Visualisations and results from HumAnimate benchmark.

Note:

Preferred browser- Firefox and Chrome.

Works on other browsers too but may be slow to load. (Optional) Due to the volume of the mp4 files, the page may be slow to load. Therefore, we request you to please wait for results to load.

Thank you for taking time to review this page and our work!


Table of Contents



Visualisation T2I-Adapter style pose sequences for Base Set of pose sequences.

Stand up and Sit On-spot jogging Standing wave Jumping jacks
Bend down and get up On-spot dance High knees Squats


Visualisation ControlNet style pose sequences for Advanced Set of pose sequences.

Walk Dance with rotation Cartwheel Moonwalk
Taichi action Knee-tuck jumps Lateral jumps Alternate lungs


Camera motion pose sequences for "Squat" action. The pixel translation stride is 5 here.

Motion direction Squats On spot jogging
Camera moves from left to right (L2R)
Camera moves towards the subject (Zoom-In)


Video outputs by four methods for subject size experiment for prompt "A man performing taichi in a park."

Subject Size Text2Video-Zero ControlVideo ConditionVideo Follow Your Pose
Small
Normal
Large


Video outputs by four methods for different sampling rates (FPS) for prompt "A man performing jumping jacks in a park."

Sampling rate (FPS) Text2Video-Zero ControlVideo ConditionVideo Follow Your Pose
2
5
8


Video outputs by four methods for "Foreground shape control" for prompt "A male < kid/adult > performing lunges while facing the camera in a park."

Age group Text2Video-Zero ControlVideo ConditionVideo Follow Your Pose
A male kid
A male adult


Video outputs by four methods for "Foreground shape control" for prompt "A < fit/fat > male performing lunges while facing the camera in a park."

Body shape Text2Video-Zero ControlVideo ConditionVideo Follow Your Pose
A fat woman
A fit woman