opencode-pollinations-plugin 6.1.0-beta.12 → 6.1.0-beta.22

package/dist/tools/pollinations/transcribe_audio.js

@@ -3,40 +3,36 @@
  *
  * Updated: 2026-02-12 - Verified API Reference
  *
- * Two STT options:
- * 1. openai-audio (DEFAULT): GPT-4o Audio Preview - uses /v1/chat/completions with modalities
- *    - Least expensive option
- *    - Can handle both audio input and output
+ * 1. whisper-large-v3 (DEFAULT): High accuracy Whisper model
+ * 2. whisper-1: Standard Whisper model
+ * 3. scribe: ElevenLabs Scribe v2
  *
- * 2. whisper: OpenAI Whisper v3 - uses /v1/audio/transcriptions
- *    - POST ONLY with multipart/form-data
- *    - Specialized for transcription
- *    - Higher accuracy for long audio
+ * All models use /v1/audio/transcriptions (POST multipart)
  */
 import { tool } from '@opencode-ai/plugin/tool';
 import * as fs from 'fs';
 import * as path from 'path';
-import { getApiKey, httpsPost, httpsPostMultipart, ensureDir, formatFileSize, AUDIO_MODELS, } from './shared.js';
+import { getApiKey, httpsPostMultipart, ensureDir, formatFileSize, getAudioModels, } from './shared.js';
 // ─── Constants ─────────────────────────────────────────────────────────────
-const DEFAULT_MODEL = 'openai-audio';
-const SUPPORTED_FORMATS = ['mp3', 'wav', 'm4a', 'webm', 'mp4', 'mpeg', 'mpga', 'oga', 'ogg'];
+const DEFAULT_MODEL = 'whisper-large-v3';
+const SUPPORTED_FORMATS = ['mp3', 'wav', 'm4a', 'webm', 'mp4', 'mpeg', 'mpga', 'ogg'];
 // ─── Tool Definition ──────────────────────────────────────────────────────
-export const transcribeAudioTool = tool({
+export const polliSttTool = tool({
     description: `Transcribe audio to text using Pollinations AI.
 
 **🎙️ Models:**
 
-| Model | Endpoint | Best For | Notes |
-|-------|----------|----------|-------|
-| openai-audio | /v1/chat/completions | Short-medium audio | **DEFAULT** - lowest cost |
-| whisper | /v1/audio/transcriptions | Long audio, high accuracy | POST multipart only |
+| Model | Supplier | Notes |
+|-------|----------|-------|
+| whisper-large-v3 | OpenAI | **DEFAULT** - High accuracy, long audio |
+| whisper-1 | OpenAI | Standard accuracy |
+| scribe | ElevenLabs | Scribe v2 model |
 
 **📁 Supported Formats:**
-mp3, wav, m4a, webm, mp4, mpeg, mpga, oga, ogg
+mp3, wav, m4a, webm, mp4, mpeg, mpga, ogg
 
 **💡 Tips:**
-- Use \`openai-audio\` for cost-effective transcription
-- Use \`whisper\` for highest accuracy on long recordings
+- Use \`whisper-large-v3\` for the highest accuracy on long recordings
 - Supports both local files and URLs
 
