La route /api/ai/generate connaissait la cause exacte de l'échec
(configuration absente, clé invalide, crédits épuisés, modèle inconnu)
mais la remplaçait par un « Failed to generate text » en 500, et
l'interface affichait de son côté un message figé. Sans accès aux logs
du serveur, l'utilisateur n'avait donc aucun moyen de savoir quoi
corriger.
Le motif renvoyé par OpenRouter est désormais extrait, tronqué et
propagé jusqu'au bandeau d'erreur, avec un code HTTP qui distingue un
problème de configuration (400) d'un refus du service tiers (502).
La clé API n'apparaît à aucun moment dans ces messages.
Même traitement pour /api/reels/generate-text, qui souffrait du même
masquage.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018uUPZ1GNpEkYB1dpXHYPek
TikTok exige de prouver la propriété du domaine ("URL properties") avant
d'autoriser la Content Posting API, en servant un fichier à la racine du
site.
Le dossier public-root/ est exposé publiquement à la racine du domaine et
monté avant le catch-all du frontend. Y déposer un fichier suffit : pas de
modification de code ni de reconstruction du client, ce qui couvre aussi
les prochaines vérifications (autre domaine, Google, Meta).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018uUPZ1GNpEkYB1dpXHYPek
Renseigne l'identité de FROUARD DISTRIBUTION (enseigne LA FOIR'FOUILLE)
comme valeur par défaut des pages /terms et /privacy : forme juridique,
capital, siège et numéro RCS. Les relecteurs TikTok ouvrent ces pages
pendant l'audit et refusent les mentions incomplètes.
Chaque champ reste surchargeable par variable d'environnement si l'outil
venait à être exploité par une autre société du groupe.
L'adresse de contact reste à fournir via LEGAL_CONTACT_EMAIL : elle ne
figure pas au registre, et le RGPD impose un point de contact pour
l'exercice des droits.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018uUPZ1GNpEkYB1dpXHYPek
L'audit de l'application développeur TikTok impose une URL publique pour
les conditions d'utilisation et pour la politique de confidentialité, et
les relecteurs les ouvrent réellement.
Les pages sont rendues en HTML côté serveur, sans authentification, pour
rester lisibles par un robot qui n'exécute pas le JavaScript du client.
La politique décrit précisément ce que l'intégration fait : données reçues
de TikTok (open_id, nom d'affichage, avatar, jetons), finalité de
publication, chiffrement des jetons au repos, suppression à la
déconnexion et marche à suivre pour retirer l'autorisation.
Les mentions de l'exploitant se configurent via LEGAL_COMPANY_NAME,
LEGAL_COMPANY_ADDRESS et LEGAL_CONTACT_EMAIL.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018uUPZ1GNpEkYB1dpXHYPek
Ajoute la gestion multi-comptes TikTok et permet d'envoyer la même vidéo
sur plusieurs pages Facebook ET plusieurs comptes TikTok en une seule
publication.
Un compte TikTok vit dans social_pages comme une page Facebook : il hérite
donc des permissions par utilisateur, de la planification et du calendrier.
Trois particularités de l'API TikTok structurent l'implémentation :
- pas de jeton saisi à la main : chaque compte passe par OAuth, et
l'access token (24h) est renouvelé via le refresh token (1 an) avant
chaque publication ;
- la publication est asynchrone : l'upload rend un publish_id, et un
poller récupère l'identifiant définitif du post ;
- creator_info doit être interrogé avant chaque envoi pour ne demander
qu'un niveau de confidentialité réellement autorisé sur le compte.
Serveur :
- service TikTok (OAuth, creator_info, upload FILE_UPLOAD par chunks,
suivi de statut) et routes de connexion/configuration
- schéma : platform 'tiktok', refresh token et scopes sur social_pages,
publish_id/publish_status sur scheduled_posts, table tiktok_config
- routage par plateforme dans les deux flux de reels et le planificateur
- cron de suivi des publications, token manager et analytics adaptés
Client :
- connexion et reconnexion des comptes TikTok depuis « Pages gérées »
- sélection combinée pages Facebook / comptes TikTok à la publication
- configuration de l'application TikTok dans les paramètres (admin)
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018uUPZ1GNpEkYB1dpXHYPek
Address Codex review: the debug log previously included the key's
first 10 characters. Replace with a non-reversible SHA-256 fingerprint
so logs remain useful for correlating support reports without ever
exposing key material.
An OpenRouter key with trailing whitespace/newline (common from
copy-paste) produces "Missing Authentication header" from their API,
which looked identical to a missing key. Trim apiKey on save (schema
level) and on use (defense in depth), fail fast with a clear message
if the stored key is empty, and log a masked key preview + model on
each generation call to make future auth failures diagnosable from
container logs.
generatePostText ignored the userId it received and always read an
arbitrary row via getAnyOpenrouterConfig() (LIMIT 1, no ORDER BY),
so a model chosen and saved in Settings could be shadowed by another
stale config row. Now it prefers the requesting user's own config,
falling back to the most recently updated shared one. Also swap the
retired anthropic/claude-3.5-sonnet default for a live OpenRouter slug.
