            && /\b(?:audio|sfx|sound|foley|music|dialogue|vo|v\.o\.|voiceover|speech|narration)\b/i.test(visibleTextHeader);
+        const explicitlyLabeledVisibleText = extractStoryboardField(audioCell, ['Visible text', 'On-screen text', 'Onscreen text', 'Text overlay', 'Text', 'CTA']);
+        const visibleText = normalizeStoryboardVisibleText(explicitlyLabeledVisibleText
+            || (visibleTextHeaderIndex >= 0 && !visibleTextHeaderIsMixedWithAudioOrVoice
+                ? compactStoryboardLine(visibleTextCell)
+                : ''));
         const audio = extractStoryboardField(audioCell, ['Audio/SFX', 'Audio', 'SFX', 'FX', 'Foley', 'Sound', 'Sounds'])
             || storyboardTableCellWithoutDialogue(soundCell || audioCell, dialogue);
         const transition = extractStoryboardField(transitionCell, ['Transition', 'Transition out', 'Transition-out', 'Edit'])
@@ -4928,34 +4943,79 @@ function splitStoryboardSections(text) {
         return inlineSections;
     return sectionHeadings.length > 0 ? sectionHeadings : tableSections;
 }
-function extractPlainNarrationScriptText(text) {
+function extractPlainNarrationScriptText(text, allowUnlabeledProse = false) {
     const source = text.trim();
     if (!source)
         return '';
     const markers = Array.from(source.matchAll(/^\s*(?:#{1,6}\s*)?(?:voice[-\s]?over\s+|narration\s+)?script\s*:\s*$/gim));
     const marker = markers[markers.length - 1];
-    if (!marker || marker.index === undefined)
+    if (marker?.index !== undefined) {
+        const body = source.slice(marker.index + marker[0].length).trim();
+        return body
+            .split(/\r?\n/)
+            .map(line => stripStoryboardMarkup(line).trim())
+            .filter(Boolean)
+            .join('\n\n')
+            .trim();
+    }
+    if (!allowUnlabeledProse)
         return '';
-    const body = source.slice(marker.index + marker[0].length).trim();
-    return body
+    return source
         .split(/\r?\n/)
         .map(line => stripStoryboardMarkup(line).trim())
         .filter(Boolean)
         .join('\n\n')
         .trim();
 }
+function countNarrativeCharacters(text) {
+    return Array.from(text)
+        .filter(character => /[\p{L}\p{N}]/u.test(character))
+        .length;
+}
 function splitPlainNarrationPhrases(text) {
     const normalized = text
         .replace(/\s+/g, ' ')
         .trim();
     if (!normalized)
         return [];
-    const sentenceMatches = normalized.match(/[^.!?]+[.!?]+(?:["')\]]+)?|[^.!?]+$/g) ?? [normalized];
+    const sentenceMatches = [];
+    let sentenceStart = 0;
+    for (let index = 0; index < normalized.length; index += 1) {
+        if (!/[.!?。！？]/u.test(normalized[index]))
+            continue;
+        const terminator = normalized[index];
+        let boundaryEnd = index + 1;
+        while (boundaryEnd < normalized.length && /["'”’»」』)\]]/u.test(normalized[boundaryEnd])) {
+            boundaryEnd += 1;
+        }
+        let nextStart = boundaryEnd;
+        while (nextStart < normalized.length && /\s/.test(normalized[nextStart]))
+            nextStart += 1;
+        const next = normalized[nextStart] ?? '';
+        const hasClosingQuote = boundaryEnd > index + 1;
+        if (hasClosingQuote && /\p{Ll}/u.test(next))
+            continue;
+        if (terminator === '.'
+            && boundaryEnd === nextStart
+            && /\d/.test(normalized[index - 1] ?? '')
+            && /\d/.test(next))
+            continue;
+        const sentence = normalized.slice(sentenceStart, boundaryEnd).trim();
+        if (sentence)
+            sentenceMatches.push(sentence);
+        sentenceStart = nextStart;
+        index = Math.max(index, boundaryEnd - 1);
+    }
+    const tail = normalized.slice(sentenceStart).trim();
+    if (tail)
+        sentenceMatches.push(tail);
+    if (sentenceMatches.length === 0)
+        sentenceMatches.push(normalized);
     return sentenceMatches
         .flatMap(sentence => sentence
         .split(/\s*(?:;|:\s+|,\s+(?=(?:and|but|so|whether|from|across|under|while|as)\b))\s*/i)
         .map(part => compactStoryboardLine(part)))
-        .filter(part => countWords(part) > 0);
+        .filter(part => countNarrativeCharacters(part) > 0);
 }
 function splitPhraseAtWordMidpoint(value) {
     const words = value.match(/\S+/g) ?? [];
@@ -5008,9 +5068,10 @@ function titleFromNarrationSegment(segment, index) {
         .join(' ');
     return words || `Narration Beat ${index + 1}`;
 }
-function synthesizeStoryboardSectionsFromPlainNarration(sourceText, frameCount, references) {
-    const script = extractPlainNarrationScriptText(sourceText);
-    if (!script || countWords(script) < 8)
+function synthesizeStoryboardSectionsFromPlainNarration(sourceText, frameCount, references, allowUnlabeledProse = false) {
+    const explicitlyLabeledNarration = /^\s*(?:#{1,6}\s*)?(?:voice[-\s]?over\s+|narration\s+)?script\s*:\s*$/im.test(sourceText);
+    const script = extractPlainNarrationScriptText(sourceText, allowUnlabeledProse);
+    if (!script || (countWords(script) < 8 && countNarrativeCharacters(script) < 24))
         return [];
     const phrases = splitPlainNarrationPhrases(script);
