PackedAvatar

A bundled SadTalker-based talking-head runtime for Hugging Face Spaces.

Upload a source image + audio, pick an AvatarBank identity, or feed in a reusable conditioning bundle. The bundled Wav2Lip checkpoint is used automatically when enabled.

Source image

Driven audio

AvatarBank selection

Gender filter

Style filter

Avatar ID

Optional conditioning bundles

Avatar conditioning bundle (.pt / .pth / .mat)

Motion conditioning bundle (.pt / .pth / .mat)

Reference video / idle mode

Use reference video

Reference video

Reference mode

Use idle mode

Idle length (seconds)

0 30

Rendering options

Remove background

Use enhancer (GFPGAN)

Still mode

Use blink

Preprocess

Face renderer

Size

Pose style

0 45

Expression scale

0.5 2

Batch size

1 8

Result directory (optional)

Output video

Status

Download generated video (click to save)