Text to Speech AI Agents Music Speech to Text Voice Cloning Voice Design Marketplace More
Report Bug / Feature Request

AI Lip Sync Video Generator

Turn a face photo and an audio clip into a talking-head video with lip sync, head pose and blinks. Currently unavailable while we look for a model whose license allows it.

Sign Up Free
Lip sync is not available right now. The SadTalker model it used was trained on the Basel Face Model, which is licensed for non-commercial research only, so we cannot offer it on a paid service. No replacement with a suitable license is installed yet.

Upload Face + Audio

1,000 characters per second

Drag & drop your file here, or browse

JPG, PNG, or short MP4/WebM. Max 10MB. One clear, well-lit face works best.

file.mp3

0 MB

Drag & drop your file here, or browse

MP3, WAV, M4A, or FLAC. Max 10MB. Free: up to 30 sec. Pro: up to 5 min.

file.mp3

0 MB

Rendering your video. This typically takes 30 seconds to 2 minutes.

Your Talking-Head Video

Download MP4

About SadTalker

SadTalker (CVPR 2023, Tencent ARC) is an open-source talking-head model that animates a single face image to speak any audio. Unlike Wav2Lip variants, SadTalker also animates head pose, blinks, and expression for a more natural result.

SadTalker's code is Apache-2.0, but its 3D face reconstruction weights were trained on the Basel Face Model, whose license allows non-commercial research only. TTS.ai is a commercial service, so the tool was switched off on September 15, 2026.

Tips for Best Results

  • Use a high-quality, well-lit portrait — eyes visible, mouth closed
  • Centered face, square or 4:5 aspect ratio works best
  • Clean speech audio (no music) yields tighter lip sync
  • Enable GFPGAN for hero shots — doubles render time but sharpens detail
  • Use the Still preset when you want a steady avatar shot

Lip Sync Video Plans

Start free, upgrade when you need more

Free
  • 30-second audio limit
  • 256 px output
  • "Still" preset only
  • No face enhancer
Most Popular
Free Account
  • 30-second audio limit
  • Both "full" and "still" presets
  • 256 / 512 px output
  • GFPGAN face enhancer
Sign Up Free
Pro
  • 5-minute audio limit
  • Priority GPU queue
  • API access (multipart upload)
  • Webhook completion callbacks
  • Currently unavailable (non-commercial model license)
Upgrade

Frequently Asked Questions

Upload a face photo and an audio clip, and the AI generates a video of that face speaking the audio with lip movements, head pose and blinks. The tool is currently unavailable: it ran SadTalker (CVPR 2023), whose 3D face reconstruction weights were trained on the Basel Face Model, which is licensed for non-commercial research only.

The face input can be a JPG or PNG image (up to 10 MB) or a short MP4/WebM driving video (we use the first frame). The driving audio can be MP3, WAV, M4A, or FLAC up to 10 MB. We resample audio to 16 kHz internally.

Free accounts: up to 30 seconds per clip. Paying users: up to 5 minutes per request. Longer audio means longer render time and higher character cost.

Lip sync video uses 1,000 characters per second of generated video. A 30-second clip = 30,000 characters. The cost is billed up front from your character balance and refunded automatically if generation fails.

No. The tool is switched off because the model license does not allow it. SadTalker code is Apache-2.0, but its bundled face reconstruction network was trained on the Basel Face Model 2009, whose license allows internal, non-commercial research only. Videos generated with it before September 15, 2026 are not for commercial use, paid plans included. You are responsible for having the rights to the source face image and audio you upload.

When the tool ran, a 5-second clip took about 30 seconds, scaling roughly linearly with audio length. It is currently unavailable while we look for a talking-head model whose license allows it.

Full preset (default) animates head pose, blinks, and expression along with the lips, producing a more natural talking-head video. Still preset locks the head in place and animates only the mouth — useful when you want a steady avatar shot.

GFPGAN is a face restoration model that sharpened facial details after rendering. It is not offered: its StyleGAN2 prior is under an NVIDIA non-commercial license and its face parsing model is CC BY-NC-SA.

SadTalker rendered at 256 px by default, with 512 px as a slower option. The tool is currently unavailable; a well-lit, high-quality portrait gives the best result with any talking-head model.

Yes. Upload an MP4 or WebM as the face input and we will use the first frame as the driving identity. For full video re-dubbing (per-frame mouth replacement), see the upcoming Dubbing Studio video pipeline.

Yes. POST a multipart request to /api/v1/lipsync/ with face and audio fields, then poll /api/v1/lipsync/result/?uuid= until status is "completed". The response contains a URL to the rendered MP4. API access requires a paid plan.

SadTalker uses face-alignment to detect and crop the most prominent face. For best results, upload a portrait with one person centered, eyes visible, and minimal occlusion. Group photos may produce unpredictable results.
5.0/5 (1)

What could we improve? Your feedback helps us fix issues.

Ready to get started?

Sign up free and get 15,000 characters. No credit card required.

Sign Up Free View Pricing