AudioLift: Audio-Driven Single Image to
Full-Head Gaussian Avatar via Multi-view Video Generation

Face crops
Back to Homepage

More Results

Face crops

FFHQ-style face crops (plus one painting) padded onto the same 832×1216 canvas, so the shoulders are not visible. Rows as on the homepage. The two long-sequence examples at the bottom use the full recordings.

(Homepage)  |  Page 1  |  Page 2  |  Page 3

Subject 07
English · speech · 6.7 s
Subject 08
English · speech · 12.0 s
Subject 09
English · speech · 12.0 s
Subject 10
English · speech · 13.5 s
Subject 11
Japanese · singing · 25.0 s
Input portrait
input portraitSubject 07 · input
input portraitSubject 08 · input
input portraitSubject 09 · input
input portraitSubject 10 · input
input portraitSubject 11 · input
Frontal
talking
Frontal
Frontal
Frontal
Frontal
Frontal
Novel views
talking while the camera moves
Novel views
Novel views
Novel views
Novel views
Novel views
Subject 12
German · singing · 9.7 s
Subject 13
English · singing · 10.6 s
Subject 14
English · speech · 22.5 s
Subject 15
Italian · singing · 9.7 s
Input portrait
input portraitSubject 12 · input
input portraitSubject 13 · input
input portraitSubject 14 · input
input portraitSubject 15 · input
Frontal
talking
Frontal
Frontal
Frontal
Frontal
Novel views
talking while the camera moves
Novel views
Novel views
Novel views
Novel views

(Homepage)  |  Page 1  |  Page 2  |  Page 3

Long-Sequence Examples

Ours on the two longest recordings, generated in one pass at the recording's full length. Use the player controls. In the camera demo the coverage sweep runs over the first 20 s, then the final frame is frozen for a 360° orbit and a pitch nod.

Long example 1 · English speech · 22.6 s

“It is a very fine old place, of red brick softened by a pale powdery lichen, which has dispersed itself with happy irregularity, so as to bring the red brick into terms of friendly companionship with the limestone ornaments surrounding the three gables, the windows, and the door place.”

Input portrait
Ours · frontal
Ours · camera demo
input portraitInput portrait
Ours · frontal
Ours · camera demo
  sound from the frontal video · coverage sweep → frontal → final-frame 360° orbit → pitch nod
Long example 2 · Japanese singing · 25.0 s

“You can stay seated for just a little longer. Take a good look at the scenery of this town.”

Input portrait
Ours · frontal
Ours · camera demo
input portraitInput portrait
Ours · frontal
Ours · camera demo
  sound from the frontal video · coverage sweep → frontal → final-frame 360° orbit → pitch nod