De l'audio au cartoon 3D animé Audio to animated 3D cartoon
Reconnaissance vocale avec détection des interlocuteurs, visèmes synchronisés sur les mots réellement prononcés, mise en scène 3D, et export vidéo.
Speech recognition with speaker diarization, visemes synced to the words actually spoken, a 3D stage, and video export.
L'hébergement gratuit ne peut pas exécuter ce projet
The free tier cannot run this project
Ce n'est pas une limite de l'application mais une question de taille. Les modèles de reconnaissance vocale pèsent à eux seuls 879 Mo, et une transcription complète atteint un pic d'environ 1,76 Go en mémoire.
This is not a limitation of the application but one of size. The speech recognition models alone weigh 879 MB, and a full transcription peaks at roughly 1.76 GB of memory.
| Offre | Plan | Mémoire | Memory | Verdict | Verdict |
|---|---|---|---|---|---|
| Render — Free | 512 Mo | ✗ les modèles seuls dépassent la RAM | ✗ the models alone exceed the RAM | ||
| Render — Starter — 7 $/mois par moisper month |
512 Mo | ✗ toujours 512 Mo | ✗ still 512 MB | ||
| Render — Standard — 25 $/mois par moisper month |
2 Go | ⚠ limite : le pic est à 1,76 Go | ⚠ limit: the peak is 1.76 GB | ||
| Render — Pro — 85 $/mois par moisper month |
4 Go | ✓ fonctionne, mais coûteux | ✓ works, but costly |
Le palier le moins cher qui pourrait fonctionner laisse presque aucune marge. La voie prise en charge est donc Docker sur votre propre machine, où la mémoire n'est pas plafonnée par un quota d'hébergement.
The cheapest plan that might work leaves almost no headroom. The supported path is therefore Docker on your own machine, where memory is not capped by a hosting quota.
À noter : les Docker Spaces de Hugging Face exigent un compte depuis juillet 2026 — ce n'est pas une alternative gratuite.
Worth noting: Hugging Face Docker Spaces have required an account since July 2026 — not a free alternative.
Comment l'exécuter en local How to run it locally +
Avec Docker (recommandé)
With Docker (recommended)
Tout est contenu dans l'image. C'est la voie la plus fiable.
Everything is contained in the image. This is the most reliable route.
-
Préparez le fichier de jetons :Prepare the token file:
cp .env.example .env
-
Renseignez
HF_TOKENdans.env:Fill inHF_TOKENin.env: -
Démarrez :Start it:
docker compose up -d
-
Suivez le démarrage (premier lancement : téléchargement des modèles) :Follow startup (first run: models download):
docker compose logs -f
Sans Docker
Without Docker
Prérequis : Python 3.10+ et FFmpeg.
Prerequisites: Python 3.10+ and FFmpeg.
python -m pip install -r requirements.txt
python -m uvicorn app.api.main:app --host 127.0.0.1 --port 8000
Ouvrez ensuite
Then open
http://127.0.0.1:8000/Au premier lancement, environ 879 Mo de modèles sont téléchargés. Prévoyez quelques Go d'espace disque et de la patience.
On first run, roughly 879 MB of models are downloaded. Allow a few GB of disk space and some patience.
Besoin du fichier Docker ?
Need the Docker file?
Écrivez-moi et je vous l'envoie, avec les instructions de démarrage.
Email me and I will send it over, with the startup instructions.