 **📋 Output:**
@@ -57,10 +53,11 @@ mp3, wav, m4a, webm, mp4, mpeg, mpga, oga, ogg
         }
         const model = args.model || DEFAULT_MODEL;
         // Validate model
-        const modelInfo = AUDIO_MODELS[model];
+        const audioModels = getAudioModels();
+        const modelInfo = audioModels[model];
         if (!modelInfo || (modelInfo.type !== 'stt' && modelInfo.type !== 'both')) {
             return `❌ Modèle STT inconnu: ${model}
-💡 Modèles STT disponibles: ${Object.entries(AUDIO_MODELS)
+💡 Modèles STT disponibles: ${Object.entries(audioModels)
                 .filter(([, info]) => info.type === 'stt' || info.type === 'both')
                 .map(([name]) => name)
                 .join(', ')}`;
@@ -130,58 +127,20 @@ mp3, wav, m4a, webm, mp4, mpeg, mpga, oga, ogg
         try {
             let transcript = '';
             let detectedLanguage = '';
-            if (model === 'openai-audio') {
-                // === OpenAI Audio: Use modalities endpoint ===
-                // Convert audio to base64
-                const base64Audio = audioBuffer.toString('base64');
-                const mimeType = fileName.endsWith('.mp3') ? 'audio/mpeg' :
-                    fileName.endsWith('.wav') ? 'audio/wav' :
-                        'audio/mp4';
-                const response = await httpsPost('https://gen.pollinations.ai/v1/chat/completions', {
-                    model: 'openai-audio',
-                    modalities: ['text', 'audio'],
-                    messages: [
-                        {
-                            role: 'user',
-                            content: [
-                                {
-                                    type: 'text',
-                                    text: args.language
-                                        ? `Transcribe this audio to text. Language: ${args.language}`
-                                        : 'Transcribe this audio to text.'
-                                },
-                                {
-                                    type: 'input_audio',
-                                    input_audio: {
-                                        data: base64Audio,
-                                        format: fileName.endsWith('.mp3') ? 'mp3' : 'wav'
-                                    }
-                                }
-                            ]
-                        }
-                    ]
-                }, {
-                    'Authorization': `Bearer ${apiKey}`,
-                });
-                const data = JSON.parse(response.data.toString());
-                transcript = data.choices?.[0]?.message?.content || '';
-            }
-            else if (model === 'whisper') {
-                // === Whisper: Use multipart endpoint ===
-                const fields = {
-                    file: audioBuffer,
-                    model: 'whisper',
-                };
-                if (args.language) {
-                    fields.language = args.language;
-                }
-                const response = await httpsPostMultipart('https://gen.pollinations.ai/v1/audio/transcriptions', fields, {
-                    'Authorization': `Bearer ${apiKey}`,
-                });
-                const data = JSON.parse(response.data.toString());
-                transcript = data.text || '';
-                detectedLanguage = data.language || '';
+            // === All STT models use multipart endpoint ===
+            const fields = {
+                file: audioBuffer,
+                model: model,
+            };
+            if (args.language) {
+                fields.language = args.language;
             }
+            const response = await httpsPostMultipart('https://gen.pollinations.ai/v1/audio/transcriptions', fields, {
+                'Authorization': `Bearer ${apiKey}`,
+            });
+            const data = JSON.parse(response.data.toString());
+            transcript = data.text || '';
+            detectedLanguage = data.language || '';
             if (!transcript) {
                 return `❌ Aucune transcription générée.
 💡 Vérifiez que l'audio contient de la parole claire.`;

package/dist/tools/power/extract_audio.js

@@ -5,6 +5,7 @@ import * as os from 'os';
 import * as https from 'https';
 import * as http from 'http';
 import { resolveOutputDir, formatFileSize, safeName, formatTimestamp, TOOL_DIRS } from '../shared.js';
+import { hasSystemFFmpeg, getFFmpegInstallInstructions, runFFmpeg, runFFprobe } from '../ffmpeg.js';
 // ─── Download helper ────────────────────────────────────────────────────────
 function downloadFile(url) {
     return new Promise((resolve, reject) => {
@@ -27,16 +28,6 @@ function downloadFile(url) {
         req.setTimeout(120000, () => { req.destroy(); reject(new Error('Timeout (120s)')); });
     });
 }
-// ─── FFmpeg check ───────────────────────────────────────────────────────────
-function hasSystemFFmpeg() {
-    try {
-        require('child_process').execSync('ffmpeg -version', { stdio: 'ignore' });
-        return true;
-    }
-    catch {
-        return false;
-    }
-}
 // ─── Tool Definition ────────────────────────────────────────────────────────
 export const extractAudioTool = tool({
     description: `Extract the audio track from a video file or URL.
@@ -63,9 +54,7 @@ Free to use — no API key needed.`,
                 `❌ FFmpeg non trouvé!`,
                 ``,
                 `Cet outil nécessite ffmpeg :`,
-                `  • Linux: sudo apt install ffmpeg`,
-                `  • macOS: brew install ffmpeg`,
-                `  • Windows: choco install ffmpeg`,
+                getFFmpegInstallInstructions().split('\n').map(l => `  • ${l}`).join('\n')
             ].join('\n');
         }
         // Resolve source
@@ -89,8 +78,14 @@ Free to use — no API key needed.`,
         }
         // Check if video has audio
         try {
-            const { execSync } = require('child_process');
-            const probe = execSync(`ffprobe -v quiet -select_streams a -show_entries stream=codec_type -of csv=p=0 "${videoPath}"`, { timeout: 10000, encoding: 'utf-8' }).trim();
+            // Using runFFprobe helper
+            const probe = runFFprobe([
+                '-v', 'quiet',
+                '-select_streams', 'a',
+                '-show_entries', 'stream=codec_type',
+                '-of', 'csv=p=0',
+                videoPath
+            ], { timeout: 10000 }).trim();
             if (!probe) {
                 if (isRemote)
                     try {
@@ -109,31 +104,30 @@ Free to use — no API key needed.`,
         const outputFile = path.join(outputDir, `${baseName}.${outputFormat}`);
         try {
             context.metadata({ title: `🎵 Extraction audio...` });
-            const { execSync } = require('child_process');
-            // Build ffmpeg command
-            let cmd = `ffmpeg -y -i "${videoPath}" -vn`;
+            // Build ffmpeg args
+            const ffmpegArgs = ['-y', '-i', videoPath, '-vn'];
             // Time range
             if (args.start)
-                cmd += ` -ss ${args.start}`;
+                ffmpegArgs.push('-ss', args.start);
             if (args.end)
-                cmd += ` -to ${args.end}`;
+                ffmpegArgs.push('-to', args.end);
             // Format-specific encoding
             switch (outputFormat) {
                 case 'mp3':
-                    cmd += ` -acodec libmp3lame -q:a 2`;
+                    ffmpegArgs.push('-acodec', 'libmp3lame', '-q:a', '2');
                     break;
                 case 'wav':
-                    cmd += ` -acodec pcm_s16le`;
+                    ffmpegArgs.push('-acodec', 'pcm_s16le');
                     break;
                 case 'aac':
-                    cmd += ` -acodec aac -b:a 192k`;
+                    ffmpegArgs.push('-acodec', 'aac', '-b:a', '192k');
                     break;
                 case 'flac':
-                    cmd += ` -acodec flac`;
+                    ffmpegArgs.push('-acodec', 'flac');
                     break;
             }
-            cmd += ` "${outputFile}"`;
-            execSync(cmd, { stdio: 'ignore', timeout: 120000 });
+            ffmpegArgs.push(outputFile);
+            runFFmpeg(ffmpegArgs, { timeout: 120000 });
             // Cleanup
             if (isRemote && fs.existsSync(videoPath)) {
                 try {
@@ -148,7 +142,12 @@ Free to use — no API key needed.`,
             // Get audio duration
             let durationStr = 'N/A';
             try {
-                const durRaw = execSync(`ffprobe -v quiet -show_entries format=duration -of csv=p=0 "${outputFile}"`, { timeout: 5000, encoding: 'utf-8' }).trim();
+                const durRaw = runFFprobe([
+                    '-v', 'quiet',
+                    '-show_entries', 'format=duration',
+                    '-of', 'csv=p=0',
+                    outputFile
+                ], { timeout: 5000 }).trim();
                 const dur = parseFloat(durRaw);
                 if (!isNaN(dur))
                     durationStr = formatTimestamp(dur);