     const segments = normalizeNarrationSegmentCount(phrases, frameCount).slice(0, frameCount);
@@ -5025,20 +5086,28 @@ function synthesizeStoryboardSectionsFromPlainNarration(sourceText, frameCount,
         'Detail insert on concrete objects, gestures, text, environment, or visual evidence named in the narration.',
         'Tracking or reveal-style composition that moves the story into the next idea.',
     ];
-    return segments.map((segment, index) => ({
-        number: index + 1,
-        heading: `Scene ${index + 1} - ${titleFromNarrationSegment(segment, index)}`,
-        body: [
-            `Purpose: Translate narration beat ${index + 1} into a concrete ordered storyboard moment.`,
-            `Visual/Action: Create a concrete visual moment using only this narration beat and the supplied references: ${segment}`,
-            `Camera/Motion: ${cameraPresets[index % cameraPresets.length]}`,
-            'Lighting/Style: Match the visual style, genre, and tone implied by the user request and supplied references; keep the frame cinematic but readable as a storyboard panel.',
-            'Transition: Maintain continuity from the previous beat through subject, setting, gesture, camera direction, color, or motion when those cues are present.',
-            `Dialogue/VO: ${segment}`,
-            'Audio/SFX: Use only audio cues implied by the narration or user request; otherwise keep ambience or music generic and unobtrusive.',
-            referenceLine,
-        ].filter(Boolean).join('\n'),
-    }));
+    return segments.map((segment, index) => {
+        const quotedSpeech = extractQuotedDialogueSegments(segment)
+            .map(item => compactStoryboardLine(item))
+            .filter(Boolean);
+        const dialogue = explicitlyLabeledNarration
+            ? segment
+            : quotedSpeech.join(' ') || '[no dialogue]';
+        return {
+            number: index + 1,
+            heading: `Scene ${index + 1} - ${titleFromNarrationSegment(segment, index)}`,
+            body: [
+                `Purpose: Translate story beat ${index + 1} into a concrete ordered storyboard moment.`,
+                `Visual/Action: Create a concrete visual moment using only this story beat and the supplied references: ${segment}`,
+                `Camera/Motion: ${cameraPresets[index % cameraPresets.length]}`,
+                'Lighting/Style: Match the visual style, genre, and tone implied by the user request and supplied references; keep the frame cinematic but readable as a storyboard panel.',
+                'Transition: Maintain continuity from the previous beat through subject, setting, gesture, camera direction, color, or motion when those cues are present.',
+                `Dialogue/VO: ${dialogue}`,
+                'Audio/SFX: Use only audio cues implied by the story or user request; otherwise keep ambience or music generic and unobtrusive.',
+                referenceLine,
+            ].filter(Boolean).join('\n'),
+        };
+    });
 }
 function storyboardSectionsHavePreservableExplicitTiming(sections) {
     return sections.length > 1
@@ -5283,7 +5352,8 @@ function normalizeStoryboardAudioSfx(value) {
 }
 function buildSceneFromSection(section, references, fallbackTiming, planningContract) {
     const combined = `${section.heading}\n${section.body}`;
-    const timing = extractStoryboardTiming(combined) ?? fallbackTiming;
+    const timingSource = `${section.heading.replace(/^\s*(?:[-*+>#_]{1,6}\s*)*(?:[*_]{1,3})?\s*(?:Scene|Shot|Beat|Panel|Frame)\s*_?\s*\d{1,2}\b\s*(?:[-:.)|]\s*)?/i, '')}\n${section.body}`;
+    const timing = extractStoryboardTiming(timingSource) ?? fallbackTiming;
     const normalizedHeading = stripStoryboardMarkup(section.heading)
         .replace(/^\s*#{1,6}\s*/, '');
     const title = compactStoryboardLine(removeStoryboardTimingText(normalizedHeading
@@ -5903,7 +5973,14 @@ export function buildStoryboardProject(options) {
     const selectedSectionsHaveExplicitTiming = sections.length > 0 && storyboardSectionsHavePreservableExplicitTiming(sections);
     const preserveAssistantExplicitTiming = options.promptAuthorship === 'assistant' && selectedSectionsHaveExplicitTiming;
     const synthesizedSections = sections.length === 0
-        ? synthesizeStoryboardSectionsFromPlainNarration(`${sourceText}\n\n${narrativeUserIntentText}`, options.frameCount, references)
+        ? [
+            { text: approvedScriptContext, allowUnlabeledProse: true },
+            { text: narrativeUserIntentText, allowUnlabeledProse: false },
+            { text: sourceText, allowUnlabeledProse: false },
+        ]
+            .filter((candidate, index, candidates) => candidate.text && candidates.findIndex(other => other.text === candidate.text) === index)
+            .map(candidate => synthesizeStoryboardSectionsFromPlainNarration(candidate.text, options.frameCount, references, candidate.allowUnlabeledProse))
+            .find(candidate => candidate.length > 0) ?? []
         : [];
     const storyboardSections = sections.length > 0 ? sections : synthesizedSections;
     const parsedScenes = storyboardSections.length > 0