Video comparison study

You will compare pairs of short generated videos of a person and say which one is better in three respects. It takes about 10 minutes.

What happens

Each screen shows a reference photo, a driving video, and two generated videos, A and B.
The videos play together and loop. Watch both, then answer three questions.
There are 22 screens. At the end you get a short code to send back.

Only your choices are recorded, inside the code. No personal data is collected and nothing is uploaded. Participation is voluntary.

How to judge

Both videos try to make the person in the reference photo copy the motion in the driving video. They come from two different methods, and their left/right order is random on every screen.

1. Identity preservation

Does the person still look like the one in the reference photo? Face shape, features, hair, skin. Ignore whether the motion is right.

2. Expression tracking

Does the face copy the expression and head motion of the driving video, at the right moments? Mouth, eyes, eyebrows, head pose. Ignore whether the person looks like the reference.

3. Overall quality

Everything together: sharpness, flicker, wobble, warping, artefacts, general realism.

Please note

Many pairs are close. Look carefully, and if it is truly a coin flip, pick the one you lean towards — you must choose one.
Questions unlock after both videos have played through once. Replay as often as you like.
A couple of screens are easy checks where one video is obviously broken.

Practice complete

That is the whole task. The real screens start now.

Saving your responses…

One moment.

Thank you — you're done

Your responses have been recorded. Nothing further is needed, and you can close this tab.

Thank you for taking the time — these judgements are genuinely useful.

Thank you — one last step

Your answers are complete, but they could not be uploaded (you may be offline). Please send this code to Moe — it contains all of them.

Please don't close this tab until you have sent the code — it is the only record of your session. Letters are not case-sensitive.

Unfinished session found

You started this study before and did not finish. Would you like to carry on where you left off?