Add GEMINI_API_KEY to docker-compose so it can be set in Portainer.
Fall back to this env var when the DB app_config has no geminiApiKey,
so Gemini TTS works without requiring the user to save the key through
the Settings UI.
https://claude.ai/code/session_01QBvwHAMZzVYfWvy1U5paat
The TTS preview route hardcoded the Gemini API key to undefined when
calling the ffmpeg service, causing the service to silently fall back to
Edge TTS even when the user selected Gemini. Mirror the lookup already
used in the reel processing route so the selected engine is honored.
https://claude.ai/code/session_01QBvwHAMZzVYfWvy1U5paat
- ttsSyncService.calculateSyncTiming now accepts ttsEngine parameter
- sync-info route extracts ttsVoice/ttsEngine from req.body instead of voice
- Previously it always fell back to Edge TTS even when Gemini was selected
(because server stored 'fr-FR-Standard-B' but route only used 'voice')
- Store Gemini API key globally in appConfig (single key for all users)
- Add /api/settings/gemini GET/POST/DELETE routes for API key management
- Backend: add tts_engine parameter ("edge" or "gemini") to FFmpeg service
- Backend: add generate_tts_gemini() using Google Cloud TTS REST API
- Frontend: Settings page shows Google Gemini API key input card
- Frontend: new-reel, mobile/new-reel, remotion-video, mobile/remotion-video
pages now have Edge/Gemini engine toggle and French voice selector
- Fix tts-preview route to extract ttsVoice from req.body instead of
undefined voice variable
- Remove piper_url from ReelRequest model and all function signatures
- Remove generate_tts_piper() function and Piper branch in generate_tts_with_subs()
- Remove /api/piper/config routes from server/routes.ts
- Remove piperConfig table, schemas and storage methods
- Remove piper_config migration
- Remove Piper TTS settings UI card
- Update "Piper TTS" labels to "TTS — voix activée"
edge_tts is now the only TTS engine, using precise word-boundary timing
Remove Minimax Speech API and Freesound integrations entirely.
Add Piper TTS as the sole TTS provider via configurable HTTP URL.
- Add piper_config DB table (url field, per-user)
- Add GET/POST /api/piper/config routes
- Python: replace generate_tts_minimax with generate_tts_piper (GET ?text=, WAV→MP3)
- Simplify generate_tts_with_subs: piper_url param replaces tts_provider+minimax fields
- Drop ttsVoice/ttsProvider from all UI, API, and background job params
- Settings page: replace Minimax+Freesound cards with single Piper URL input
- Remove freeSoundService init from server startup and CSP headers
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Pass GroupId query param to Minimax T2A v2 API — required for paid
plan quota allocation. Without it, Minimax defaults to (0/0 used).
- shared/schema.ts: groupId field on minimaxConfig table
- server/migrate.ts: ADD COLUMN IF NOT EXISTS group_id
- server/routes/reels.ts: fetch and pass groupId alongside apiKey
- server/services/ffmpeg.ts: minimax_group_id in request interface/body
- ffmpeg-service/main.py: GroupId in URL, threaded through all call sites
- client/src/pages/settings.tsx: Group ID input in Minimax settings card
- Python: capture tts_error_msg on exception, return in response
- Python: log tts_provider, minimax_api_key presence before call
- ffmpeg.ts: read tts_error from response, log and return it
- reels.ts: store TTS error in post.generationError for visibility
- Add .min(1) validation to insertOpenrouterConfigSchema
- Strip empty/whitespace-only apiKey before fallback to existing key
- Return 400 if resulting apiKey is empty (forces user to enter valid key)
- Add minimax_config table (migration + schema + storage CRUD)
- Add GET/POST /api/minimax/config routes
- Pass tts_provider + minimax_api_key through ffmpeg service
- Add generate_tts_minimax() in Python using Minimax T2A v2 API
- Fall back to ffsubsync for subtitle sync (no WordBoundary events)
- Add Minimax config card in Settings page
- Add provider toggle (Edge TTS / Minimax) + French voices in new-reel
GET /api/v1/posts - list upcoming scheduled posts with filters
PATCH /api/v1/posts/:id - edit content, schedule, or image
DELETE /api/v1/posts/:id - remove scheduled post and cascading relations
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- New POST /api/v1/publish endpoint: download image from URL, create post and schedule it per page
- New GET /api/v1/pages endpoint: list available pages for external callers
- API key auth via X-API-Key header, configurable from admin settings (stored in DB)
- New app_config table (migration included) to store the external API key
- Admin-only settings section (desktop + mobile) to set/revoke the key
- Fallback to EXTERNAL_API_KEY env var if no DB key is configured
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- facebook.ts: use resolvePublicUrl + getMediaBuffer for reels so missing local files fall back to HTTP fetch.
