Move to OpenRouter and add filters

Author: HugoGresse

scripts/.env.example

@@ -1,5 +1,14 @@
 YOUTUBE_PLAYLIST_ID=YOUTUBE_PLAYLIST_ID
 OPENPLANNER_EVENT_ID=
 OPENPLANNER_API_KEY=
+
+# Speech-to-text (youtubeSubtitleEdit.js)
 GLADIA_API_KEY=
-OPENAI_API_KEY=
+GLADIA_MODEL=solaria-3
+
+# Keyword extraction via OpenRouter (youtubeSubtitleEdit.js)
+OPENROUTER_API_KEY=
+OPENROUTER_MODEL=z-ai/glm-5.2
+
+# Thumbnails source dir for youtubeBatchEdit.js (defaults to youtube-thumbnails/output/<eventId>)
+THUMBNAILS_DIR=

scripts/package.json

@@ -8,7 +8,9 @@
         "test": "echo \"Error: no test specified\" && exit 1",
         "typecheck": "tsc --noEmit",
         "generate:thumbnails": "tsx youtube-thumbnails/generateThumbnails.ts",
-        "youtube:batch-edit": "tsx youtube/youtubeBatchEdit.js"
+        "youtube:batch-edit": "tsx youtube/youtubeBatchEdit.js",
+        "youtube:subtitle-edit": "tsx youtube/youtubeSubtitleEdit.js",
+        "youtube:subtitle-upload": "tsx youtube/youtubeSubtitleUpload.js"
     },
     "author": "",
     "license": "ISC",

scripts/youtube/utils/youtubeAPI.js

@@ -158,15 +158,22 @@ const getChannel = async (auth) => {
 export const getVideosFromPlaylist = async (auth, channelId, playlistId) => {
     var service = google.youtube('v3')
 
-    // get all videos in playlist
-    const playlistItems = await service.playlistItems.list({
-        auth: auth,
-        part: 'snippet,contentDetails',
-        playlistId: playlistId,
-        maxResults: 50,
-    })
+    // get all videos in playlist, paging through 50-item batches (API max per page)
+    const items = []
+    let pageToken = undefined
+    do {
+        const playlistItems = await service.playlistItems.list({
+            auth: auth,
+            part: 'snippet,contentDetails',
+            playlistId: playlistId,
+            maxResults: 50,
+            pageToken: pageToken,
+        })
+        items.push(...playlistItems.data.items)
+        pageToken = playlistItems.data.nextPageToken
+    } while (pageToken)
 
-    return playlistItems.data.items
+    return items
 }
 
 export const listVideoCategories = async (auth) => {

scripts/youtube/youtubeBatchEdit.js

@@ -95,7 +95,12 @@ const main = async () => {
 
     const videosWithValidSession = joinYoutubeAndOpenPlannerData(videos, openPlannerContent)
 
+    // Optional: restrict the run to specific session titles passed as CLI args
+    // e.g. node youtube/youtubeBatchEdit.js "Title one" "Title two"
+    const titleFilter = process.argv.slice(2)
+
     const videosWithValidSessionAndDescription = videosWithValidSession
+        .filter((video) => titleFilter.length === 0 || titleFilter.some((title) => video.session.title === title))
         .map((video) => {
             return {
                 ...video,
@@ -104,6 +109,12 @@ const main = async () => {
         })
         .filter((video) => !!video)
 
+    if (titleFilter.length > 0) {
+        console.log(
+            `Filtering to ${titleFilter.length} title(s): matched ${videosWithValidSessionAndDescription.length} video(s)`
+        )
+    }
+
     // Update video metadata
     for (const video of videosWithValidSessionAndDescription) {
         const tagBasedOnOpenPlannerCategory = openPlannerContent.event.categories.find((category) => {

scripts/youtube/youtubeSubtitleEdit.js

@@ -8,7 +8,7 @@ import { joinYoutubeAndOpenPlannerData } from './utils/joinYoutubeAndOpenPlanner
 import 'dotenv/config'
 
 const POLLING_INTERVAL = 5000 // 5 seconds
-const CONCURRENT_JOBS = 5
+const CONCURRENT_JOBS = 1
 