package/dist/tools/power/extract_frames.js

@@ -5,12 +5,17 @@ import * as os from 'os';
 import * as https from 'https';
 import * as http from 'http';
 import { resolveOutputDir, formatFileSize, safeName, formatTimestamp, TOOL_DIRS } from '../shared.js';
+import { hasSystemFFmpeg, getFFmpegInstallInstructions, runFFmpeg, runFFprobe } from '../ffmpeg.js';
 function extractMetadata(videoPath) {
     try {
-        const { execSync } = require('child_process');
         // Use ffprobe JSON output for reliable parsing
-        const probeCmd = `ffprobe -v quiet -print_format json -show_format -show_streams "${videoPath}"`;
-        const raw = execSync(probeCmd, { timeout: 15000, encoding: 'utf-8' });
+        const raw = runFFprobe([
+            '-v', 'quiet',
+            '-print_format', 'json',
+            '-show_format',
+            '-show_streams',
+            videoPath
+        ]);
         const data = JSON.parse(raw);
         const videoStream = data.streams?.find((s) => s.codec_type === 'video');
         const audioStream = data.streams?.find((s) => s.codec_type === 'audio');
@@ -82,38 +87,32 @@ function downloadVideo(url) {
         req.setTimeout(120000, () => { req.destroy(); reject(new Error('Timeout téléchargement (120s)')); });
     });
 }
-// ─── FFmpeg availability ────────────────────────────────────────────────────
-function hasSystemFFmpeg() {
-    try {
-        const { execSync } = require('child_process');
-        execSync('ffmpeg -version', { stdio: 'ignore' });
-        return true;
-    }
-    catch {
-        return false;
-    }
-}
 // ─── Frame extraction ───────────────────────────────────────────────────────
 function extractWithSystemFFmpeg(videoPath, outputDir, baseName, options) {
-    const { execSync } = require('child_process');
     const outputs = [];
-    let cmd = `ffmpeg -y -i "${videoPath}"`;
     if (options.at_time) {
+        // Single Frame Extraction
         const singleOutput = path.join(outputDir, `${baseName}_at_${options.at_time.replace(/:/g, '-')}.png`);
-        cmd += ` -ss ${options.at_time} -frames:v 1 "${singleOutput}"`;
-        execSync(cmd, { stdio: 'ignore', timeout: 60000 });
+        runFFmpeg([
+            '-y', '-i', videoPath,
+            '-ss', options.at_time,
+            '-frames:v', '1',
+            singleOutput
+        ], { timeout: 60000 });
         if (fs.existsSync(singleOutput))
             outputs.push(singleOutput);
     }
     else {
-        if (options.start)
-            cmd += ` -ss ${options.start}`;
-        if (options.end)
-            cmd += ` -to ${options.end}`;
+        // Range Extraction
         const fps = options.fps || 1;
         const outputPattern = path.join(outputDir, `${baseName}_%03d.png`);
-        cmd += ` -vf "fps=${fps}" "${outputPattern}"`;
-        execSync(cmd, { stdio: 'ignore', timeout: 120000 });
+        const args = ['-y', '-i', videoPath];
+        if (options.start)
+            args.push('-ss', options.start);
+        if (options.end)
+            args.push('-to', options.end);
+        args.push('-vf', `fps=${fps}`, outputPattern);
+        runFFmpeg(args, { timeout: 120000 });
         fs.readdirSync(outputDir)
             .filter(f => f.startsWith(baseName) && f.endsWith('.png'))
             .sort()
@@ -146,9 +145,7 @@ Free to use — no API key needed.`,
                 `❌ FFmpeg non trouvé!`,
                 ``,
                 `Cet outil nécessite ffmpeg. Installez-le :`,
-                `  • Linux: sudo apt install ffmpeg`,
-                `  • macOS: brew install ffmpeg`,
-                `  • Windows: choco install ffmpeg`,
+                getFFmpegInstallInstructions().split('\n').map(l => `  • ${l}`).join('\n')
             ].join('\n');
         }
         // Resolve source: URL → download, path → validate

