tuna-agent 0.1.125 → 0.1.126

package/dist/daemon/analyze-video-handler.js

@@ -174,30 +174,50 @@ export async function analyzeVideo(url, onProgress) {
         const finalSlots = sceneSlots.slice(0, MAX_SCENES);
         progress(`Đang cắt ${finalSlots.length} frames và phân tích...`);
         console.log('[analyze_video] Building', finalSlots.length, 'scenes (segments:', segments.length, ', duration:', durationSec, 's)');
-        const scenes = [];
+        // Step 1: Extract all frames sequentially (ffmpeg can't run in parallel on same file efficiently)
+        const frameBuffers = [];
         for (let i = 0; i < finalSlots.length; i++) {
             const slot = finalSlots[i];
             const midpoint = (slot.start + slot.end) / 2;
             const framePath = path.join(framesDir, `scene-${String(i).padStart(3, '0')}.jpg`);
             try {
-                progress(`Đang phân tích scene ${i + 1}/${finalSlots.length}...`);
                 await run(FFMPEG, ['-y', '-ss', String(midpoint), '-i', videoPath, '-vframes', '1', '-vf', 'scale=640:-1', '-q:v', '5', framePath, '-loglevel', 'error']);
                 const buf = await fs.readFile(framePath);
-                const visual_description = await visionDescribe(buf.toString('base64'), slot.voiceover);
-                scenes.push({
-                    scene_number: i + 1,
-                    timestamp_start: Math.round(slot.start * 10) / 10,
-                    timestamp_end: Math.round(slot.end * 10) / 10,
-                    thumbnail_base64: buf.toString('base64'),
-                    voiceover: slot.voiceover,
-                    visual_description,
-                });
+                frameBuffers.push({ idx: i, buf, slot });
             }
             catch (err) {
                 const msg = err instanceof Error ? err.message : String(err);
                 console.warn('[analyze_video] Frame extract failed for scene', i, msg);
             }
         }
+        // Step 2: Vision describe all frames in parallel (batch of 5)
+        progress(`Đang phân tích ${frameBuffers.length} scenes song song...`);
+        const BATCH_SIZE = 5;
+        const sceneResults = [];
+        for (let b = 0; b < frameBuffers.length; b += BATCH_SIZE) {
+            const batch = frameBuffers.slice(b, b + BATCH_SIZE);
+            progress(`Đang phân tích scene ${b + 1}-${Math.min(b + BATCH_SIZE, frameBuffers.length)}/${frameBuffers.length}...`);
+            const results = await Promise.all(batch.map(async ({ idx, buf, slot }) => {
+                try {
+                    const visual_description = await visionDescribe(buf.toString('base64'), slot.voiceover);
+                    return {
+                        scene_number: idx + 1,
+                        timestamp_start: Math.round(slot.start * 10) / 10,
+                        timestamp_end: Math.round(slot.end * 10) / 10,
+                        thumbnail_base64: buf.toString('base64'),
+                        voiceover: slot.voiceover,
+                        visual_description,
+                    };
+                }
+                catch (err) {
+                    const msg = err instanceof Error ? err.message : String(err);
+                    console.warn('[analyze_video] Vision describe failed for scene', idx, msg);
+                    return null;
+                }
+            }));
+            sceneResults.push(...results.filter((r) => r !== null));
+        }
+        const scenes = sceneResults.sort((a, b) => a.scene_number - b.scene_number);
         return {
             duration_sec: Math.round(durationSec),
             language: transcript.language || 'unknown',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tuna-agent",
-  "version": "0.1.125",
+  "version": "0.1.126",
   "description": "Tuna Agent - Run AI coding tasks on your machine",
   "bin": {
     "tuna-agent": "dist/cli/index.js"