 // This whole is here to generate subtitles for a youtube video
 // using Gladia & ChatGPT. It won't upload subtitles to youtube
@@ -18,8 +18,8 @@ const CONCURRENT_JOBS = 5
 //  - Fill the .env file with the following variables:
 //    - GLADIA_API_KEY
 //    - GLADIA_MODEL (optional, defaults to solaria-3; use solaria-1 for 100+ languages)
-//    - OPENAI_API_KEY
-//    - OPENAI_MODEL (optional, defaults to gpt-4o-mini)
+//    - OPENROUTER_API_KEY
+//    - OPENROUTER_MODEL (optional, defaults to openai/gpt-4o-mini)
 //    - YOUTUBE_PLAYLIST_ID
 //    - OPENPLANNER_EVENT_ID
 //  - Ensure you have youtube credentials for API in ~/.credentials/youtube.credentials.json
@@ -31,18 +31,21 @@ const CONCURRENT_JOBS = 5
 //  - If any SRT or keywords are already generated, they won't be recreated.
 
 const GLADIA_API_KEY = process.env.GLADIA_API_KEY
-const OPENAI_API_KEY = process.env.OPENAI_API_KEY
+const OPENROUTER_API_KEY = process.env.OPENROUTER_API_KEY
 
-if (!GLADIA_API_KEY || !OPENAI_API_KEY) {
-    throw new Error('GLADIA_API_KEY and OPENAI_API_KEY must be set')
+if (!GLADIA_API_KEY || !OPENROUTER_API_KEY) {
+    throw new Error('GLADIA_API_KEY and OPENROUTER_API_KEY must be set')
 }
 
+const OPENROUTER_ENDPOINT = 'https://openrouter.ai/api/v1/chat/completions'
+
 const GLADIA_TRANSCRIPTION_ENDPOINT = 'https://api.gladia.io/v2/pre-recorded'
 // Gladia speech-to-text model. Defaults to solaria-3 (newest, tuned for European languages
 // incl. French). Override with GLADIA_MODEL, e.g. solaria-1 for 100+ languages / code-switching.
 const GLADIA_MODEL = process.env.GLADIA_MODEL || 'solaria-3'
-// OpenAI model used to extract keywords. Override with OPENAI_MODEL.
-const OPENAI_MODEL = process.env.OPENAI_MODEL || 'gpt-4o-mini'
+// OpenRouter model used to extract keywords. Override with OPENROUTER_MODEL
+// (any OpenRouter model slug, e.g. openai/gpt-4o-mini, anthropic/claude-3.5-haiku).
+const OPENROUTER_MODEL = process.env.OPENROUTER_MODEL || 'z-ai/glm-5.2'
 
 async function getTranscriptionIdFromGladia(audioUrl, customVocabulary) {
     const headers = {
@@ -128,31 +131,33 @@ async function generateKeywords(session) {
     let keywords = []
     try {
         const response = await axios.post(
-            'https://api.openai.com/v1/chat/completions',
+            OPENROUTER_ENDPOINT,
             {
-                model: OPENAI_MODEL,
+                model: OPENROUTER_MODEL,
                 messages: [
                     { role: 'system', content: 'You are a helpful assistant.' },
                     { role: 'user', content: prompt },
                 ],
-                max_tokens: 500,
+                max_tokens: 2000,
                 temperature: 0,
             },
             {
                 headers: {
-                    Authorization: `Bearer ${OPENAI_API_KEY}`,
+                    Authorization: `Bearer ${OPENROUTER_API_KEY}`,
                     'Content-Type': 'application/json',
                 },
             }
         )
 
-        if (response.data && response.data.choices && response.data.choices.length > 0) {
-            const rawKeywords = response.data.choices[0].message.content
-            keywords = rawKeywords.replaceAll('`', '').replace('json', '')
-            return JSON.parse(keywords)
+        const rawKeywords = response.data?.choices?.[0]?.message?.content
+        if (!rawKeywords) {
+            throw new Error('Empty content in model response')
         }
 
-        throw new Error('No keywords found in response')
+        // Extract the JSON array from the response (models may wrap it in ```json fences or prose)
+        const match = rawKeywords.match(/\[[\s\S]*\]/)
+        keywords = JSON.parse(match ? match[0] : rawKeywords.replaceAll('`', '').replace('json', ''))
+        return keywords
     } catch (error) {
         console.error(`❌ Error generating keywords for session: ${session.title}`, error.message, keywords)
         return keywords