package/dist/tools/power/remove_background.js

@@ -4,12 +4,13 @@ import * as https from 'https';
 import * as fs from 'fs';
 import * as path from 'path';
 import { resolveOutputDir, formatFileSize, TOOL_DIRS } from '../shared.js';
+import { getConfigDir } from '../../server/config.js';
 // ─── Provider Defaults ───────────────────────────────────────────────────────
 const CUT_API_URL = 'https://cut.esprit-artificiel.com';
 const CUT_API_KEY = 'sk-cut-fkomEA2026-hybridsoap161200';
 const BACKGROUNDCUT_API_URL = 'https://backgroundcut.co/api/v1/cut/';
 // ─── Key Storage ─────────────────────────────────────────────────────────────
-const KEYS_FILE = path.join(process.env.HOME || process.env.USERPROFILE || '/tmp', '.pollinations', 'backgroundcut_keys.json');
+const KEYS_FILE = path.join(getConfigDir(), 'backgroundcut_keys.json');
 function loadKeys() {
     try {
         if (fs.existsSync(KEYS_FILE)) {

package/dist/tools/power/rmbg_keys.js

@@ -2,7 +2,8 @@ import { tool } from '@opencode-ai/plugin/tool';
 import * as fs from 'fs';
 import * as path from 'path';
 // ─── Shared Logic (Duplicated from rmbg to avoid circular deps if not using shared.ts for this) ──
-const KEYS_FILE = path.join(process.env.HOME || process.env.USERPROFILE || '/tmp', '.pollinations', 'backgroundcut_keys.json');
+import { CONFIG_DIR } from '../../server/config.js';
+const KEYS_FILE = path.join(CONFIG_DIR, 'backgroundcut_keys.json');
 function loadKeys() {
     try {
         if (fs.existsSync(KEYS_FILE)) {

package/dist/tools/shared.js

@@ -29,9 +29,15 @@ export function resolveOutputDir(subdir, customPath) {
     if (customPath) {
         // If customPath is absolute, use it directly
         // If relative, resolve from cwd
-        dir = path.isAbsolute(customPath)
-            ? customPath
-            : path.resolve(process.cwd(), customPath);
+        // Handle tilde ~ manual expansion for cross-platform support
+        if (customPath.startsWith('~')) {
+            dir = path.join(os.homedir(), customPath.slice(1));
+        }
+        else {
+            dir = path.isAbsolute(customPath)
+                ? customPath
+                : path.resolve(process.cwd(), customPath);
+        }
     }
     else {
         dir = path.join(DEFAULT_BASE, subdir);

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "opencode-pollinations-plugin",
     "displayName": "Pollinations AI (V5.9)",
-    "version": "6.1.0-beta.12",
+    "version": "6.1.0-beta.22",
     "description": "Native Pollinations.ai Provider Plugin for OpenCode",
     "publisher": "pollinations",
     "repository": {
@@ -61,4 +61,4 @@
         "@types/qrcode": "^1.5.6",
         "typescript": "^5.0.0"
     }
-}
+}