- scheduler.ts: only delete local video files after the last pending scheduled post for that postId is published.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- ffmpeg-service/main.py: replace ffsubsync path with exact word-boundary timing from edge_tts for TTS subtitles. Karaoke styling preserved.
- server/services/ttsSync.ts: fix word count to match TTS-cleaned text, remove artificial punctuationPause subtraction, strip punctuation tokens from count.
- server/routes/reels.ts: remove redundant ttsSyncService calls in preview/background; word_duration is ignored by Python, these only wasted TTS generations.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Calculate optimal word_duration based on actual TTS audio duration and text punctuation.
- server/services/ttsSync.ts: new service to measure TTS audio and compute sync timing
- server/routes/reels.ts: integrate sync into preview and background processing
- client: auto-calculate sync info widget in desktop and mobile Reel creation
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Facebook resumable upload START/FINISH phases require multipart/form-data
(not application/x-www-form-urlencoded). Add full raw response logging for
each phase to diagnose any remaining issues (subcode, etc).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace simple multipart upload (which fails with error 6000 on large
files) with the 3-phase Resumable Upload API (start → transfer → finish).
This is the recommended Facebook approach for files > ~50 MB and is much
more reliable regardless of file size.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add slow zoom/pan (Ken Burns) effect to each image slide, cycling through
8 deterministic presets per image index for natural variety
- CapCut-style captions: semi-transparent pill background + yellow glow on
active word; background/text now only visible while voice is speaking
- Fix Facebook error 6000: pass Node.js Buffer directly instead of unsafe
ArrayBuffer pool-slice conversion which could corrupt the upload payload
- Fix TypeScript error: replace u-flag emoji regex with BMP surrogate pairs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remotion's headless Chromium cannot reliably fetch http://localhost
inside Docker due to network sandboxing. Instead of HTTP URLs, convert
all local /uploads/... files to data:image/...;base64,... before passing
to inputProps. This is reliable in all environments (Docker, dev, CI).
Audio files remain as HTTP URLs since they are loaded via Web Audio API
which has different network handling.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
req.file.buffer is undefined with diskStorage — must read the temp file
from req.file.path using fs.readFileSync instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Logo stored as relative /uploads/logos/... was resolving to Remotion's
bundle server (port 3000) instead of the app (port 5555).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remotion's headless Chromium runs inside the same container as the app,
so it must use http://localhost:PORT to fetch local media files.
The Docker service hostname (socialflow-app) is only reachable from
other containers on the Docker network, not from within the container itself.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remotion's headless Chromium cannot load relative /uploads/... paths.
Library images were passed as-is, causing "Error loading image" at render time.
Now resolves them using INTERNAL_APP_URL (Docker) or the request host (dev).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Images uploaded between the local storage switch and the relative URL
fix have http://hostname/uploads/... stored in DB, which is blocked by
CSP. POST /api/media/fix-urls (admin only) updates all affected rows
to use relative /uploads/... paths.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Media files were stored with absolute http://localhost:5555/... URLs,
which got blocked by the Content Security Policy when the app runs at
https://socialflow.fnancy.fr. Now stores relative /uploads/... paths
and resolves to absolute URLs only when needed by external APIs
(Facebook, FFmpeg internal service).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Media is now stored directly on the server disk (no external service).
Files are served via Express static middleware at /uploads/*.
Cleanup rules:
- Videos deleted immediately after successful publish to Facebook/Instagram
- Videos older than 7 days purged daily by cron job
- Images older than 30 days purged daily by cron job
Changes:
- server/services/minio.ts: replaced S3 client with local fs read/write
- server/index.ts: added static serving for /uploads/media, /logos, /stories
- server/services/media-purge.ts: new rules (images 30d, videos 7d)
- server/services/scheduler.ts: delete local videos after successful publish
- docker-compose.yml: add media_uploads, logos_uploads, stories_uploads volumes
- settings UI: removed credentials card, shows local storage info
- package.json: removed googleapis (unneeded)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Store endpointUrl and publicUrl in cloudinary_config DB table so
MinIO connection can be fully configured from the app settings
without needing environment variables.
- DB migration: adds endpoint_url and public_url columns
- Schema: adds endpointUrl/publicUrl to cloudinaryConfig table
- MinIO service: reads endpoint from DB (falls back to MINIO_ENDPOINT env)
- buildMinioUrl: accepts optional publicUrl override from DB config
- All routes updated to pass config.publicUrl to buildMinioUrl
- Settings UI (desktop + mobile): adds Endpoint URL and Public URL fields
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude
Michael
MichaelandClaude Opus 4.7