01
Upload a portrait
A clear, front-facing face works best. Avoid tiny faces, heavy occlusion, and group shots.
Free talking-photo demo
SadTalker Online
Envoyez un portrait et un audio. SadTalker anime le visage selon la voix. Cette page embarque des démos publiques, sans GPU ni clé API.
Essai gratuit Pas de clé API Photo + audio → video
Essayer SadTalker gratuitement Interactive generator
Générer une vidéo parlante
Start with the recommended demo. If it is busy or asleep, switch to a backup. Only the selected demo loads.
Démo recommandée
Public demos can sleep, queue, or go offline without notice.
Loading the SadTalker demo
How to use SadTalker
A three-step talking-photo workflow. Keep clips short so the public queue has a chance to finish.
02
Upload speech audio
Clean speech is easier than noisy music. Shorter clips are kinder to the free GPU window.
03
Generate and download
Submit, keep the tab open through the queue, preview the MP4, and save it from the demo.
What SadTalker is for
SadTalker (CVPR 2023) drives a still face from audio. It is a talking-head tool, not a full-body video generator.
Talking photo
A single portrait plus a voice clip becomes a short lip-synced talking-head video.
Audio-driven motion
Lip motion and expression follow the driven audio rather than a text prompt.
Still and pose options
Many demos expose still mode, preprocess, pose style, and expression scale.
No local GPU setup
The embedded demo talks to a public Space. Wan2.Video does not run the model on our servers.
SadTalker vs Wan 2.2 S2V
Both animate a still with audio, but the job and output feel different.
| Feature | SadTalker | Wan 2.2 S2V |
|---|---|---|
| Job | Talking-head animation of one face | Fuller speech-to-video performance |
| Inputs | Portrait + driven audio | Image + speech or singing audio |
| Best for | Classic talking photo, explainers, avatars | Talking or singing shots with more body motion |
SadTalker FAQ
Is SadTalker free here?
You can try it without creating an account on this site. The compute comes from public ZeroGPU demos, so queues, daily limits, and downtime are expected.
Why can I only generate once?
Anonymous ZeroGPU time is about two GPU minutes per day, and this SadTalker demo often reserves at least 90 seconds per job. A second run may be refused until the window resets, or until you try another backup demo.
Do I need my own API key or GPU?
Not for this page. Official commercial APIs exist but are paid. The official Hugging Face demo is currently broken; community demos are what actually run.
Can this be a production backend?
No. Public Spaces can change, sleep, or disappear. For real traffic, use a paid API or host SadTalker yourself. The project is Apache 2.0.
What should I not upload?
Private photos, voices without consent, and anything sensitive. Files are processed by a third-party public demo.